/* ===================================================================
   MANROPE, FOR THE WORDMARK — 11 Oct 2026
   ===================================================================

   Guillaume: "the CoRoute still looks different on the home page and
   on the mission page.. please fix it so they are identical". It did:
   /our-mission set the wordmark in Manrope at 28px, while the
   homepage had no Manrope at all and fell back to system-ui at 22px.
   One page showed a designed wordmark, the other bold system text.

   SELF-HOSTED, NOT GOOGLE FONTS. CSP default-src 'self' blocks the
   Google stylesheet outright, and loading it would hand every
   visitor's IP to Google — which the privacy notice's
   no-third-party-tracking promise does not allow. The files are
   already in this repo for /our-mission and are OFL-licensed; this
   adds no new request type, only one more consumer.

   Declarations copied verbatim from mission.css so the two cannot
   drift: same variable range, same unicode-range, same swap. */
@font-face {
  font-family: 'Manrope';
  font-style: normal;
  font-weight: 400 800;
  font-display: swap;
  src: url(/static/fonts/manrope-latin.woff2) format('woff2');
  unicode-range: U+0000-00FF, U+0131, U+0152-0153, U+02BB-02BC, U+02C6,
    U+02DA, U+02DC, U+0304, U+0308, U+0329, U+2000-206F, U+20AC, U+2122,
    U+2191, U+2193, U+2212, U+2215, U+FEFF, U+FFFD;
}

@font-face {
  font-family: 'Manrope';
  font-style: normal;
  font-weight: 400 800;
  font-display: swap;
  src: url(/static/fonts/manrope-latin-ext.woff2) format('woff2');
  unicode-range: U+0100-02BA, U+02BD-02C5, U+02C7-02CC, U+02CE-02D7,
    U+02DD-02FF, U+0304, U+0308, U+0329, U+1D00-1DBF, U+1E00-1E9F,
    U+1EF2-1EFF, U+2020, U+20A0-20AB, U+20AD-20C4, U+2113, U+2C60-2C7F,
    U+A720-A7FF;
}

/* ===================================================================
   DM SANS, FOR BODY TEXT — 7 Nov 2026
   ===================================================================

   THESE TWO FACES WERE DECLARED ONLY IN mission.css, AND THAT MADE
   THE FIRST ATTEMPT AT THIS CHANGE SILENTLY WRONG. Setting
   `font-family: 'DM Sans'` on body in this file named a family the
   browser had no file for on any page but /our-mission, which is the
   one page that loads mission.css. Every other page fell back — and
   it fell back INVISIBLY, because getComputedStyle still reports
   "DM Sans": the computed value is what the cascade resolved, not
   what the rasteriser found.

   MEASURED, which is the only way this shows up. Rendering
   "Handgloves 123" at 400 40px through canvas measureText:

       page           DM Sans    Manrope    system-ui
       home            261.1      285.2       300.8     <- fallback
       privacy         261.1      285.2       300.8     <- fallback
       /our-mission    283.7      285.2       300.8     <- the real face

   22.6px apart for the same named family on the same browser in the
   same run. The network panel said the same thing more bluntly: only
   /our-mission requested dmsans-latin.woff2 at all.

   SO THE DECLARATIONS LIVE HERE NOW, beside Manrope's, and every page
   that loads app.css can actually paint the face it asks for.
   mission.css keeps its own copies: it is a self-contained port and
   duplicate @font-face rules for an identical src are harmless — the
   browser fetches the file once. See the note at its head.

   Declarations copied verbatim from mission.css so the two cannot
   drift: same variable range, same unicode-range, same swap. */
@font-face {
  font-family: 'DM Sans';
  font-style: normal;
  font-weight: 400 700;
  font-display: swap;
  src: url(/static/fonts/dmsans-latin.woff2) format('woff2');
  unicode-range: U+0000-00FF, U+0131, U+0152-0153, U+02BB-02BC, U+02C6,
    U+02DA, U+02DC, U+0304, U+0308, U+0329, U+2000-206F, U+20AC, U+2122,
    U+2191, U+2193, U+2212, U+2215, U+FEFF, U+FFFD;
}

@font-face {
  font-family: 'DM Sans';
  font-style: normal;
  font-weight: 400 700;
  font-display: swap;
  src: url(/static/fonts/dmsans-latin-ext.woff2) format('woff2');
  unicode-range: U+0100-02BA, U+02BD-02C5, U+02C7-02CC, U+02CE-02D7,
    U+02DD-02FF, U+0304, U+0308, U+0329, U+1D00-1DBF, U+1E00-1E9F,
    U+1EF2-1EFF, U+2020, U+20A0-20AB, U+20AD-20C4, U+2113, U+2C60-2C7F,
    U+A720-A7FF;
}

/* Desktop and mobile are equal first-class targets (D-requirement).
   Not mobile-first with desktop reflowed, nor the reverse: the layout
   is designed at both ends and checked at the awkward widths between. */

/* ===================================================================
   DESIGN TOKENS — "Open sky": navy, sky blue, cream, orange
   ===================================================================

   WEBSITE V2, 22 OCT 2026. Guillaume: "keep the old design, just the
   new colour stuff." The palette below replaces the purple/orange
   direction this file carried from 22 Sep to 22 Oct 2026. NOTHING
   ELSE CHANGED — layout, copy, section order and behaviour are the
   approved v1 design, and every rule beneath this block still has the
   geometry it was signed off with. The design canvas page
   "Website v2 · Open sky (light + dark)" is the source, and its
   v2rules note says the same thing in Guillaume's own words.

   WHERE THE VALUES COME FROM. Copied from v2-src/css/d2.css on the
   canvas, which is the reviewed preview and the reference
   implementation. The three blocks below — :root, the
   prefers-color-scheme block and the [data-theme="dark"] block —
   are all three deliberate and none is redundant:

     :root                      light, and the default
     @media (prefers-color-scheme: dark)
       :root:not([data-theme="light"])
                                the NO-JS FALLBACK. Without it a
                                visitor with JS off and a dark OS gets
                                the light theme, because nothing sets
                                data-theme. It is guarded by
                                :not([data-theme="light"]) so an
                                explicit Light choice still wins.
     :root[data-theme="dark"]   the EXPLICIT choice, set by
                                /static/theme-init.js before first
                                paint and by the header's theme
                                switch afterwards.

   THE PURPLE IS GONE AND SO ARE ITS REASONS. The notes that used to
   stand here explained why --brand DEEPENED in dark mode (#5B21B6
   over a lighter #8B5CF6 that measured 4.05:1) and why links went
   lavender. Both were arguments about a purple that no longer
   exists; keeping them would have left this file explaining a
   decision it no longer implements, which is the exact failure mode
   CLAUDE.md names. The navy equivalents are recorded below instead.

   TWO ROLES THAT ARE STILL TWO. The split the purple palette
   introduced survives the recolour unchanged and for the same
   reason: orange cannot be both a fill and a text colour. White on
   #FFAA66 is 1.8:1, dark ink on it is 7.61:1 — so the CTA is an
   orange FILL with NAVY ink (--on-cta: #142B49, never white), while
   links, focus rings and control borders take --accent, which is the
   navy in light mode and a light blue in dark.

   MEASURED RATIOS, LIGHT / DARK, all computed not eyeballed:

     --ink on --paper            14.27:1 / 12.03:1
     --ink on --page             13.64:1 / 14.21:1
     --ink-soft on --page         7.27:1 /  9.65:1
     --ink-soft on --panel        6.44:1 /  7.10:1
     --on-cta on --cta            7.61:1 /  7.64:1
     --ink on --sky               8.11:1 / 13.69:1   (hero copy)
     --field-border on --paper    7.60:1 / 10.18:1   (needs 3:1)
     --accent on --page          13.64:1 /  9.06:1   (needs 3:1)

   --logo (#E8762A on cream) MEASURES 2.84:1 AND THAT IS DELIBERATE.
   It draws the ring C and the "o" of the wordmark, which is a
   LOGOTYPE — explicitly exempt from WCAG 1.4.11 ("Logotypes: text
   that is part of a logo or brand name has no contrast
   requirement"). It is never used for body text, a control edge or
   a focus ring, and the wordmark's accessible name comes from the
   markup, not from the glyph being legible at a given ratio. */

:root {
  color-scheme: light;

  --page:       #FFF9F1;   /* warm off-white reading surface */
  --paper:      #FFFFFF;   /* raised card surface */
  --panel:      #DCEFFA;   /* pale sky explanatory panel */
  --section-alt:#EAF4FA;   /* the second, cooler section ground */
  --warm:       #F4ECDF;   /* the warm band (mission strip, etc.) */
  --ink:        #142B49;
  --ink-soft:   #40566B;
  --line:       #CBD8E1;   /* decorative rules only, never a control edge */

  --brand:      #142B49;
  --on-brand:   #FFFFFF;
  --brand-edge: #06111F;
  --cta:        #FFAA66;
  --cta-hover:  #FF9E52;
  --cta-edge:   #C8682A;
  --on-cta:     #142B49;

  /* --accent is the LINK/FOCUS/BORDER colour. It is deliberately not
     the CTA fill; see the note above. Kept under the old name because
     the rules below reference it and they all mean "the colour that
     text and outlines are drawn in", which is still true — only the
     hue changed, from purple to navy. */
  --accent:     #142B49;
  --accent-ink: #FFFFFF;
  /* ACCENT AS *TEXT*, which is not --accent and not --cta. It is the
     colour an emphasised WORD takes, paired with --mark underneath
     it; see .h1-accent for how the two work together and why the
     pair is one rule rather than two designs. */
  --accent-text:#142B49;
  --mark:       #FF9E52;   /* underline/marker stroke under headlines */
  /* The orange rule under the links in the flow nav. Its own token
     because --mark goes transparent in dark and --accent is the navy
     here; see .flownav-links a. Named for "Home" because it landed
     there first on 29 Oct 2026 and went to every item hours later --
     a known misnomer, left rather than renamed across both themes
     for a colour that did not change. */
  --nav-home-rule: #FF9E52;
  --logo:       #E8762A;   /* ring C and the "o" — logotype only */

  /* THE HERO'S OWN GROUND, and --sky is the artwork's painted sky.
     The hero fades from --hero-bg into --sky so the art's feathered
     top edge melts into the same blue rather than showing a seam.
     Sampled from the shipped art: day/hero-home-wide.webp is
     rgb(115,197,254) at its top centre, night is rgb(1,31,82). */
  --hero-bg:    #DCEFFA;
  --hero-text:  #142B49;
  --hero-soft:  #2F4660;
  --sky:        #84CBFC;

  /* The dark bands (the contextual section, the footer). */
  --band:       #142B49;
  --band-2:     #1D3A5E;
  --band-text:  #FFF9F1;
  --band-soft:  #C9DAEA;
  --band-line:  rgba(255,249,241,.16);
  --footer-bg:  #142B49;
  --footer-text:#FFF9F1;

  --ok-bg:      #EAF7EE;  --ok-ink:  #17643A;
  --err-bg:     #FDECEC;  --err-ink: #8C1D1D;
  --warn-ink:   #8A5200;  --warn-bg: #FFF1E2;
  --good-solid: #17643A;  --good-soft: #BFF0CF;

  --pill-bg:    #FFE6D1;  --pill-line: #FFC79B;  --pill-ink: #7A3606;
  --chip-dark:  rgba(12,24,42,.92);

  /* THE FORM FIELD'S OWN SURFACE. Separate from --paper because a
     field is not a card: the v2 boards declare --field-bg #FFFFFF in
     light and #101D30 in dark, which is the PAGE ground rather than
     the card ground, so a field reads as a well rather than as
     another raised surface. --field-border is further down this file
     with --up and --down. */
  --field-bg:   #FFFFFF;

  /* PRODUCT UI — the portal mock-up, the AI chat card, the drive
     review. These are a screenshot of software, not of this website,
     so they get their own neutral surface set rather than borrowing
     the marketing palette. */
  --ui-side:   #142B49;  --ui-side-ink: #C9DAEA;
  --ui-bg:     #F3F7FA;  --ui-card:     #FFFFFF;  --ui-line: #E0E8EE;
  --ui-ink:    #142B49;  --ui-soft:     #55697D;
  --ui-accent: #1F5FA8;  --ui-accent-soft: #E3EEF8;  --ui-bar: #EAF0F5;

  /* The drawn street map in the portal's trip thumbnails, carried from
     HomeV2.dc.html, which classes every part instead of inlining a hex.
     Named after what each part DRAWS rather than after a role in this
     palette, and given a dark set of its own: a map re-tinted by a
     filter turns grass pink. */
  --map-bg:    #EDF1F4;  --map-block: #E3E9EE;  --map-road:  #FFFFFF;
  --map-main:  #F7E3B5;  --map-water: #CFE3F5;  --map-park:  #DDEFD9;

  --radius:     0.75rem;   /* controls */
  --radius-panel: 1.25rem; /* cards and panels */
  --measure:    34rem;

  /* ---- THE SPACING SCALE ---------------------------------------
     Added 6 Oct 2026. Until then app.css had NO spacing tokens at
     all — every gap was a literal, which is how the rhythm drifted
     from the boards without anyone being able to see that it had.

     UNCHANGED BY THE v2 RECOLOUR. These are geometry, and v2 is
     colours only.

     NAMED FOR THEIR ROLE, NOT THEIR SIZE. A token called
     --space-14 is the same literal with extra steps: it tells the
     next reader nothing about when to use it, and it has to be
     renamed when the value changes. --field-gap survives a change
     of value, which is the point of having it.

     IN REM, NOT PX. Guillaume, 6 Oct: "things should be relative as
     well". Base is 16px, so 0.875rem is the board's 14px — and at
     200% text the rhythm scales with the type instead of leaving
     doubled text crammed into a fixed gutter.

     TWO SETS. These are the DESKTOP values, which are signed off
     and must not move. The phone boards are drawn on a tighter
     rhythm and the block at the 30rem breakpoint re-points these
     same tokens at the board's values — so the phone rhythm is
     changed by editing tokens in one place, not by hunting
     literals through 7,000 lines.

     The quote-layout skill's R5 documents both scales and says
     which is which. The boards are the source of truth for the
     phone column; see the note at the phone block.

     THESE ARE THE VALUES THE DESKTOP ALREADY RENDERS, measured on
     the running page at 1280 before the tokens existed — NOT the
     values R5's table named. Four differed from that table, and the
     PAGE is what is signed off, so the page won:

       field -> next field     16px   (R5's table said 24)
       section head -> field   28px   (said 16)
       section -> next         32px   (said 48)
       label -> control         6px   (said 4)

     Taking the table's numbers would have silently re-spaced every
     desktop form on the way to fixing the phone — which is exactly
     what happened on the first attempt here, measured as 24/40/48
     before it was caught. */
  --label-gap:    0.375rem;  /*  6px  label -> control -> hint     */
  --option-gap:   0.75rem;   /* 12px  options within one group     */
  --field-gap:    1rem;      /* 16px  field -> next field          */
  --heading-gap:  0.75rem;   /* 12px  legend margin -> 28px rendered */
  --section-gap:  2rem;      /* 32px  section -> next section      */
  --gutter:       1.5rem;    /* 24px  page gutter at >= 48rem      */
  --card-pad-y:   0.875rem;  /* 14px  radio card                   */
  --card-pad-x:   1rem;      /* 16px                               */
  --btn-pad-y:    0.8125rem; /* 13px  button                       */
  --btn-pad-x:    1.25rem;   /* 20px                               */
  --ctrl-pad-y:   0.75rem;   /* 12px  text input and select        */
  --ctrl-pad-x:   0.875rem;  /* 14px                               */

  /* The homepage board's two extra surfaces.
     --secondary-panel-bg backs the two built step cards; the board
     uses a deeper tone for --statement-bg, which is the difference
     these two names carry.

     --statement-bg NO LONGER BACKS A HOMEPAGE PANEL. It was the
     box behind "Not your marital status." until 4 Oct 2026, when
     that section was restyled onto the page ground (centred, no
     box — Guillaume: the old one was "strangely formatted"). Its one
     remaining consumer is .noask on the advisory result page, which
     makes the same argument in the same voice; the token is kept for
     that rather than retired, and the dark-mode pair below with it.

     ⚠️ --statement-bg IS GREEN-GREY, NOT A SECOND BLUE, 23 Oct 2026.
     The quote flow v2 boards (MainV2/VehicleV2/ResultV2, and the
     dark and phone variants) draw the "What we didn't ask" panel in
     #DCEBDD light / #223D57 dark, and .noask is this token's one
     consumer -- so the token moves with it.

     WHY THE BOARD PARTS FROM THE BLUE HERE. Every other surface in
     the flow is a tint of the page or the sky; this panel is the one
     that says what the form did NOT ask, and the board gives it a
     distinct hue so it reads as an aside rather than another field
     group. #CCE4F4 was the v1 tone and was a pale blue like
     everything around it.

     MEASURED ON THE RENDERED PANEL, not estimated: --statement-text
     on --statement-bg is 11.53:1 light / 9.41:1 dark, read off
     .noask on step 2 (its only consumer) with the colours composited
     on canvas. An earlier draft of this note guessed 13.51 and 11.74;
     the real figures are these. --ink on --secondary-panel-bg is
     12.79:1. Both dark-mode counterparts are below. */
  --secondary-panel-bg:   #EAF4FA;
  --statement-bg:         #DCEBDD;
  --statement-text:       #142B49;
  /* 1200px, the board's content width, UP TO 90rem. Above that the
     container widens to 1440px — see --site-max-wide below. */
  --site-max:             75rem;

}
/* ---- WIDE SCREENS GET A WIDER CONTAINER -------------------------
   16 Oct 2026. Guillaume, on a ~2525px window: the hero's video card
   "seems to be an issue in this particular browser... even after I
   clear the cache".

   IT WAS NEITHER THE BROWSER NOR THE CACHE, and that matters for
   anyone reading this later: the same render came back from a clean
   headless Chrome at the same width. Measured across widths, the
   card is ALWAYS 36.7% of the content container — the board's
   proportion, holding exactly. What changes is the margin around it:

     viewport   container   card   empty each side
     1280       1200        440    33px
     1920       1200        440    353px
     2525       1200        440    655px

   So the card never shrinks; the PAGE stops growing at 1200px while
   the window keeps going, and at 2525 the card is 17% of the screen
   with 655px of bare hero ground beside it. Nothing was stale, which is
   why clearing the cache could not help.

   The fix is to let the container grow on screens big enough to use
   it. 1440px keeps the same 36.7% card (now ~530px) and cuts the
   waste to 535px a side, while staying inside a readable measure for
   the copy column — the reason a cap exists at all. Guillaume chose
   this over letting the card take a bigger share of a 1200px box,
   which would have left the margins untouched.

   THE THRESHOLD IS 90rem (1440px), not the new width itself: the
   container may only widen once the viewport can hold it plus
   gutters, or it would simply fill the screen edge to edge on a
   1440px laptop.

   THE HERO ART FOLLOWS AUTOMATICALLY and that is intended. Its right
   edge is calc()'d from --site-max (see the long note on .hero-art),
   so it stays locked to the content column rather than drifting —
   which is the behaviour Guillaume demanded on 9 Oct: "THE CAR
   SHOULD NOT MOVE RELATIVE TO THE TEXT AND PLAYING VIDEO." */
@media (min-width: 90rem) {
  :root { --site-max: 90rem; }   /* 1440px */
}


/* ---- THE PHONE SPACING SCALE ------------------------------------

   THE BOARDS ARE THE SOURCE OF TRUTH AND THEY ARE TIGHTER THAN THE
   DESKTOP. Guillaume, 6 Oct 2026, comparing the live mobile view
   against MainPhone: "board on design is always source of truth" and
   "things should be relative as well". He was right that the page was
   loose — the board reaches "Town or city / Postcode" in the height
   the page needed to reach "Where do you live?".

   THIS IS A DIFFERENT SYSTEM, NOT A FEW STRAY VALUES, and that is why
   it gets its own set of tokens rather than a handful of overrides.
   Counted across all three phone boards: 173 of 355 px spacing
   declarations are off the desktop scale, and the three commonest
   off-scale values are 13px (x57), 10px (x48) and 14px (x34).

   The quote-layout skill's R5 and R6 were rewritten in the same
   commit. They used to name one scale and forbid 10, 14, 22 and 28
   outright — three of which the boards use throughout — so a
   contract that said "off-scale values are forbidden" was, as
   written, an instruction to ignore the boards. R6 now forbids
   inventing a value that is on NEITHER scale, which is the rule it
   was always trying to be.

   RE-POINTING THE SAME TOKENS, not adding phone-only ones. Every rule
   that consumes --field-gap gets the phone rhythm here for free, and
   the rhythm is changed by editing these ten lines rather than by
   hunting literals. Every value is rem so it scales with the user's
   text size; the one px left in the phone work is the 3px hairline
   rule on the keeper indent, because a hairline is a hairline.

   Values read off the boards by counting every declaration, not
   eyeballed. */
@media (max-width: 30rem) {
  :root {
    --label-gap:    0.25rem;   /*  4px */
    --option-gap:   0.625rem;  /* 10px */
    --field-gap:    0.875rem;  /* 14px */
    --heading-gap:  1rem;      /* 16px */
    --section-gap:  1.75rem;   /* 28px */
    --gutter:       1.25rem;   /* 20px */
    --card-pad-y:   0.8125rem; /* 13px */
    --card-pad-x:   0.875rem;  /* 14px */
    --btn-pad-y:    0.9375rem; /* 15px */
    --btn-pad-x:    1.5rem;    /* 24px */
    /* 13px + 16px text + 13px = the board's ~44px control, which is
       also the minimum touch target. Desktop keeps 12px/52px. */
    --ctrl-pad-y:   0.8125rem; /* 13px */
    --ctrl-pad-x:   0.875rem;  /* 14px */
  }
}

/* ---- DARK MODE, TWICE, AND BOTH COPIES ARE LOAD-BEARING ---------

   The two blocks below carry THE SAME VALUES under two different
   selectors, and neither can be dropped:

     @media (prefers-color-scheme: dark) :root:not([data-theme="light"])
       is the NO-JS FALLBACK. With JavaScript off nothing ever sets
       data-theme, so without this block a visitor whose OS is dark
       gets the light theme. The :not([data-theme="light"]) guard is
       what lets an EXPLICIT Light choice still win over a dark OS —
       without it the media query would beat the attribute and the
       Light radio would do nothing for those users.

     :root[data-theme="dark"]
       is the EXPLICIT choice, written by /static/theme-init.js
       before first paint and by the theme switch afterwards. It must
       win over a light OS, which is why it is a plain attribute
       selector outside any media query and comes last.

   DUPLICATION IS THE POINT, not an oversight. CSS has no way to say
   "these two selectors share a declaration block" without a
   preprocessor, and this project has no build step (D10). The
   alternative — one block keyed only on the attribute — would make
   dark mode a JavaScript feature, which it is not.

   KEEP THEM IN STEP. A value changed in one and not the other
   produces a page that is dark for OS-dark visitors and a different
   dark for people who clicked Dark. TestV2DarkBlocksAgree in
   web/internal/quote/css_integrity_test.go compares them literally
   so they cannot drift apart silently.

   MEASURED, DARK:
     --ink (#E5ECF3) on --page (#101D30)       14.21:1
     --ink on --paper (#182B43)                12.03:1
     --ink-soft (#B7C5D6) on --page             9.65:1
     --ink-soft on --panel (#1E3552)            7.10:1
     --on-cta (#142B49) on --cta (#FFAA70)      7.64:1
     --accent (#FFAA70) on --page               9.06:1   (needs 3:1)
     --on-brand (#142B49) on --brand (#A9DCFA)  9.72:1
     --ok-ink on --ok-bg                        9.10:1
     --err-ink on --err-bg                      7.94:1
     --warn-ink on --warn-bg                    7.27:1

   THE ORANGE CTA KEEPS ITS NAVY INK IN DARK MODE. #142B49 on #FFAA70
   is 7.64:1; inverting it would put a 1.8:1 white on orange. This was
   true of the purple palette for the same arithmetic and it is still
   true. */
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) {
    color-scheme: dark;

    --page:       #101D30;
    --paper:      #182B43;
    --panel:      #1E3552;
    --section-alt:#142640;
    --warm:       #17263B;
    --ink:        #E5ECF3;
    --ink-soft:   #B7C5D6;
    --line:       #2F4A68;

    --brand:      #A9DCFA;
    --on-brand:   #142B49;
    --brand-edge: #5E93B5;
    --cta:        #FFAA70;
    --cta-hover:  #FFBB8A;
    --cta-edge:   #B9622A;
    --on-cta:     #142B49;

    --accent:     #FFAA70;
    --accent-ink: #142B49;
    --accent-text:#FFAA70;
    --mark:       transparent;
    --nav-home-rule: #FFAA70;
    --logo:       #FFB264;

    --hero-bg:    #142B49;
    --hero-text:  #E5ECF3;
    --hero-soft:  #C9DAEA;
    --sky:        #021E4A;

    --band:       #0B1626;
    --band-2:     #182B43;
    --band-text:  #E5ECF3;
    --band-soft:  #B7C5D6;
    --band-line:  rgba(229,236,243,.14);
    --footer-bg:  #0B1626;
    --footer-text:#E5ECF3;

    --ok-bg: #16301F; --ok-ink: #8FE0AD;
    --err-bg: #341A1A; --err-ink: #F2A1A1;
    --warn-ink: #E0B070; --warn-bg: #33281A;
    --good-solid: #1F7A4A; --good-soft: #BFF0CF;

    --pill-bg: rgba(255,170,112,.14);
    --pill-line: rgba(255,170,112,.45);
    --pill-ink: #FFCB9A;
    --chip-dark: rgba(6,14,26,.9);
    --field-bg:   #101D30;

    --ui-side:   #0B1626;  --ui-side-ink: #B7C5D6;
    --ui-bg:     #101D30;  --ui-card:     #182B43;  --ui-line: #2A4260;
    --ui-ink:    #E5ECF3;  --ui-soft:     #A7B8CA;
    --ui-accent: #A9DCFA;  --ui-accent-soft: #20406A;  --ui-bar: #142640;

    --map-bg:    #1B2C42;  --map-block: #22364F;  --map-road:  #2C4561;
    --map-main:  #5C4B2E;  --map-water: #21405F;  --map-park:  #1E3A33;

    /* Both homepage panels collapse to near-neighbours in dark mode:
       the light theme's pale-sky pair does not survive inversion as
       two clearly distinguishable tones, and inventing a third dark
       blue to keep the distinction would be decoration nobody asked
       for. --ink on #1E3552 is 10.45:1. */
    --secondary-panel-bg: #142640;
    --statement-bg:       #223D57;
    --statement-text:     #E5ECF3;
  }
}
:root[data-theme="dark"] {
  color-scheme: dark;

  --page:       #101D30;
  --paper:      #182B43;
  --panel:      #1E3552;
  --section-alt:#142640;
  --warm:       #17263B;
  --ink:        #E5ECF3;
  --ink-soft:   #B7C5D6;
  --line:       #2F4A68;

  --brand:      #A9DCFA;
  --on-brand:   #142B49;
  --brand-edge: #5E93B5;
  --cta:        #FFAA70;
  --cta-hover:  #FFBB8A;
  --cta-edge:   #B9622A;
  --on-cta:     #142B49;

  --accent:     #FFAA70;
  --accent-ink: #142B49;
  --accent-text:#FFAA70;
  --mark:       transparent;
  --nav-home-rule: #FFAA70;
  --logo:       #FFB264;

  --hero-bg:    #142B49;
  --hero-text:  #E5ECF3;
  --hero-soft:  #C9DAEA;
  --sky:        #021E4A;

  --band:       #0B1626;
  --band-2:     #182B43;
  --band-text:  #E5ECF3;
  --band-soft:  #B7C5D6;
  --band-line:  rgba(229,236,243,.14);
  --footer-bg:  #0B1626;
  --footer-text:#E5ECF3;

  --ok-bg: #16301F; --ok-ink: #8FE0AD;
  --err-bg: #341A1A; --err-ink: #F2A1A1;
  --warn-ink: #E0B070; --warn-bg: #33281A;
  --good-solid: #1F7A4A; --good-soft: #BFF0CF;

  --pill-bg: rgba(255,170,112,.14);
  --pill-line: rgba(255,170,112,.45);
  --pill-ink: #FFCB9A;
  --chip-dark: rgba(6,14,26,.9);
  --field-bg:   #101D30;

  --ui-side:   #0B1626;  --ui-side-ink: #B7C5D6;
  --ui-bg:     #101D30;  --ui-card:     #182B43;  --ui-line: #2A4260;
  --ui-ink:    #E5ECF3;  --ui-soft:     #A7B8CA;
  --ui-accent: #A9DCFA;  --ui-accent-soft: #20406A;  --ui-bar: #142640;

  --map-bg:    #1B2C42;  --map-block: #22364F;  --map-road:  #2C4561;
  --map-main:  #5C4B2E;  --map-water: #21405F;  --map-park:  #1E3A33;

  --secondary-panel-bg: #142640;
  --statement-bg:       #223D57;
  --statement-text:     #E5ECF3;
}

*, *::before, *::after { box-sizing: border-box; }

/* THE ROOT TAKES THE FOOTER'S COLOUR, NOT THE PAGE'S. 28 Oct 2026.

   A short page — the confirmation screen is the only one — ended
   with a band of cream BELOW the navy footer at 1080px tall.

   Two things cause that and only the second is obvious. body is now
   a 100dvh column with the footer on margin-top:auto, which puts the
   footer's BOTTOM EDGE exactly at the viewport bottom. But <html>
   still paints the canvas behind everything, and the browser
   propagates BODY's background to the canvas when html has none —
   so the area below body took --page, the cream, and showed as a
   strip under the footer wherever body was shorter than the
   scrollable area (a sub-pixel rounding of dvh is enough).

   Declaring a background HERE stops the propagation: html's own
   background wins the canvas, body's paints only body's box. Set to
   the footer colour so that any overscroll or rounding gap continues
   the footer rather than interrupting it. On a long page nothing
   below the footer is ever visible, so this is invisible there. */
html {
  -webkit-text-size-adjust: 100%;
  background: var(--footer-bg);
}

/* --page, not --paper: the reading surface is the warm off-white and
   --paper is reserved for cards raised off it. Body copy is 18px
   rising to 19px, which is the design system's stated range, not the
   16-17px this used to be. */
body {
  margin: 0;
  background: var(--page);
  color: var(--ink);
  /* 'DM SANS', NOT system-ui. 7 Nov 2026.

     THIS USED TO END `system-ui, -apple-system, "Segoe UI", Roboto,
     sans-serif`, and it was the reason the site rendered in a
     different typeface on every operating system — Cantarell or
     Ubuntu on Linux, Segoe UI on Windows, SF on a Mac. Nothing was
     broken; the site simply had no typeface of its own on any page
     but /our-mission, which is a port and brought DM Sans and Manrope
     with it.

     So the site ran THREE typefaces: whatever the visitor's OS
     supplied, plus DM Sans and Manrope on one page. Guillaume asked
     for one set across the site and chose the pair /our-mission
     already uses.

     IT COSTS NOTHING TO SERVE. Both families are self-hosted in
     /static/fonts/, declared at the head of this file and of
     mission.css, and were already downloaded by anyone who visited
     /our-mission. No new request type, no new origin, no CSP change —
     which matters, because default-src 'self' blocks the Google
     Fonts stylesheet outright and loading it would hand every
     visitor's IP to Google, against the privacy notice's promise.

     THE FALLBACK STACK IS STILL THE SYSTEM ONE, for the swap window
     and for the rare client that refuses woff2. font-display: swap on
     the @font-face means that text paints immediately in the fallback
     and reflows once DM Sans arrives.

     mission.css:714 RECORDED THE OPPOSITE DECISION — "that is
     deliberate and it STAYS -- Guillaume's instruction is the nav,
     not the page" — and that note was right on 23 Oct, when the ask
     was to stop the two navs reading as two components. This is the
     wider ask he did not make then. The nav exception in mission.css
     is now redundant rather than wrong: it pins the nav to the stack
     the rest of the site used, which is the stack the rest of the
     site no longer uses. See the note there. */
  font: 400 clamp(1.125rem, 1.1rem + 0.15vw, 1.1875rem)/1.55
        'DM Sans', system-ui, -apple-system, "Segoe UI", Roboto, sans-serif;
  /* STICKY FOOTER. 28 Oct 2026.

     The site had none. Every page was simply long enough that the
     footer fell below the fold, so nobody saw the gap — until the
     confirmation page, which is short, and ended with a band of
     cream page background BELOW the navy footer. Guillaume found it
     on a 1080-tall window.

     min-height on BODY rather than a wrapper, because this site has
     no single wrapper element: the templates put the banner, nav,
     main and footer as siblings directly in <body>. 100dvh, not
     100vh, so mobile browsers measure the visible viewport rather
     than the one behind the collapsing address bar — 100vh there
     leaves a strip of background exactly like the one this fixes.

     The footer takes margin-top:auto below (.sitefooter), which is
     what pushes it down; without the auto margin this rule only
     stretches the body and changes nothing. */
  min-height: 100dvh;
  /* ~~GRID, NOT FLEX COLUMN~~ -- REVERSED 29 Oct 2026, see the
     flex rule at the foot of this block. The paragraph below is
     kept because its DIAGNOSIS was right and still governs: it is
     the reason `body > * { min-width: 0 }` exists.

     28 Oct 2026: `display:flex; flex-direction:column` made every
     direct child of <body> size to its CONTENT instead of to the
     body's width, so a child wider than the viewport widened the
     page -- at 390px the nav and the hero ran off the right edge on
     EVERY page. Caught by screenshotting the homepage at phone
     width while chasing a different fault.

     WHAT THAT DIAGNOSIS MISSED is that this is the default
     `min-width: auto` on a flex item, not a property of flex
     columns, and one line fixes it. Grid was the right call with
     the information available; it was the wrong tool for the job,
     because a grid cannot say "every row to its content except
     one". See the measurements below. */
  /* AUTO ROWS, NOT `1fr auto`. Corrected 28 Oct 2026.

     `grid-template-rows: 1fr auto` assumed <body> has exactly two
     children in exactly one order: content, then footer. That is
     true of most pages and NOT true of the confirmation page on a
     DARK deployment, where the template drops the nav and the
     footer and the children become banner / skiplink / main /
     legal-line. The BANNER then landed in the 1fr row and stretched
     to fill the viewport: a 350px navy void above the page, with
     everything shoved below the fold.

     Guillaume saw it on production and I had not, because I checked
     every render against LOCAL, where HOLDING_PAGE_ONLY is unset and
     .Dark is therefore false -- the one configuration this page is
     NOT served in on production.

     Letting every row size to its content and giving the footer
     margin-top:auto (see .sitefooter) keeps the sticky-footer
     behaviour without asserting anything about how many children
     there are or which one is last.

     THAT PROPERTY IS PRESERVED BY THE FLEX RULE BELOW -- it names
     the content element, never a position or a count -- so the
     28 Oct void cannot come back. */
  /* AND THE SLACK GOES TO THE CONTENT, WHICH THE 28 Oct FIX LEFT OUT.
     29 Oct 2026.

     Auto rows do not say where a SHORT page's leftover height goes.
     With min-height:100dvh above, a page whose content is shorter
     than the viewport has spare space, and auto rows let the grid
     stretch EVERY row to absorb it -- so the banner and the nav grew
     with the rest.

     Measured on /quote/email/{id} at 1440x1000, which is short (main
     605px, total content ~890): the banner rendered 81px against 44
     on every other page and the nav 112 against 75, with the inner
     contents still their correct 44 and 74. Guillaume: "this looks
     pretty bad" -- a deep empty band above the heading.

     THE FOOTER'S margin-top:auto COULD NOT FIX THIS, and that is the
     part the 28 Oct note missed. An auto margin only consumes space
     the grid has already given to ITS row; it cannot claim space the
     grid handed to the banner's row instead.

     THE FIX IS A FLEX COLUMN, WITH THE GUARD THAT WAS MISSING THE
     FIRST TIME. A grid cannot express "every row to its content
     except one", which is exactly what a page shell needs: four
     candidates were measured and all four failed --
     grid-auto-rows:min-content pins main's row too and leaves a
     149px gap below the footer; align-content:start leaves the free
     space outside every track, where no margin can reach it;
     align-self:start on the chrome shrinks the boxes but leaves them
     sitting in oversized tracks, so the bands separate; and a
     calc() min-height on main hardcodes the chrome's heights, which
     breaks the moment the banner wraps.

     flex-direction:column does express it: `flex: 1 0 auto` on the
     content element and nothing on the chrome.

     THIS WAS TRIED ON 28 Oct AND REVERTED, and the note above says
     why -- every child sized to its CONTENT instead of the body's
     width, so a child wider than the viewport widened the page and
     the nav ran off the right edge at 390. That is the default
     `min-width: auto` on a flex item, and `body > * { min-width: 0 }`
     is the one line that was missing. Re-measured at 360, 390, 480,
     768, 993, 1280 and 1440 after adding it: horizontal overflow is
     ZERO at every width.

     Measured at 1440x1000 on the page that exposed this: banner 44
     (was 81), nav 75 (was 112), main grows 568 -> 717, footer flush
     with the bottom and no gap. */
  display: flex;
  flex-direction: column;
}
/* THE GUARD THAT MAKES THE FLEX COLUMN SAFE. Without it a flex item
   defaults to min-width:auto and refuses to shrink below its content,
   so one wide child widens the whole page -- which is what reverted
   the first attempt at this on 28 Oct 2026. */
body > * {
  min-width: 0;
}
/* THE ONE CHILD THAT GROWS on a short page, so the chrome keeps its
   natural height and the footer still reaches the bottom edge.

   .mission is named because /our-mission uses <section class="mission">
   as its content element rather than <main>. A page with neither is
   unaffected: nothing grows, and .sitefooter's margin-top:auto (which
   DOES work in a flex column) still pushes the footer down. */
body > main,
body > .mission {
  flex: 1 0 auto;
}

/* Money is tabular everywhere it appears, so digits line up between
   the breakdown rows and between the two ends of a range. */
.bignumber, .amount, .working tfoot td, .savings-bar-figure {
  font-variant-numeric: tabular-nums;
}

/* 75rem = 1200px, the content width the design system specifies, with
   32-48px of horizontal padding. Individual panels still cap their own
   measure well below this — a 1200px line of body copy is unreadable —
   so this widens the page, not the prose. */
.wrap {
  width: 100%;
  max-width: 75rem;
  margin-inline: auto;
  padding-inline: clamp(1.25rem, 4vw, 3rem);  /* 20px gutter minimum */
}

header.site, footer.site { border-block: 1px solid var(--line); }
header.site { border-top: 0; padding-block: 1rem; }
footer.site { margin-top: 4rem; padding-block: 1.5rem; border-bottom: 0; }
footer.site p { margin: 0; color: var(--ink-soft); font-size: 0.875rem; }

.mark { color: var(--accent); margin-right: 0.5rem; }
.wordmark { font-weight: 600; letter-spacing: -0.01em; }

.hero { padding-block: clamp(2.5rem, 8vw, 5rem) 1rem; }

/* HEADINGS ARE MANROPE, BODY IS DM SANS. 7 Nov 2026.

   The pairing /our-mission already uses, now applied site-wide — see
   the long note on `body` above for why, and for why it costs no new
   request.

   WHY A SEPARATE RULE RATHER THAN A FAMILY ON EACH HEADING. There was
   no global heading rule in this file at all: h1 set size, line-height
   and measure, and h2/h3 were styled only in context (.prose h2,
   .signup h2, .breakdown h2 and a dozen more). Every one of them
   inherited its family from body. So one rule here reaches all of
   them, and a new heading added later is covered without anyone
   remembering to set it.

   LETTER-SPACING IS NOT SET HERE. Manrope at display sizes wants
   tightening and the existing rules already do it per-heading — h1 is
   -0.025em, .prose h2 and the rest carry their own. A blanket value
   would fight them at equal specificity and win or lose by source
   order, which is how the step-indicator font bug at the foot of this
   file happened. mission.css:527 sets -.055em on its own headings;
   that page keeps its own rhythm.

   .sitenav .brand ALREADY NAMED MANROPE and still does, a few hundred
   lines down, with its own size and weight. It is not made redundant:
   it is the wordmark, not a heading, and it is an <a>. */
h1, h2, h3, h4, h5, h6 {
  font-family: Manrope, system-ui, -apple-system, "Segoe UI", Roboto, sans-serif;
}

h1 {
  margin: 0 0 1rem;
  font-size: clamp(2rem, 1.4rem + 3vw, 3.25rem);
  line-height: 1.1;
  letter-spacing: -0.025em;
  max-width: 18ch;
}

.lede { max-width: var(--measure); color: var(--ink-soft); font-size: 1.125rem; }

.signup { padding-block: 1rem 2rem; max-width: var(--measure); }
.signup h2 { font-size: 1.375rem; letter-spacing: -0.01em; margin-block: 2rem 1rem; }

label { display: block; font-weight: 500; margin-bottom: 0.375rem; }

input[type="email"], input[type="text"] {
  width: 100%;
  padding: var(--ctrl-pad-y) var(--ctrl-pad-x);
  font: inherit;
  color: var(--ink);
  background: var(--paper);
  border: 1px solid var(--line);
  border-radius: var(--radius);
}

input[type="email"]:focus-visible,
input[type="checkbox"]:focus-visible,
button:focus-visible,
a:focus-visible {
  outline: 2px solid var(--accent);
  outline-offset: 2px;
}

/* THE CONSENT BLOCK. The WORDING inside the label is CONSENT_TEXT,
   rendered verbatim and stored as the consent record (D7/G4) — it is
   not editable here and nothing below changes a character of it.
   Only the box around it is styled.

   It sits on a panel so it reads as a distinct decision rather than
   as another paragraph, and the label is --ink rather than the
   muted --ink-soft: this is the sentence a person is agreeing to,
   and agreement text set in grey is the oldest dark pattern there
   is. 14.12:1 on the panel. */
/* `minmax(0, 1fr)`, NOT `1fr`. A bare 1fr has an auto minimum, so the
   label column cannot shrink below its longest word and the block
   pushed the page sideways by 33px at 200% text on a 360px phone
   (WCAG 1.4.10). The consent TEXT is untouched -- it is an audit
   record and is never edited for layout; only the track it sits in
   changes. */
.consent {
  display: grid;
  grid-template-columns: auto minmax(0, 1fr);
  /* The column gap is capped for the same reason as the padding
     below: at 200% a 0.75rem gap is 24px taken from a label that had
     23px left. Rows keep the full rem gap. */
  gap: 0.75rem min(0.75rem, 3vw);
  align-items: start;
  margin-block: 1.25rem 0;
  /* SIDE PADDING CAPPED AT LARGE TEXT (WCAG 1.4.10). 1.125rem became
     36px a side at 200%, which with the card's padding outside it and
     the checkbox track beside it left the consent label 23px to hold
     175px of text. min(…, 4vw) is the designed 18px at every width
     this project tests and shrinks only where the line would
     otherwise break out of the viewport. */
  padding: 1rem min(1.125rem, 4vw);
  background: var(--panel);
  border-radius: var(--radius) var(--radius) 0 0;
  max-width: var(--measure);
}
/* 1.5rem rather than 1.1rem: the box is the control, and the label
   beside it is already a click target through label[for]. */
.consent input {
  margin-top: 0.1rem;
  width: 1.5rem;
  height: 1.5rem;
  flex: none;
  accent-color: var(--brand);
}
.consent label {
  font-weight: 400;
  margin: 0;
  color: var(--ink);
  font-size: 1rem;
  cursor: pointer;
}

/* The primary action is the orange field with DARK ink on it
   (8.67:1). White on this orange is 1.8:1 and must never be used —
   that is why --cta carries its own --on-cta rather than reusing
   --accent-ink.

   THE BORDER IS --accent, NOT --cta, and that is an accessibility
   fix rather than decoration. WCAG 1.4.11 wants the control itself
   distinguishable from its surroundings at 3:1, and the orange fill
   measures only 1.79:1 against the cream page — the button's text
   passes comfortably while the button's SHAPE does not. An --accent
   edge at 13.64:1 gives the control a discernible boundary without
   changing the approved orange fill.

   THE FIGURE WAS 7.54:1 UNDER THE PURPLE PALETTE and is recomputed
   for v2 rather than carried: --on-cta is navy #142B49, which is
   13.64:1 on the cream page. The orange fill's own 1.79:1 is
   unchanged, because the fill did not move.

   ⚠️ IT IS --on-cta, NOT --accent, SINCE 23 OCT 2026, and the
   difference only shows in dark mode. --accent is navy in light but
   ORANGE (#FFAA70) in dark, so the edge became orange-on-orange
   there: measured rgb(255,170,112) border on a rgb(255,170,112)
   fill, which is no edge at all. The quote-flow v2 boards set it
   explicitly in BOTH themes --

       .cta-btn{border-color:#142B49 !important;color:#142B49 !important}

   -- and Guillaume's brief says it in as many words: "anything on an
   orange fill keeps NAVY text and border in both themes (--on-cta
   #142B49)". --on-cta is navy by definition in both, which is
   exactly the guarantee this edge needs.

   THE OLD NOTE SAID THE DARK BORDER "COSTS NOTHING EITHER", which
   was true of contrast and wrong about the shape: the fill is 9.85:1
   against the dark page, so the control is findable, but it had no
   discernible edge and did not match the board. Both now hold. */
button {
  width: 100%;
  margin-top: 0.5rem;
  padding: var(--btn-pad-y) var(--btn-pad-x);
  font: inherit;
  font-weight: 700;
  color: var(--on-cta);
  background: var(--cta);
  border: 2px solid var(--on-cta);
  border-radius: var(--radius);
  cursor: pointer;
  min-height: 44px;           /* touch target */
}
/* The generic hover, for buttons that are NOT one of the two CTA
   roles (the mail viewer's controls, the car-list "Change", and so
   on). The raised roles set their own hover background and must not
   also get a filter: it multiplies the colour the hover just chose
   and it repaints the 4px slab, which made the edge shimmer. */
button:not(.cta-btn):not(.cta-sec):not(.action-join):not(.btn-primary):not(.action-decline):not(.btn-secondary):hover {
  filter: brightness(1.08);
}

@media (min-width: 34rem) {
  button { width: auto; min-width: 14rem; }
}

.fineprint { margin-top: 0.875rem; font-size: 0.875rem; color: var(--ink-soft); }

/* ⚠️ A LINK INSIDE .fineprint HAD NO RULE, 8 Nov 2026 — THE SAME
   FAULT AS .err-jump BELOW, FOUND THE SAME WAY.

   .fineprint set a colour on the PARAGRAPH, and an <a> inside it does
   not inherit colour: it takes the browser's default link blue,
   rgb(0,0,238) in light. Guillaume screenshotted the email step and
   "Read our Privacy Notice" was bright blue under grey body text,
   beside a navy "Back to home" in the same header.

   .contact-fineprint a already had this rule, which is why the
   contact page looked right and this one did not — the class with the
   rule was the one anybody had looked at.

   --accent, matching .contact-fineprint a, so the two fineprint
   blocks are one treatment rather than two. The underline STAYS: this
   is body-size text in a paragraph, where colour alone would carry
   the affordance and fail WCAG 1.4.1. */
.fineprint a { color: var(--accent); }

.result {
  padding: 0.875rem 1rem;
  border-radius: var(--radius);
  margin-bottom: 1rem;
}
.result.ok  { background: var(--ok-bg);  color: var(--ok-ink); }
.result.err { background: var(--err-bg); color: var(--err-ink); }
/* ⚠️ THE "Go to the field" LINK HAD NO RULE AT ALL, 23 Oct 2026.
   .err-jump carried a class and nothing styled it, so it fell back
   to the browser's default link blue -- rgb(0,0,238) in light and
   rgb(158,158,255) in dark, measured on the real error state.

   IT WAS NOT A CONTRAST FAILURE (8.98:1 and 7.08:1, both past AA)
   and that is exactly why nothing caught it: the contrast walker
   reads ratios and was content, every test passed, and the colour
   was simply not ours. It showed up as a purple link inside a red
   box in the dark-mode screenshot of a submitted empty form -- a
   state no page load reaches.

   currentColor, NOT a token of its own: the link sits inside
   .result.err and should read as part of that sentence, which is
   already --err-ink on --err-bg. The underline is what marks it as
   a link, so colour is not carrying that job alone (1.4.1). */
.err-jump { color: currentColor; text-decoration: underline; }

.notice {
  padding: 1.25rem;
  border: 1px dashed var(--line);
  border-radius: var(--radius-panel);
}
.notice h2 { margin-top: 0; }

/* Honeypot: off-screen rather than display:none, which some bots detect. */
.hp {
  position: absolute;
  left: -9999px;
  width: 1px; height: 1px;
  overflow: hidden;
}

.back { color: var(--accent); }

@media (prefers-reduced-motion: reduce) {
  *, *::before, *::after { animation: none !important; transition: none !important; }
}

/* Staging banner. Only rendered when ENVIRONMENT_BANNER is set, which
   production never sets. Loud on purpose: this exists so nobody mistakes
   a test deployment for the real site. */
.envbanner {
  background: #b4210a;
  color: #fff;
  font-weight: 600;
  text-align: center;
  padding: .6rem 1rem;
  letter-spacing: .01em;
  position: sticky;
  top: 0;
  z-index: 100;
}

/* Long-form legal copy (/privacy). The other pages are a headline and a
   paragraph; this one needs headings, a table and lists, so the rules
   live here rather than being inherited from .lede.

   WIDER THAN --measure SINCE 10 Oct 2026. Guillaume, reading /privacy on
   a desktop: "the privacy notice on the desktop feels like its mobile".
   He was right, and --measure is the reason: 34rem = 544px, which at a
   1280px viewport is 42% of the screen and on a 2560px monitor is a
   narrow column in a field of white.

   544px was also slightly TIGHT as a measure, not merely narrow. Body
   text here is clamp(1.125rem, 1.1rem + 0.15vw, 1.1875rem) = 18-19px,
   so 544px carries about 58 characters. The long-form reading range is
   60-75; this sat just under it.

   42.5rem = 680px carries about 70 characters at the same size, which
   is mid-range and reads as a document rather than a phone page.

   --MEASURE ITSELF IS NOT TOUCHED, DELIBERATELY. It is used by .lede,
   .signup, the quote flow (app.css:348) and others, all of which were
   measured against the boards and are correctly narrow. Widening the
   token would stretch the homepage lede and the signup column as a side
   effect. This is a legal-page rule, so it lives on .prose.

   The table still collapses to stacked rows on narrow screens rather
   than scrolling sideways. */
.prose {
  max-width: 42.5rem;
  padding-bottom: 2rem;
}

/* .prose INSIDE THE SHARED CHROME, 9 Nov 2026 — /about (D27).

   .prose above is a STANDALONE-PAGE rule: /privacy has its own header
   and footer, no .wrap around it, so the column starts at the page
   gutter and looks right there. /about takes the shared chrome
   instead, because it is in the nav — so its .prose sits inside
   .wrap, which is 75rem. Without this the 42.5rem column sits hard
   against the left gutter while the nav and footer span the full
   width, leaving the right half of a 1280px page empty.

   MEASURED, 1280px, before this rule: .prose ran left 81 to right
   761 inside a .wrap running 33 to 1233. /get-in-touch's column, the
   SAME 42.5rem, ran 293 to 973 — centred. The page looked misaligned
   against its own siblings.

   THIS IS THE FAULT .contactpage's comment ALREADY DESCRIBES ("the
   column sits hard against the left gutter — which is exactly the
   fault found on 3 Oct 2026"). Same cause, same fix, recorded here
   too because the next prose page in the chrome will hit it again.

   WHY NOT just add margin-inline to .prose: that would centre
   /privacy's column as well, and /privacy is a standalone page that
   is signed off as it is. A modifier changes the new page only.

   WHY NOT reuse .contactpage: it is the right behaviour under the
   wrong name, and a third page using a class called "contactpage"
   is how a stylesheet stops being readable. */
.prose-centred {
  margin-inline: auto;
}

.prose h2 {
  /* 2.75rem top, 10 Oct 2026: with body gaps at 1.25rem a 2rem heading gap
     was only 0.75rem more than an ordinary paragraph break, so sections did
     not read as sections. The small bottom margin is deliberate and stays —
     a heading belongs to the text under it. */
  margin-block: 2.75rem 0.5rem;
  font-size: 1.1875rem;
  letter-spacing: -0.01em;
}

/* Sub-headings inside a section — the privacy notice uses them to
   separate "if you get a quote" from "if you join the waitlist", which
   are two genuinely different collections. Sized below h2 so the
   hierarchy is visible, not just semantic. */
.prose h3 {
  margin-block: 2rem 0.5rem;
  font-size: 1.0625rem;
  letter-spacing: -0.005em;
}

.prose h4 {
  margin-block: 1.25rem 0.375rem;
  font-size: 1rem;
}

/* VERTICAL RHYTHM, OPENED UP 10 Oct 2026. Guillaume, on the desktop page:
   "spacing is still rediculously small on the website desktop view".
   Widening the measure earlier the same day did not touch this, and the
   gaps were the real complaint.

   0.75rem = 12px between paragraphs, against a LINE that is ~29px tall
   (18-19px type at 1.55 line-height). A gap under half a line does not
   separate anything: consecutive paragraphs read as one block of text.
   1.25rem = 20px is about two thirds of a line, which reads as a break
   without floating the paragraphs apart. */
.prose p, .prose ul { margin-block: 1.25rem; }
.prose ul { padding-left: 1.5rem; }
.prose li { margin-block: 0.5rem; }
.prose a { color: var(--accent); }

/* THE UPDATE BANNERS. Nine <blockquote>s carry the dated "Updated 7
   October 2026" notices at the top of /privacy, and until 10 Oct 2026
   NOTHING IN THIS FILE STYLED THEM. They ran on the browser default,
   `margin: 1em 40px`, which is why Guillaume saw the horizontal width
   as "strange": the h1 and the "Last updated" line sat at the column's
   left edge while every banner paragraph was indented 40px past it and
   lost 80px of measure. Two different left margins on one page reads as
   broken rather than narrow, and widening .prose could not fix it
   because the indent is applied inside .prose.

   They are now flush with the column and marked as an aside by a rule
   and a tint instead of by indentation — the same device the page
   already uses for .notyet, so the banners read as the same KIND of
   thing. padding-left carries the text clear of the rule; the text
   itself still starts at the column edge plus that padding, which is
   12px rather than 40px. */
.prose blockquote {
  margin: 1.5rem 0;
  padding: 0.75rem 1rem;
  border-left: 3px solid var(--accent);
  background: color-mix(in srgb, var(--accent) 6%, transparent);
  border-radius: 0 6px 6px 0;
}

.prose blockquote > p:first-child { margin-block-start: 0; }
.prose blockquote > p:last-child { margin-block-end: 0; }

.prose .updated {
  color: var(--ink-soft);
  font-size: 0.875rem;
  margin-top: 0;
}

.prose .address { color: var(--ink-soft); line-height: 1.7; }

.prose code {
  font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
  font-size: 0.9em;
  padding: 0.1em 0.3em;
  background: var(--line);
  border-radius: 4px;
}

.prose table {
  width: 100%;
  border-collapse: collapse;
  margin-block: 1rem;
  font-size: 0.9375rem;
}
.prose th, .prose td {
  text-align: left;
  vertical-align: top;
  padding: 0.625rem 0.75rem 0.625rem 0;
  border-bottom: 1px solid var(--line);
}
.prose th { font-weight: 600; }
.prose td:last-child { color: var(--ink-soft); }

/* Below ~34rem a two-column table of sentences is unreadable, so the
   cells stack and the header row is dropped. */
@media (max-width: 34rem) {
  .prose thead { position: absolute; left: -9999px; }
  .prose tr {
    display: block;
    padding-block: 0.75rem;
    border-bottom: 1px solid var(--line);
  }
  .prose td {
    display: block;
    border: 0;
    padding: 0;
  }
  .prose td:first-child { font-weight: 600; }
  .prose td:last-child { margin-top: 0.25rem; }
}


/* ===================================================================
   The advisory quote flow (docs/product/quote-flow-spec.md)
   ===================================================================

   Desktop and mobile equally, as everywhere else here: the breakdown
   is designed as a table at both ends and the layout checked at the
   awkward widths between, not reflowed from one to the other.

   Contrast ratios below were computed, not eyeballed, the same way the
   code review's Part 4 did it. Every value quoted is light/dark. */

/* A4 from the code review: --line fails WCAG 1.4.11's 3:1 for the
   boundary of a form control, so control edges use --field-border and
   --line is left for decorative rules only. That split is unchanged;
   the VALUES moved to the Open-sky palette and were recomputed.

   --panel is NOT redefined here any more. It used to be, with a grey
   #f7f8fb that silently overrode anything set above — and it now
   carries the pale sky from the token block, which is the surface the
   handoff specifies for explanatory panels.

   RECOMPUTED FOR V2, 22 Oct 2026, light / dark:
     --field-border on --paper  14.27:1 /  9.76:1  (needs 3:1)
     --up   on --page            6.11:1 /  8.56:1
     --down on --page            6.28:1 /  9.72:1
     --up   on --panel           5.41:1 /  6.29:1
     --down on --panel           5.56:1 /  7.15:1

   --up and --down are UNCHANGED BY V2 and that is deliberate: they are
   the "costs more" / "costs less" pair in the advisory breakdown, and
   their meaning is amber-vs-green, not brand-vs-brand. Recolouring
   them to the new brand would have thrown away the one place on this
   site where hue carries information rather than identity. Only their
   measured ratios moved, because the surfaces beneath them did.

   THE DARK VALUES APPEAR TWICE for the same reason the main palette's
   do: the media query is the no-JS fallback and the attribute selector
   is the explicit choice. See the long note at the token block. */
/* ⚠️ --down-track AND --down-line ARE THE SAVINGS BARS' TWO TINTS,
   23 Oct 2026, named because the v2 flow boards name them.

   WHAT THEY REPLACE. The bars tinted --down inline:

       .savings-track        color-mix(in srgb, var(--down) 12%, transparent)
       .savings-track--today color-mix(in srgb, var(--down) 28%, transparent)

   That WORKED -- color-mix tracks the theme, so dark mode was never
   broken here -- but it put two palette decisions inside two layout
   rules, where the flow boards expect a token. Naming them means the
   tints can be tuned, or diverge from a straight tint of --down,
   without editing the bar rules.

   THE DARK VALUES ARE THE BOARDS' OWN: rgba(127,214,164,0.16) and
   rgba(127,214,164,0.30) -- that is --down at 16% and 30%. THE LIGHT
   BOARD DOES NOT GIVE VALUES: ResultV2.dc.html defines all three as
   `--down:var(--down)`, a self-reference, which is invalid and
   resolves to nothing. So the light values are DERIVED at the dark
   board's own opacities from light --down, which is what the dark
   board is. Reported to Guillaume rather than silently invented.

   THE OPACITIES MOVE 12%->16% AND 28%->30%, taking the boards'
   figures over the shipped ones. Composited on canvas against
   --ok-bg, the card they sit on -- a tint cannot be judged from its
   token text, only from what it resolves to over its ground:

       --down-track on --ok-bg   1.27:1 light / 1.43:1 dark
       --down-line  on --ok-bg   1.59:1 light / 2.02:1 dark
       --down (the FILL)         5.96:1 light / 8.16:1 dark

   NEITHER TINT REACHES 3:1 AND NEITHER NEEDS TO. 1.4.11 binds on a
   graphic that is "required to understand the content"; these two
   are the unfilled remainder of a bar and the neutral reference bar
   behind it. The information is carried by the FILL, which is --down
   at full strength and passes comfortably, and by the figure printed
   beside every bar in text -- the bars are a redundant picture of a
   number that is already written. Removing both tints entirely would
   lose decoration, not meaning.

   THE EARLIER VERSION OF THIS NOTE CLAIMED 3.18:1 AND 3.41:1 and
   concluded the line "passes". Those numbers were estimated, not
   measured; the real figures are above and they are roughly half.
   The conclusion changes with them: the line does not pass 3:1 and
   does not have to, which is a different argument from the one that
   was written here. */
:root {
  --field-border: #142B49;
  --up:           #8A5200;
  --down:         #0F6B3F;
  --down-track:   rgba(15, 107, 63, 0.16);
  --down-line:    rgba(15, 107, 63, 0.30);
}

@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) {
    --field-border: #A9DCFA;
    --up:           #E0B070;
    --down:         #7FD6A4;
    --down-track:   rgba(127, 214, 164, 0.16);
    --down-line:    rgba(127, 214, 164, 0.30);
  }
}
:root[data-theme="dark"] {
  --field-border: #A9DCFA;
  --up:           #E0B070;
  --down:         #7FD6A4;
  --down-track:   rgba(127, 214, 164, 0.16);
  --down-line:    rgba(127, 214, 164, 0.30);
}

/* A3: skip link. Visible on focus, and it must sit above the sticky
   staging banner or it lands underneath it. */
.skiplink {
  position: absolute;
  left: -9999px;
  top: 0;
  z-index: 200;
  padding: 0.75rem 1rem;
  background: var(--paper);
  color: var(--accent);
  border: 2px solid var(--accent);
  border-radius: 0 0 var(--radius) 0;
}
.skiplink:focus {
  left: 0;
}

/* A4: every field boundary, not just the new ones. */
input[type="email"], input[type="text"], input[type="number"], select {
  border-color: var(--field-border);
}

/* type="number" is new in this flow, so it misses the base rule above
   that sizes email and text. Without this the age field renders as a
   ~110px box — narrower than the label above it, and reading as
   disabled rather than as the first question on the page.

   The width is capped rather than 100%: an age is at most three
   characters, and a full-width box for three characters invites the
   thought that something longer is wanted. 8rem is wide enough for the
   spinner controls Chrome and Safari add. */
input[type="number"] {
  width: 100%;
  max-width: 8rem;
  padding: var(--ctrl-pad-y) var(--ctrl-pad-x);
  font: inherit;
  color: var(--ink);
  background: var(--paper);
  border: 1px solid var(--field-border);
  border-radius: var(--radius);
  min-height: 44px;
}
input[type="checkbox"], input[type="radio"] {
  accent-color: var(--accent);
}

input[type="number"]:focus-visible,
input[type="text"]:focus-visible,
input[type="radio"]:focus-visible,
select:focus-visible,
summary:focus-visible,
[role="combobox"]:focus-visible {
  outline: 2px solid var(--accent);
  outline-offset: 2px;
}

/* D5 / C1. Persistent, never collapsible, and always above the number.
   §10 requires it visible without scrolling at 360px, so it is plain
   flow content at the top of <main> — not sticky, not dismissible, and
   not something that can scroll out from under the figure.

   THE HOMEPAGE NO LONGER USES THIS. Since the board rebuild on
   22 Sep 2026 it carries the same notice, the same words, as
   .topbanner — a full-width dark bar above the nav, which is what the
   board shows. This rule still styles the notice on the other eleven
   templates and is not dead. The two must not diverge in WORDING; a
   test asserts the homepage's copy verbatim. */
.notyet {
  margin: 1.25rem 0 0;
  padding: 0.875rem 1rem;
  background: var(--panel);
  border-left: 4px solid var(--accent);
  border-radius: 0 var(--radius) var(--radius) 0;
  color: var(--ink);
  font-size: 0.9375rem;
  max-width: 44rem;
}

/* D9. The definition of the coined term. Deliberately not small print:
   a term we invent carries whatever meaning the reader supplies unless
   we supply it, so it reads at body size. */
.definition {
  max-width: var(--measure);
  color: var(--ink-soft);
  margin: 0.5rem 0 0;
}
.definition-standalone {
  margin-block: 1.5rem;
  padding-left: 1rem;
  border-left: 3px solid var(--line);
  color: var(--ink);
}

/* ---- Screen 2: the questions ---------------------------------- */

/* --measure is a READING measure, 34rem, and capping the form column
   with it was the bug behind three separate complaints.

   Measured at 993px: .formgrid offered the main column 914px and
   .questions capped its contents at 544, leaving ~400px of empty page
   beside a cramped form. It did that at EVERY width from 900 to 1440 —
   the column never reclaimed the space the stacked sidebar gave back.

   It also squeezed everything inside: radio cards to 173px on both
   steps (not just step 1 — step 2's 266px was measured at 660px, where
   the column is narrower than the cap, so the cap was invisible), the
   title select to 49px, and the three-up name row to 171px per field.

   A form column is not prose. Its width should come from the grid; the
   grid is what decides how much room a form gets. The measure still
   governs PROSE inside it — see .questions p below — which is what
   --measure is actually for. */
.questions { padding-block: 1.5rem 3rem; }

/* Paragraphs inside the form keep the reading measure. A line of help
   text 900px wide is hard to read; a row of radio cards is not. */
.questions > p,
.questions .flow-lede,
.questions .infonote,
.questions .basis { max-width: var(--measure); }
.questions h1 { font-size: clamp(1.75rem, 1.3rem + 2vw, 2.5rem); }

/* NOT tokenised, deliberately. A .field OUTSIDE a fieldgroup — the
   single email field on the join and result screens — keeps the
   1.5rem it has always had. The scale below governs the
   QUESTIONNAIRE's rhythm, which is what the phone boards draw; it
   is not a licence to re-space every form in the app.

   `.fieldgroup .field` overrides this with var(--field-gap) and is
   the rule that actually sets the questionnaire's rhythm. */
.field { margin-block: 1.5rem; }
.field label { display: block; font-weight: 500; margin-bottom: var(--label-gap); }

.help {
  margin: 0.375rem 0 0;
  font-size: 0.875rem;
  color: var(--ink-soft);   /* 7.96:1 / 8.25:1 */
}
.help code {
  font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
  font-size: 0.9em;
  padding: 0.1em 0.3em;
  background: var(--line);
  border-radius: 4px;
}

/* The per-field error. Carries its own text rather than relying on
   colour (WCAG 1.4.1), and is referenced by aria-describedby. */
.fielderr {
  margin: 0.5rem 0 0;
  padding: 0.5rem 0.75rem;
  background: var(--err-bg);
  color: var(--err-ink);     /* 8.0:1 / 7.9:1 */
  border-radius: var(--radius);
  font-size: 0.9375rem;
}

input[aria-invalid="true"] {
  border-color: var(--err-ink);
  border-width: 2px;
}

select {
  width: 100%;
  padding: var(--ctrl-pad-y) var(--ctrl-pad-x);
  font: inherit;
  color: var(--ink);
  background: var(--paper);
  border: 1px solid var(--field-border);
  border-radius: var(--radius);
  min-height: 44px;
}

/* ---- The car combobox ------------------------------------------
   `hidden` is honoured explicitly because the flex/grid rules below
   would otherwise override it — a display value on a [hidden] element
   is the classic way to un-hide something by accident. */
[hidden] { display: none !important; }

.combo { position: relative; }
.combo-label {
  position: absolute;
  left: -9999px;
}

/* The listbox scrolls, and Chrome makes any scrollable container a
   keyboard tab stop ("keyboard-focusable scrollers") even with
   tabIndex -1. That put an extra Tab stop between the car field and
   the fallback disclosure, carrying a default browser outline and
   breaking the ARIA 1.2 combobox pattern, in which focus stays on the
   INPUT and aria-activedescendant tracks the highlighted option.

   The options are reachable exactly as the pattern intends — arrow
   keys, Enter, Escape, all handled in quote.js — so removing this
   stop removes a duplicate route, not the only one.

   The scrolling this would otherwise provide is not lost: arrow keys
   move the active option and quote.js scrolls it into view. */
.listbox {
  list-style: none;
  margin: 0.25rem 0 0;
  padding: 0;
  max-height: 16rem;
  overflow-y: auto;
  border: 1px solid var(--field-border);
  border-radius: var(--radius);
  background: var(--paper);
  display: none;
}
.listbox.is-open { display: block; }

.option {
  padding: 0.6875rem 0.875rem;
  cursor: pointer;
  border-bottom: 1px solid var(--line);
  min-height: 44px;           /* touch target */
}
.option:last-child { border-bottom: 0; }
.option.is-active {
  background: var(--accent);
  color: var(--accent-ink);
}
.option-name { display: block; }
.option-note {
  display: block;
  font-size: 0.8125rem;
  color: var(--ink-soft);
}
.option.is-active .option-note { color: var(--accent-ink); }
.option-empty { cursor: default; }

.chosen {
  margin: 0.5rem 0 0;
  font-size: 0.9375rem;
}
.linkbtn {
  width: auto;
  min-width: 0;
  margin: 0 0 0 0.5rem;
  padding: 0.25rem 0.5rem;
  min-height: 0;
  background: none;
  border: 0;
  color: var(--accent);
  font: inherit;
  text-decoration: underline;
  cursor: pointer;
}

/* C9b, the always-available fallback. A <details> because it is
   genuinely supplementary — but it is beneath the field, always
   visible, and never the only route to the five categories. */
.fallback {
  margin-top: 1rem;
  border: 1px solid var(--line);
  border-radius: var(--radius);
  padding: 0.75rem 1rem;
}
/* The summary is a control, so it keeps a 44px touch target — but it
   must not read as an empty input. A disclosure triangle and a tighter
   box distinguish it from the fields above it; without them a bordered
   44px-tall rectangle containing text is exactly what the postcode
   field looks like. */
.fallback {
  padding: 0;
}
.fallback summary {
  cursor: pointer;
  font-weight: 500;
  min-height: 44px;
  display: flex;
  align-items: center;
  gap: 0.5rem;
  padding: 0.5rem 1rem;
  color: var(--accent);
}
.fallback summary::marker { content: ""; }
.fallback summary::-webkit-details-marker { display: none; }
.fallback summary::before {
  content: "\25B8";
  flex: none;
  transition: transform 0.15s;
}
.fallback[open] summary::before { transform: rotate(90deg); }
.fallback .catgroup { padding: 0 1rem 0.75rem; }

.catgroup, .reasons {
  border: 0;
  margin: 0.5rem 0 0;
  padding: 0;
}
.catgroup legend, .reasons legend {
  padding: 0;
  font-weight: 500;
  margin-bottom: 0.5rem;
}

.radio {
  display: flex;
  align-items: center;
  gap: 0.625rem;
  padding: 0.5rem 0;
  font-weight: 400;
  min-height: 44px;
  cursor: pointer;
}
.radio input { width: 1.1rem; height: 1.1rem; flex: none; }

/* ---- Screen 3: the advisory quote ------------------------------ */

.quoteblock { padding-block: 2rem 0; }

/* ---- The email and waitlist card (5 Oct 2026) -------------------

   Result.dc.html: white paper, a 2px brand edge, a 20px radius and a
   720px cap, immediately under the price.

   IT IS A CARD AND NOT A BAND because it is one ask among the page's
   other blocks, all of which are cards. The brand edge rather than the
   panel fill the other panels use: this is the only block on the
   page that asks for something, and the edge is what separates an ask
   from the explanations around it without making it louder than the
   number above it.

   45rem AND NOT --measure. --measure is 34rem, a READING measure, and
   this card holds a form: capping a form at a reading measure is the
   mistake quote-layout R2 exists to stop — it put the step-1 column at
   544px at every width from 900 to 1280. 45rem is 720px, which is the
   board's figure.

   NO max-width ON THE BUTTONS. They are .action inside .actions, which
   is flex: 1 1 14rem — the same rule that enforces §7.7's equal weight
   on the pair this replaced. Nothing here touches .action, deliberately:
   a size or a weight set on one of the two buttons from this block
   would be the quiet way the equal-weight rule gets undone. */
.emailcard {
  margin-block: 2rem;
  padding: 1.5rem min(1.625rem, 5vw);
  max-width: 45rem;
  background: var(--paper);
  /* --accent, the token every focusable edge on this page uses:
     7.54:1 on --paper light, 11.56:1 dark. The decorative --line at
     1.25:1 would not read as a boundary. */
  border: 2px solid var(--accent);
  border-radius: 1.25rem;
}
/* .emailcard-service WAS HERE AND IS DELETED, 9 Oct 2026.
   ------------------------------------------------------------------
   It styled the C33 paragraph — "Your advisory quote is on its way by
   email..." — which stood above the form between 7 and 9 Oct with a
   border-bottom separating it from the consent checkbox below.

   Boards version 134 draws one card with no such paragraph, so the
   markup went and these rules went with it. The class appears nowhere
   in web/templates/ now; `grep -rn emailcard-service web/` returns
   this comment and a passing mention in the color-mix note below.

   Deleted rather than left in place: a selector with no element is a
   claim that the page has a block it does not have, and the next
   person to style this card would read it as the house pattern for
   the ask. The REASONING it carried — that a disclosure must not read
   as a caption to a consent checkbox — is still live and is recorded
   in quote_result.html where the removal is explained. */
.emailcard-h {
  margin: 0 0 0.375rem;
  font-size: clamp(1.1875rem, 1.05rem + 0.6vw, 1.375rem);
  font-weight: 800;
  line-height: 1.2;
  /* The heading is a question and it is allowed to wrap; it is not a
     figure. No max-width: the card already caps it. */
  max-width: none;
}
.emailcard-lede {
  margin: 0 0 1rem;
  font-size: 0.9375rem;
  color: var(--ink-soft);   /* 7.96:1 / 8.25:1 */
  line-height: 1.5;
}
/* The email field inside the card uses the SHARED .field rule. Only
   the 34rem cap is lifted, because here the container is the cap and
   two different widths inside one card reads as a mistake.

   `.emailcard .consent` WAS IN THIS SELECTOR LIST AND IS GONE, 9 Oct
   2026. The card no longer renders a .consent block: the audit string
   now sits in .optin-consent below the tick, and .consent itself is
   still live on index.html and quote_join.html, where it is unchanged.
   .consent-link stays — the Privacy Notice link is still rendered
   here, outside the label (review A5) — but it needs MORE than the
   cap lifted now. See the override below. */
/* THE PRIVACY LINK IS A PLAIN LINE ON THIS CARD, NOT A PANEL FOOTER.
   ------------------------------------------------------------------
   FOUND BY LOOKING AT THE BOARD-VS-PAGE IMAGE, 9 Oct 2026, which is
   the only check that would have caught it.

   The shared .consent-link rule is built to CONTINUE the .consent
   panel directly above it: it repeats --panel as its background,
   rounds only its BOTTOM corners, and indents its text by
   calc(1.5rem + 0.75rem + 1.125rem) so it lines up past the checkbox
   with the label beside it. All three are correct on index.html and
   quote_join.html, where a .consent panel really is above it.

   ON THIS CARD THERE IS NO .consent PANEL ANY MORE. The audit string
   moved into .optin-consent, which is plain text. So the shared rule
   drew a panel box with two square top corners and a 54px left
   indent, floating under an unrelated paragraph — it read as a second
   control that had half-rendered.

   Every text check passed it: the right words, in the right order, no
   overflow. It is a SHAPE fault, which is the class of thing only the
   side-by-side image shows.

   So here it is a line of text that follows the consent sentence,
   with no fill, no radius and no indent. The shared rule is untouched
   for the two pages that still want the panel. */
.emailcard .consent-link {
  max-width: none;
  margin: 0 0 0.25rem;
  padding: 0;
  background: none;
  border-radius: 0;
  font-size: 0.8125rem;
}
.emailcard .field:first-of-type { margin-top: 0; }
/* `.emailcard .actions` AND ITS 17.5rem DESKTOP CAP ARE DELETED,
   9 Oct 2026.
   ------------------------------------------------------------------
   They sized the single "Join the waitlist" button of the C34 card:
   full width on phone, capped at 280px above 48rem, because
   Result.dc.html drew it that way inside a two-card layout.

   Boards version 134 replaces that button with ONE full-width raised
   orange CTA and the card no longer renders a .actions row at all —
   see .emailcard-actions below, which is its replacement and
   deliberately not a reuse of .actions.

   WHY NOT REUSE .actions: it is a flex row built for a PAIR of
   equal-weight buttons (§7.6), still live on the decline screen, with
   a 2rem top margin and a 44rem cap this card wants none of. Pointing
   a one-button card at a two-button rule is how the 280px cap came to
   need a media query in the first place. */

/* ---- THE WAITLIST OPT-IN, inside the email card -------------------
   ADDED 9 OCT 2026 (Guillaume, boards version 134). The card is ONE
   card with ONE button again, and the waitlist is an optional tick
   between the field and the button rather than a second card with its
   own heading and its own form.

   .optin IS A <label> WRAPPING BOTH THE BOX AND ITS TEXT, which is
   why it is not .consent. The two are deliberately different things
   and must not be merged:

     .consent  — the audit record. Its label is {{ .ConsentText }},
                 rendered verbatim, and its SHA-256 is what is stored
                 against every consent event (G4). Still used by the
                 homepage form and quote_join.html.
     .optin    — the UI control for this card. Its text is "Also put
                 me on the 2027 waitlist", a UI label the board wrote,
                 and the verbatim consent string sits BELOW it in
                 .optin-consent.

   A label wrapping the input means the whole panel is the hit target
   without a `for`/`id` round trip, and it is how ResultPhone draws it.
   The `for` attribute is set anyway: belt and braces costs nothing and
   survives somebody later moving the input out of the label.

   VALUES FROM ResultPhone.dc.html: --panel fill, 12px radius, 12px
   14px padding, 12px gap, align-items flex-start so the box sits with
   the FIRST LINE of a two-line title rather than centring against the
   whole block.

   THE RADIUS AND PADDING ARE LITERALS FROM THE BOARD, in rem. 0.75rem
   is the board's 12px and matches --radius' intent at this size;
   0.75rem/0.875rem is 12px/14px. quote-layout R6 forbids inventing a
   value on neither scale — these are the board's own, which is the
   source R5 names for the phone column.

   SIDE PADDING CAPPED AT LARGE TEXT, the same WCAG 1.4.10 problem
   .consent documents: at 200% a fixed 14px a side plus the card's own
   padding plus the checkbox track squeezes the label past breaking.
   min(…, 3.5vw) is the designed 14px at every width this project
   tests and shrinks only where the line would otherwise break out of
   the viewport. */
.optin {
  display: flex;
  align-items: flex-start;
  gap: min(0.75rem, 3vw);
  margin-block: 1rem 0.75rem;
  padding: 0.75rem min(0.875rem, 3.5vw);
  background: var(--panel);
  border-radius: 0.75rem;
  cursor: pointer;
  max-width: none;
}
/* 20px, the board's figure. .consent uses 1.5rem/24px because it is
   the only control in its block; here the box sits beside a bold
   title and a note, and 24px overpowers them.

   flex: none so it never shrinks when the title wraps. accent-color
   paints the CHECKED state --brand, which is the board's value and
   the one place the brand colour appears inside this card. */
.optin input[type="checkbox"] {
  width: 1.25rem;
  height: 1.25rem;
  flex: none;
  /* Optically aligns the box with the cap-height of the title beside
     it rather than its line-box top. */
  margin-top: 0.125rem;
  accent-color: var(--brand);
  cursor: pointer;
}
.optin-text {
  display: flex;
  flex-direction: column;
  gap: 0.125rem;
  min-width: 0;             /* lets a long title wrap instead of pushing out */
}
.optin-title {
  font-weight: 700;
  color: var(--ink);
  line-height: 1.3;
}
.optin-note {
  font-size: 0.8125rem;     /* the board's 13px */
  color: var(--ink-soft);   /* 7.96:1 / 8.25:1 */
  line-height: 1.45;
}
/* THE VERBATIM CONSENT STRING. 13px per the board, --ink-soft, and it
   is a <p> rather than a <label>: it is not the control's name, it is
   the record of what ticking the control means. Making it a label
   would put two labels on one input. */
.optin-consent {
  margin: 0 0 0.375rem;
  font-size: 0.8125rem;
  color: var(--ink-soft);
  line-height: 1.55;
  max-width: none;
}
/* The button row. NOT .actions: that rule is a flex row built for a
   PAIR of equal-weight buttons (§7.6, still live on the decline
   screen) with a 2rem top margin and a 44rem cap. This card has one
   full-width button under a consent line it should stay close to.

   FULL WIDTH IN BOTH COLUMNS, which is a change from the button this
   replaced. ResultPhone draws `width: 100%`, and Result.dc.html's old
   280px cap belonged to the two-card layout where the button sat
   under a short heading. Guillaume's instruction for this card is a
   full-width raised orange CTA, so there is no desktop cap and no
   @media block here at all. */
.emailcard-actions {
  margin-top: 1rem;
  max-width: none;
}
.emailcard-actions .cta-btn {
  display: block;
  width: 100%;
}

.quotelabel {
  font-size: clamp(1.5rem, 1.2rem + 1.5vw, 2rem);
  margin-bottom: 0.5rem;
  max-width: none;
}

/* D3: the range, and the LARGEST ELEMENT ON THE PAGE. The design
   system asks for 64-72px on desktop; the lower bound is set so a
   long range ("£1,450 – £1,750") still fits a 360px line without
   wrapping mid-figure, which would break the number in half.

   Wrapping is prevented rather than shrunk-to-fit: the range is one
   idea and half of it on a second line reads as two numbers. */
.bignumber {
  margin: 0.75rem 0 0.5rem;
  font-size: clamp(2.125rem, 1.1rem + 5vw, 4.5rem);
  font-weight: 700;
  letter-spacing: -0.035em;
  line-height: 1.05;
  color: var(--ink);
  /* text-wrap: nowrap WAS HERE AND WAS A WCAG 1.4.10 FAILURE.

     The intent above is right — a range is one idea and half of it on
     a second line reads as two numbers — but `nowrap` enforces it by
     making the page scroll sideways instead.

     Measured, Galaxy S24 (360px) at 200% text, which is the 1.4.10
     test condition: the result page scrolled 141px. This element was
     the source at 181px over its container; .resulthead-copy,
     .quoteblock and .wrap were all just inheriting it.

     `balance` keeps the range on one line whenever it fits — which is
     every case except very large text on a narrow screen — and lets it
     break when the alternative is a horizontal scrollbar. The non-
     breaking spaces in the template keep "£1,400 – £1,750" from
     splitting at the spaces around the dash, so if it does break it
     breaks between the two figures and not inside one. */
  text-wrap: balance;
  /* A long range at 200% text still must not push the page out. */
  overflow-wrap: break-word;
}

/* LONG SOURCE NAMES MUST BREAK RATHER THAN PUSH THE PAGE OUT.

   The last 16px of the same 1.4.10 failure was a single token:
   "(MoneySuperMarket," — 18 characters with no break opportunity, in a
   200px column at 32px text. Measured: .improved-copy scrollWidth 296
   against clientWidth 200.

   overflow-wrap: break-word only breaks a word that would otherwise
   overflow, so ordinary prose is untouched at ordinary sizes. This is
   applied where citations live — the panels that name a source — and
   not globally, because breaking words mid-syllable is a real
   readability cost and should be paid only where it buys reflow.

   .improved-copy USED TO BE NAMED HERE and is gone with the panel it
   belonged to (9 Oct 2026). The measurement above is kept because it
   is the evidence for the rule, not a description of a live selector:
   the panel that produced the 296-against-200 overflow was
   .improved-copy, and .savings-part-note is the element that now
   carries a citation in a narrow column, so it inherits the rule. */
.savings-part-note,
.sourceline,
.linesource,
.breakdown .source { overflow-wrap: break-word; }

/* D4: provenance at the point of claim — directly beneath the number,
   same visual weight, NEVER in a footer. "Same visual weight" is read
   here as full body size and body colour, not the 0.875rem the rest of
   the fine print uses. */
.sourceline {
  margin: 0 0 1.5rem;
  max-width: var(--measure);
  color: var(--ink);
  font-size: 1rem;
}


/* ---- THE BREAKDOWN — the centrepiece ---------------------------
   Always expanded, never behind a toggle. There is deliberately no
   collapsed state defined here at all: no max-height, no .is-collapsed,
   nothing for a future edit to switch on. See the comment in
   quote_result.html and spec §7.7. */
/* ⚠️ --paper, AND IT WAS MISSING ENTIRELY UNTIL 8 Nov 2026.
   Guillaume, on the rendered result page: "background should be
   white".

   THE BOARD HAS ALWAYS SAID SO. ResultV2.dc.html draws this panel as
   `background: var(--paper); border: 1px solid var(--line)`. The page
   had the border and no background, so the panel sat on the page's
   cream and only the hairline separated it — the centrepiece of the
   whole result page (§7.7) reading as a faint outline rather than as
   a surface.

   A MISSING DECLARATION IS INVISIBLE TO EVERY CHECK THIS PROJECT HAS.
   The contrast walker measures colours that ARE declared; the board
   comparison matches STRINGS, and every word in this panel was
   present and correct. Only looking at it finds this, which is the
   case design-canvas/SKILL.md makes in its own words: "A bar chart
   still contains the word Premium."

   --paper, not a literal: it is #FFFFFF in light and the lifted navy
   in dark, so the panel stays a surface in both themes. */
.breakdown {
  margin-block: 2.5rem;
  padding: 1.25rem;
  background: var(--paper);
  border: 1px solid var(--line);
  border-radius: var(--radius-panel);
  max-width: 44rem;
}
.breakdown h2 {
  margin: 0 0 1rem;
  font-size: 1.25rem;
  letter-spacing: -0.01em;
}

.working {
  width: 100%;
  border-collapse: collapse;
}
.working thead th {
  text-align: left;
  font-size: 0.8125rem;
  text-transform: uppercase;
  letter-spacing: 0.04em;
  color: var(--ink-soft);
  font-weight: 600;
  padding-bottom: 0.5rem;
  border-bottom: 1px solid var(--line);
}
.working thead th:last-child { text-align: right; }

.working tbody th {
  text-align: left;
  font-weight: 500;
  padding: 0.875rem 1rem 0.875rem 0;
  border-bottom: 1px solid var(--line);
  vertical-align: top;
}
.working tbody td {
  padding: 0.875rem 0;
  border-bottom: 1px solid var(--line);
  vertical-align: top;
}

/* §5.2 rule 5: each line carries its own source. Always in the DOM and
   always readable — not a title attribute and not hover-only, both of
   which are invisible to a keyboard user and to a touch screen. */
.linesource {
  display: block;
  margin-top: 0.25rem;
  font-size: 0.8125rem;
  font-weight: 400;
  color: var(--ink-soft);
  max-width: 34rem;
}

.amount {
  text-align: right;
  font-variant-numeric: tabular-nums;
  font-weight: 600;
  white-space: nowrap;
}
/* The +/− sign carries the meaning; colour only reinforces it, so this
   does not rely on colour alone (WCAG 1.4.1). */
.amount.is-up   { color: var(--up); }
.amount.is-down { color: var(--down); }
.amount.is-base { color: var(--ink); }

.working tfoot th {
  text-align: left;
  padding-top: 1rem;
  font-size: 1.0625rem;
}
.working tfoot td {
  padding-top: 1rem;
  font-size: 1.0625rem;
  font-weight: 700;
}

.caveat {
  margin: 1rem 0 0;
  font-size: 0.875rem;
  color: var(--ink-soft);
  max-width: var(--measure);
}

/* THE TABLE STAYS A TABLE AT 390px, 6 Oct 2026.

   This block used to turn every row into stacked blocks below 30rem,
   hide the header row off-screen, and put the amount UNDER its label.
   The reason given was that "the label needs the full width" — and
   ResultPhone, which is drawn at exactly 390, keeps two columns with
   the label left and the amount right on the same line, under a
   WHAT YOU TOLD US / WHAT IT CHANGED header row.

   The board is right and the premise was wrong. The amounts are four
   to six characters (£1,813, +£222, −£441) and carry white-space:
   nowrap, so the amount column costs about 60px of 350 and the label
   keeps the rest — which it needs, but it has it. Stacking spent a
   whole extra line per row to save nothing.

   WHAT IS KEPT FROM THE OLD RULE: the per-row border, so each answer
   still reads as one unit, and the source line still sits under its
   label rather than beside it. What is dropped is the off-screen
   header, which was the real cost — the two column headings are the
   only thing on the page that says what the right-hand number IS, and
   a sighted phone user could not see them. */
@media (max-width: 30rem) {
  .working thead th { font-size: 0.6875rem; }
  .working tbody th, .working tfoot th { padding-right: 0.75rem; }
  /* NO `white-space: nowrap` ON THE AMOUNT. A first build added it,
     reasoning that "£1,813" should never break — and at 200% text
     that pinned the amount column at 251px inside a 360px viewport
     and scrolled the whole page sideways by 206px.

     The amounts do not need it. They are short enough to stay on one
     line on their own at ordinary sizes, and when the text is large
     enough that one must wrap, wrapping is the correct outcome and a
     sideways scroll is not. The nowrap that matters is the one on
     .amount's own rule — pre-existing, right at desktop, and the
     actual cause here: measured at 360px with 200% text it pinned the
     amount column at 251px of a 360px viewport and the table at 479px.

     So it is relaxed FOR THIS BREAKPOINT ONLY. At ordinary phone text
     the amounts still fit on one line without being told to (measured
     at 393px: amount column 126px, every figure on one line), so the
     board's two-column shape is unaffected; it is only when the text
     is doubled that a figure wraps, which is what should happen. */
  .working .amount { white-space: normal; }
  .working { table-layout: auto; }

  /* A LAST RESORT FOR VERY LARGE TEXT, and the only one the layout
     rules allow. Measured at 360px with 200% text after the two rules
     above: .breakdown's content box is 280px and the table's own
     minimum content width is 344px, so it burst out of its section by
     65px and scrolled the page.

     A table cannot shrink below its content, so the choice is to
     break the two-column shape the board draws or to let the TABLE
     scroll inside its own box. The layout rules allow exactly this
     exception — "only tables, diagrams and code blocks may be wider,
     each inside its own overflow-x: auto container" — and it keeps
     the board's shape for everyone whose text is not doubled.

     The scroller is the SECTION, not a new wrapper element: adding a
     div would be regenerating structure, which the layout contract
     forbids, and the section already boxes exactly this content. */
  .breakdown { overflow-x: auto; }
}

/* ---- The savings card — "After a year of good driving" -----------

   NEW 9 OCT 2026, from ResultPhone.dc.html. It replaces the .improved
   panel (a panel box reading "we'd aim to bring that to £X – £Y"
   beside the result-road artwork), whose rules were removed from this
   file in the same change along with its two 40rem media queries.

   THE BOARD'S DECLARATIONS, AND WHERE EACH VALUE COMES FROM.
   ResultPhone draws this card with hardcoded hexes; they are built as
   TOKENS here because every one of them already exists and because a
   hardcoded green does not survive dark mode, which no board draws.

     board            token            light      dark
     #E7F6EC fill     --ok-bg          #EAF7EE    #16301F
     #0F6B3F edge     --down           #0F6B3F    #7FD6A4
     #0F6B3F text     --down           same
     20px radius      --radius-panel   1.25rem
     20px padding     1.25rem
     14px gap         0.875rem

   #E7F6EC vs --ok-bg #EAF7EE is a 3-unit difference in two channels,
   invisible side by side, and --ok-bg is the token the rest of this
   file already uses for "a good-news surface". Inventing a sixteenth
   near-identical green would be the drift the token block exists to
   stop.

   MEASURED CONTRAST, both themes, because a green card is the easiest
   place on this page to fail AA without noticing:

     --down on --ok-bg        5.96:1 light / 8.16:1 dark  (needs 4.5)
     --ink on --ok-bg        14.75:1 light / 13.59:1 dark
     --ink-soft on --ok-bg    7.90:1 light /  7.79:1 dark
     --down border on --page  6.28:1 light / 10.61:1 dark (needs 3)

   THE 2px EDGE IS --down AND NOT --line. The board draws a 2px solid
   green border and it is doing real work: this is the only card on the
   page that is not white or the panel tone, and the edge is what stops the
   pale green reading as a disabled surface. --line at 1.25:1 would
   not read as a boundary at all. */
/* 2.5rem, NOT var(--section-gap), and the reason is margin collapse.
   .breakdown above it and .emailcard below it are 2.5rem and 2rem, and
   adjacent vertical margins collapse to the larger — so a 2rem card
   between two 2.5rem ones sits 40px below and 40px above regardless of
   its own value, while the 28px phone value would be silently
   discarded. Matching the siblings makes the declared value the
   rendered one, which is what makes the rhythm checkable.
   quote-layout R7 wants the gap on a parent; this page has no flex
   parent for its sections (it is a plain block .wrap), and introducing
   one would be regenerating the shell, which the contract forbids
   outright. So the cards carry matched margins and that is the
   pre-existing system here, not a new one. */
.savings {
  margin-block: 2.5rem;
  padding: 1.25rem;
  background: var(--ok-bg);
  border: 2px solid var(--down);
  border-radius: var(--radius-panel);
  max-width: 44rem;
  display: flex;
  flex-direction: column;
  gap: 0.875rem;
}

/* The board's 14px/700/uppercase/0.06em label. It is a <p> and not a
   heading: the heading of this section is the 40px line below it, and
   two headings in a row would give a screen reader a phantom level. */
.savings-kicker {
  margin: 0;
  font-size: 0.875rem;
  font-weight: 700;
  color: var(--down);
  text-transform: uppercase;
  letter-spacing: 0.06em;
  /* The board sets no measure on it; it is three words and must not
     wrap into the headline's space on a narrow screen. */
  max-width: none;
}

/* The board's 40px / 1.05 / 800 / -0.02em headline, FLAT 2.5rem.

   A clamp() was built here first — clamp(1.75rem, 1.35rem + 1.6vw,
   2.5rem) — on the reasoning that 40px is the phone board's size and
   the card also renders at 1280. Measured, that resolved to 28px at
   393, which is 12px under what the only board drawing this card
   specifies. The board leads: ResultPhone says 40px, so it is 40px,
   and the desktop size is the same because no desktop board draws
   this card at all yet (Result.dc.html still shows the panel this
   replaced — reported, not silently resolved).

   2.5rem and not 40px, so it scales with the user's text size.
   Checked for WCAG 1.4.10 at 360px/200% text, which is the condition
   that caught .bignumber's text-wrap: nowrap — see the numbers in
   the report. text-wrap: balance plus overflow-wrap below let it
   break between words, or inside one if a translation ever needs it,
   rather than pushing the page sideways.

   margin-top is 0.25rem and the rest of the gap comes from .savings'
   flex gap (R7: a component never carries its own margin), EXCEPT
   that the kicker and this line are ONE BLOCK on the board — so the
   4px between them is a margin here rather than a second flex
   container, which would be structure invented for 4px. */
.savings-headline {
  margin: 0.25rem 0 0;
  font-size: 2.5rem;
  line-height: 1.05;
  font-weight: 800;
  letter-spacing: -0.02em;
  color: var(--down);
  max-width: none;
  /* A long translated headline must break rather than push the page
     out — the 1.4.10 lesson from .bignumber, applied here before it
     can happen rather than after. */
  text-wrap: balance;
  overflow-wrap: break-word;
}

/* ---- The two bars -----------------------------------------------

   The board's structure: two stacked blocks, each a label/figure row
   over a 14px rounded track, 10px apart.

   THE FIGURES ARE THE DATA AND THE BARS ARE A RESTATEMENT. Both
   tracks are aria-hidden and carry no text, so a screen reader reads
   "Today, £1,550 – £1,900. After a year, £950 – £1,300." and is not
   read two empty boxes. That is also why the bars may be quantised
   (see savingsFillClass in engine.go) without any information being
   lost: the exact numbers are always present as text. */
.savings-bars {
  display: flex;
  flex-direction: column;
  gap: 0.625rem;
}
/* IT WRAPS TO TWO LINES RATHER THAN SCROLLING THE PAGE, and this was
   measured rather than anticipated.

   The board draws label and figure on one line, justified apart, and
   that is what this renders at every ordinary size. Built first with
   `flex-wrap` unset and `white-space: nowrap` on the figure — the
   obvious reading of the board — it put the document 28px wide at
   360px/200% text: "£1,550 – £1,900" is 205px at 32px text, the label
   takes its share of a 309px row, and nowrap refuses to give.

   That is WCAG 1.4.10, and it is the SAME failure .bignumber already
   carries a long note about: text-wrap: nowrap there scrolled the
   result page 141px before it was removed. Repeating it 400 lines
   lower would have been this project's recorded failure mode — a
   lesson written down in one place and not applied in the next.

   So the row is allowed to wrap. At every width and text size where
   the two fit on one line they are on one line, justified apart, as
   the board draws them; when they cannot, the figure drops below its
   label instead of pushing the page sideways. */
.savings-bar-head {
  display: flex;
  flex-wrap: wrap;
  justify-content: space-between;
  align-items: baseline;
  gap: 0.25rem 0.75rem;
  margin: 0 0 0.25rem;
  font-size: 0.875rem;
  max-width: none;
}
.savings-bar-label { color: var(--ink-soft); }
.savings-bar-figure {
  /* tabular-nums comes from the shared rule near the top of this file,
     which this selector is named in — so "£950 – £1,300" lines up
     under "£1,550 – £1,900" digit for digit, which is the whole
     reason the two bars read as a comparison.

     NO white-space: nowrap. A range is one idea and should not split
     across lines, but enforcing that with nowrap is what produced the
     28px of horizontal scroll described above. The non-breaking
     behaviour that matters — not breaking INSIDE a figure — is what
     overflow-wrap: break-word gives, and the row wrapping as a whole
     keeps "£1,550" and "£1,900" together in every case measured. */
  color: var(--ink);
  overflow-wrap: break-word;
}
.savings-bar-figure--after { color: var(--down); }

/* The track. 14px tall, fully rounded, as the board draws it.
   A fixed px height and not a rem: this is a hairline-class graphic
   rule like the keeper indent's 3px — at 200% text a 28px bar would
   be a slab, and the bar carries no text to scale with. */
.savings-track {
  height: 14px;
  border-radius: 99px;
  /* --down-track, 23 Oct 2026. This was
     color-mix(in srgb, var(--down) 12%, transparent) inline -- still
     a tint of the card's own green, still theme-tracking, but now
     named as the v2 flow boards name it. The opacity moved 12% -> 16%
     with that change, which is the boards' figure. */
  background: var(--down-track);
}

/* THE "TODAY" TRACK IS FULL WIDTH AND HAS NO FILL ELEMENT, because it
   IS the 100% the other bar is a proportion of.

   THE BOARDS DRAW IT IN --statement-bg AND THAT IS STILL NOT BUILT,
   deliberately. This is the one place the page departs from the
   board, and the v2 flow boards (23 Oct 2026) draw it the same way,
   so the departure carries forward rather than being resolved.

   RE-MEASURED AGAINST THE NEW TOKEN, because --statement-bg moved
   from the v1 lilac to the v2 boards' green-grey on 23 Oct and the
   old arithmetic no longer described anything:

       --statement-bg on --ok-bg   1.12:1 light / 1.27:1 dark   (now)
       --statement-bg on --ok-bg   1.19:1 light / 1.14:1 dark   (v1)
       --down-line    on --ok-bg   1.59:1 light / 2.02:1 dark   (built)

   THE CONCLUSION IS UNCHANGED AND THE CASE IS STRONGER IN LIGHT.
   A 1.12:1 bar is not a visible bar: the "Today" reference would all
   but disappear and the remaining green bar would read as a
   full-length one, which inverts the comparison the card exists to
   make. --down-line is the card's own green at 30% and is the
   clearest of the three in both themes, while staying plainly a
   neutral reference rather than a second data bar.

   Dark mode is on no board (see the btnraise note on the canvas,
   which records the same category of departure for the buttons), so
   this is reported rather than silently preferred. */
.savings-track--today {
  /* --down-line, 23 Oct 2026, was 28% inline and is the boards' 30%.
     This is the full-width "today" reference the other bar is a
     proportion of. Measured 1.59:1 light / 2.02:1 dark on --ok-bg,
     which is below 1.4.11's 3:1 and is correct here -- see the token
     block for why a redundant picture of a printed number does not
     bind. */
  background: var(--down-line);
}

/* The fill. Width comes from one of the classes below and NEVER from
   an inline style: the page is served under style-src 'self' with no
   'unsafe-inline', so Chrome drops a style attribute and the bar
   renders full width — measured on this project on 6 Oct 2026 for the
   phone step indicator, which shipped a progress bar filled to 100%
   at step 1. See the long note in engine.go and in _chrome.html. */
.savings-fill {
  height: 100%;
  border-radius: 99px;
  background: var(--down);
}

/* FIFTEEN DECLARED WIDTHS, 50% to 85% in 2.5% steps, named as the
   percentage x10. savingsFillClass() clamps to exactly this range, so
   it can never name a class that is not here — an undeclared class
   would be a fill with no width, which is an invisible bar and would
   pass every text check.

   Measured 9 Oct 2026 across 236 valid answer combinations: the real
   midpoint ratio runs 61.3% to 67.9%, so only 625/650/675 occur
   today. The other twelve are declared anyway, because the band moves
   if ImprovedFactor or DrivingDiscount* ever change and a missing
   class then fails silently.

   2.5% of a 350px phone bar is under 9px on a 14px-tall rounded bar,
   which is finer than the eye reads. The board's own figure is a flat
   65%, which is its example driver's proportion and wrong for most
   readers — see the note on the card in quote_result.html for why a
   bar whose length contradicts the figures beside it is the exact
   fault that passes text comparison, hashing and overflow checks. */
.savings-fill--500 { width: 50%; }
.savings-fill--525 { width: 52.5%; }
.savings-fill--550 { width: 55%; }
.savings-fill--575 { width: 57.5%; }
.savings-fill--600 { width: 60%; }
.savings-fill--625 { width: 62.5%; }
.savings-fill--650 { width: 65%; }
.savings-fill--675 { width: 67.5%; }
.savings-fill--700 { width: 70%; }
.savings-fill--725 { width: 72.5%; }
.savings-fill--750 { width: 75%; }
.savings-fill--775 { width: 77.5%; }
.savings-fill--800 { width: 80%; }
.savings-fill--825 { width: 82.5%; }
.savings-fill--850 { width: 85%; }

/* ---- The two components, above the divider's rule ---------------

   The board's 1px solid rgba(15,107,63,0.25) divider and two rows,
   each a bold term, a 12px note under it, and a figure on the right.

   A <dl> with a wrapping <div> per pair: that is the only valid way
   to group a dt/dd pair in HTML, and the grouping is what lets each
   row be its own flex container. */
.savings-parts {
  margin: 0;
  padding-top: 0.75rem;
  border-top: 1px solid color-mix(in srgb, var(--down) 25%, transparent);
  display: flex;
  flex-direction: column;
  gap: 0.5rem;
  font-size: 0.875rem;
  line-height: 1.45;
}
.savings-part {
  display: flex;
  justify-content: space-between;
  gap: 0.75rem;
}
.savings-part dt {
  font-weight: 700;
  /* min-width: 0 so the note's longest word cannot push the row wider
     than the card — the same flex rule the removed .improved-copy
     carried, and the reason overflow-wrap is on .savings-part-note. */
  min-width: 0;
}
.savings-part dd {
  margin: 0;
  font-weight: 700;
  color: var(--down);
  white-space: nowrap;
}
/* The source under "No claims for a year" and the aim under "Good
   driving". IN THE DOM, never a title attribute — §7.7's rule for the
   breakdown's sources, and it applies for the same reason: a title is
   invisible to a keyboard user and to a touchscreen.

   The two notes are the only thing distinguishing a published figure
   from a product decision with no source (see product_decisions.go),
   so they are not optional decoration. */
.savings-part-note {
  display: block;
  font-weight: 400;
  font-size: 0.75rem;
  color: var(--ink-soft);
}

/* The board's 12px footnote. "An aim, not a promise" is an FCA
   fair-clear-not-misleading hedge on a pre-launch money claim, half of
   which rests on an unconfirmed product decision — it is load-bearing
   copy, not a disclaimer to shrink away. Full line-height and the
   ordinary soft ink, not a 10px grey. */
.savings-foot {
  margin: 0;
  font-size: 0.75rem;
  color: var(--ink-soft);
  line-height: 1.5;
  max-width: none;
}

/* ONE EXTRA RULE, AND IT IS ONLY THE SIDE PADDING.

   A first build also bumped .savings-parts to 0.9375rem here, on the
   instinct that 14px source rows look small on a 1280 page. THE BOARD
   SAYS 14px AND THAT WAS INVENTED — R6 forbids a value that is on
   neither of R5's two scales, reached by nudging something until it
   looked right, and 15px is exactly that. Removed.

   The side padding is a different case: the board is a 390px artboard
   and its 20px gutter is the phone column of R5. At 1280 the card is
   704px wide and 20px of side padding against a 44rem measure reads as
   no padding at all, so it takes the desktop gutter. The VERTICAL
   padding stays at the board's 20px, because that is what sets the
   card's rhythm against its own contents.

   48rem is one of the contract's four breakpoints (R1); no fifth is
   introduced, and 40rem — which the removed .improved rules used — is
   gone from this page entirely. */
@media (min-width: 48rem) {
  .savings { padding: 1.25rem 1.5rem; }
}

/* ---- The omission panel — the differentiator, stated ----------- */
.omissions {
  margin-block: 2.5rem;
  padding: 1.25rem;
  background: var(--panel);
  border-radius: var(--radius-panel);
  max-width: 44rem;
}
.omissions h2 { margin: 0 0 1rem; font-size: 1.25rem; }

/* TWO COLUMNS AT EVERY WIDTH, including 390px. ResultPhone draws this
   panel as `grid-template-columns: 1fr 1fr` — six short noun phrases
   in two columns of three, which is the shape that makes the point
   ("look how little we asked") at a glance.

   It used to be one column below 30rem, on the ordinary assumption
   that a phone takes one column. The longest item here is "Where you
   park overnight"; measured at 390px the track is 155px, which wraps
   it to two lines and nothing overflows. A six-item single column was
   also simply longer than the panel it sits in needs to be. */
.omission-list {
  list-style: none;
  margin: 0 0 1rem;
  padding: 0;
  display: grid;
  gap: 0.5rem 1.5rem;
  /* TWO COLUMNS THAT CAN FALL BACK TO ONE. A flat `1fr 1fr` matched
     the board at 390px and then scrolled the page sideways by 206px
     at 200% text (measured: each <li> 210px wide in a 360px
     viewport). An equal-fraction track does not shrink below its
     content, so doubled text pushed both columns past the edge.

     auto-fit with a min() floor gives the board's two columns at
     every ordinary size and ONE column when the text is large enough
     that two cannot fit — the same pattern, and the same reasoning,
     as the radio-card grid in R8. */
  grid-template-columns: repeat(auto-fit, minmax(min(9rem, 100%), 1fr));
}
/* Icon and label on one baseline-aligned row. The icons are inline
   SVG (Lucide, ISC) and aria-hidden: each repeats the word beside
   it, so announcing them would double every item.

   The generated ✕ that used to sit here is gone — with an icon per
   row it read as two marks saying the same thing. The heading
   already establishes that these are what we did NOT ask. */
.omission-list li {
  display: flex;
  align-items: flex-start;
  gap: 0.625rem;
}
.omission-icon {
  flex: none;
  width: 1.375rem;
  height: 1.375rem;
  margin-top: 0.15rem;
  color: var(--ink-soft);
}

.omissions p { margin: 0.75rem 0 0; max-width: var(--measure); }
.principle { font-weight: 600; color: var(--ink); }

.nameddriver {
  margin-block: 2rem;
  max-width: var(--measure);
  color: var(--ink-soft);
  padding-left: 1rem;
  border-left: 3px solid var(--line);
}

/* ---- Two actions, EQUAL WEIGHT (§7.7 point 3) -------------------

   This is the rule most likely to be undone by someone optimising for
   conversion, so it is worth being explicit about what equal means
   here: same padding, same font size, same weight, same min-height,
   same flex-basis. The only difference is fill versus outline, which
   distinguishes them without ranking them.

   Do not give .action-decline a smaller font, a link appearance, or
   `margin-left: auto`. A forced or awkward choice produces noise, not
   signal — people join to escape the page, which puts unengaged
   addresses on the list and inflates the one metric we most need to
   trust. */
.actions {
  display: flex;
  flex-wrap: wrap;
  gap: 0.75rem;
  margin-block: 2rem 0.75rem;
  max-width: 44rem;
}

.action {
  flex: 1 1 14rem;
  display: inline-flex;
  align-items: center;
  justify-content: center;
  text-align: center;
  padding: var(--btn-pad-y) var(--btn-pad-x);
  min-height: 44px;
  font: inherit;
  font-weight: 600;
  border-radius: var(--radius);
  text-decoration: none;
  cursor: pointer;
  width: auto;
  margin-top: 0;
}

/* Filled orange versus outlined brand. Same box, same padding, same
   font-weight, same min-height, same flex-basis — the fill/outline
   difference distinguishes them without ranking them, which is what
   the handoff and §7.7 both ask for. */
.action-join {
  background: var(--cta);
  color: var(--on-cta);         /* 8.67:1 / 8.67:1 */
  /* --on-cta, not --cta and NOT --accent: see the note on `button`
     above, which this rule was changed with on 23 Oct 2026. The
     orange fill is 1.79:1 against the cream page, so the outline is
     what makes the control's boundary discernible (13.64:1). It also
     keeps the two actions the SAME box — both carry a 2px edge and
     only the fill differs.

     --accent WAS ORANGE IN DARK MODE, so this edge disappeared into
     its own fill there — the same fault as `button`, found on step 3
     a moment after the step 1 one was fixed, because the two orange
     controls are different elements with their own rules. --on-cta
     is navy in both themes by definition, which is what the v2 flow
     boards draw and what the brief requires. */
  border: 2px solid var(--on-cta);
}
/* .action-decline's FILL, INK AND EDGE moved to the shared secondary
   rule below, 5 Oct 2026 — it was `background: transparent`, which on
   the cream result page read as a hole rather than as a control, and
   it is now --paper like every other secondary. The contrast figures
   that were recorded here still hold: --ink on --paper is 16.27:1, and
   the 2px --accent edge is 7.54:1 light / 11.56:1 dark against a 3:1
   requirement. */

/* NO `filter: brightness()` ON .action:hover ANY MORE.

   It applied to BOTH actions and multiplied whatever
   background-color the hover state had just set — so the primary's
   specified #FF9E52 hover came out brighter than #FF9E52, and the
   secondary got a wash rather than the specified #F3ECFD. A filter
   also creates a containing block and repaints the slab, which made
   the 4px edge shimmer on hover. The two roles now each set their own
   hover background explicitly, below. */

/* ==================================================================
   THE RAISED BUTTON TREATMENT — ONE RULE, EVERY PAGE
   ==================================================================

   Guillaume's spec, 5 Oct 2026, and the fault it fixes: the same two
   roles were defined THREE TIMES and only one carried the relief.

   MEASURED BEFORE THE FIX, at 1280, with SIGNUP_ENABLED=true:

     homepage hero       .action .action-join   raised  (4px slab)
     homepage ready CTA  .action .action-join   raised
     homepage waitlist   bare <button>          FLAT, and orange-on-
                                                white because nothing
                                                gave it the secondary
     step 1 Continue     .btn .btn-primary      FLAT — .btn-primary is
                                                not defined in this
                                                file at all; it took
                                                the base `button` rule
     step 2 primary      .btn .btn-primary      FLAT
     step 2 Back         .btn .btn-secondary    FLAT, and a 1px border
                                                where every sibling
                                                has 2px
     step 3 join         .action .action-join   raised
     step 3 decline      .action .action-decline FLAT

   So a visitor pressed a raised button on the homepage, a flat one on
   step 1 and step 2, and a raised one again on step 3. That is the
   bug, and it is a bug about SHARING A DEFINITION rather than about
   any one value.

   .cta-btn AND .cta-sec ARE GUILLAUME'S NAMES, and they are also
   already on the boards: Home.dc.html and HomePhone.dc.html carry
   `class="cta-btn"` and `class="cta-sec"` with these exact
   declarations in their <helmet> CSS. Using them here means the board
   and the page now share a vocabulary, not just a look.

   THE ROLE CLASSES ARE ALIASED, NOT REPLACED. .action-join,
   .btn-primary, .action-decline and .btn-secondary all still exist
   and still mean what they meant — each is now a selector ON these
   two rules rather than a definition of its own. That matters for a
   specific reason: TestDeclineIsAsProminentAsJoin greps the rendered
   templates for the literal strings "action action-join" and
   "action action-decline", so renaming the step 3 classes would fail
   a §7.7 test. Aliasing keeps the markup the test pins and still
   leaves ONE definition per role.

   THE 4px BOTTOM EDGE IS THE WHOLE EFFECT. `0 4px 0` is not a drop
   shadow — it is a hard-edged slab under the button, a darker shade
   of the fill, which reads as thickness. The soft shadow behind it is
   the shadow proper and lifts the control off its ground.

   PRESSED MUST EAT THE EDGE. On :active the button moves down 3px and
   the slab drops to 1px, so the gap between button and ground almost
   closes: it reads as pressed rather than merely moved. Moving it
   down while keeping a 4px edge would look like the whole control
   slid, which is the usual mistake.

   THE SOFT SHADOW IS 0 10px 22px rgba(20,43,73,.18), WHICH IS A
   CHANGE. The old primary carried 0 12px 26px rgba(11,22,38,.35) —
   nearly twice the opacity of a shadow that now has to sit beside a
   white secondary button. Guillaume specified the lighter pair and
   the "Ready when you are" button ON THE BOARD already carries it.
   The hero button on that same board still carries the heavier old
   value, so the board is internally inconsistent; the spec and the
   newer of the two board values agree, and that is what is built.

   V2 RECOLOUR, 22 Oct 2026: the rgba() above was rgba(41,20,73,.18),
   an 18% tint of the old ink purple. It is now the same 18% tint of
   the navy ink. The GEOMETRY of every shadow in this section — the
   4px slab, the 10px/22px soft shadow, the 3px press — is untouched;
   v2 is colours only. */

/* THE EDGE COLOURS ARE TOKENS BECAUSE DARK MODE MOVES --accent.

   Guillaume's v1 spec named #C8682A and #6624C9 literally; v2 keeps
   the mechanism and changes the second hex. #C8682A is fine in both
   themes, because the orange fill does not change and neither should
   its edge. The ACCENT is not — in light mode --accent is navy
   #142B49, and in dark mode it becomes the light blue #A9DCFA (navy
   on the near-black dark page is 1.3:1 and fails the 3:1 a control
   boundary needs). A hardcoded navy slab under a light-on-dark
   button would be an invisible smudge, and the focus ring would be
   worse. So the two secondary values are tokens that follow the
   theme, exactly as they did under the purple palette. */
:root {
  --cta-edge: #C8682A;        /* darker shade of --cta */
  --cta-hover: #FF9E52;       /* --cta, one step warmer */
  --sec-fill: #FFFFFF;        /* = light --paper, the board's slab */
  --sec-edge: #142B49;        /* = light --accent, per the spec */
  --sec-hover: #DCEFFA;       /* = light --panel, per the spec */
  --cta-lift: 0 10px 22px rgba(20, 43, 73, 0.18);
  --sec-lift: 0 10px 22px rgba(20, 43, 73, 0.16);
}
/* THE DARK SECONDARY VALUES, TWICE — the media query is the no-JS
   fallback, the attribute selector is the explicit Dark choice. Same
   values in both; see the long note at the token block for why the
   duplication is deliberate and what keeps them in step. */
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) {
    --sec-edge: #A9DCFA;
    --sec-hover: #182B43;
    --cta-lift: 0 10px 22px rgba(0, 0, 0, 0.45);
    --sec-lift: 0 10px 22px rgba(0, 0, 0, 0.40);
    --sec-fill: #1E3552;
  }
}
/* --sec-edge follows --accent to the light blue; --sec-hover is the
   dark paper, so hover still reads as "this one" rather than as a
   hole. The lift goes to near-black: a navy-tinted shadow on a
   #101D30 page is invisible, and the relief has to come from the
   slab.

   --sec-fill EXISTS BECAUSE --paper DOES NOT LIFT ENOUGH IN DARK
   MODE. Guillaume, 13 Oct 2026, on his phone: the "Read our mission"
   button is hard to see. It was not a contrast failure and that is
   why nothing caught it — the 2px --accent border measured 11.56:1
   against the page, so every automated check passed.

   THE DEFECT IS HIERARCHY, NOT CONTRAST. In light mode the fill is
   #FFFFFF on a #FFF9F1 page: only 1.05:1 of lift, but white on warm
   cream still reads as a solid slab because the hue shifts. With
   --paper as the fill in dark mode the body of the button nearly
   vanishes into the page and only the outline and the 4px slab
   survive, so the eye reads the under-slab as the object and the
   button itself as a hole.

   RE-MEASURED FOR THE V2 PALETTE, 22 Oct 2026. The old table's
   numbers were all against purple surfaces and none of them survives
   the recolour, so it is recomputed rather than carried over:

     fill        lift/page  --ink  --sec-edge on fill
     #182B43       1.18     12.03        9.76   (= --sec-hover)
     #1E3552       1.36     10.45        8.48   <- this
     #243C5C       1.51      9.39        7.63
     #20406A       1.61      8.80        7.14

   1. LIFT. 1.36:1 against the page, the same order of separation the
      purple fill gave, and enough for the body to read as a slab.
   2. HOVER MUST STILL ANSWER. --sec-hover is #182B43, so taking
      --paper as the fill would make the rest state and the hover
      state the SAME COLOUR and hover would do nothing at all. That
      still rules out the obvious choice, exactly as it did before.
   3. THE SLAB MUST SURVIVE ON THE FILL. --sec-edge is the 4px relief
      directly under the fill; below 3:1 against it the thickness
      stops reading and the button goes flat again. THIS CONSTRAINT
      HAS STOPPED BINDING: the light blue #A9DCFA measures 8.48:1 on
      #1E3552 against the purple pair's 3.04:1, so every candidate in
      the table above clears it comfortably. The choice is now made
      on lift and on hover alone, which is worth recording — a future
      reader tightening this should know the slab is no longer what
      limits it.

   The orange --cta stays the only primary: it lifts 9.06:1 off the
   dark page against this button's 1.36:1, so the two do not compete,
   which is what the note above this block requires.

   The FOCUS RING is unaffected either way — it is --sec-edge at 3px
   with a 3px offset, so it is drawn on the PAGE, not on the fill, and
   measures 11.53:1 there. */
:root[data-theme="dark"] {
  --sec-edge: #A9DCFA;
  --sec-hover: #182B43;
  --cta-lift: 0 10px 22px rgba(0, 0, 0, 0.45);
  --sec-lift: 0 10px 22px rgba(0, 0, 0, 0.40);
  --sec-fill: #1E3552;
}

/* ---- PRIMARY (orange) -------------------------------------------- */
.cta-btn,
.action-join,
.btn-primary {
  box-shadow: 0 4px 0 var(--cta-edge), var(--cta-lift);
  transition: transform 120ms ease, box-shadow 120ms ease,
              background-color 120ms ease;
}
.cta-btn:hover,
.action-join:hover,
.btn-primary:hover {
  background: var(--cta-hover);
  transform: translateY(-2px);
}
.cta-btn:active,
.action-join:active,
.btn-primary:active {
  transform: translateY(3px);
  box-shadow: 0 1px 0 var(--cta-edge);
}
/* 3px, per the spec, and the COLOUR IS --ink SINCE V2.

   IT WAS A LITERAL CREAM #FFF9F1 and the comment here argued that
   "cream is the only thing in the palette that reads on the hero AND
   on the cream band where the other copies live". That was true of
   the purple design and is FALSE of v2 in both halves:

     cream on the light hero (pale sky #DCEFFA)   1.13:1
     cream on the cream band (--page #FFF9F1)     1.00:1

   The second figure was always 1.00 -- the ring was invisible on the
   band even under the purple palette, and only the hero copy was
   ever doing any work. The light hero now makes the hero copy
   invisible too, so the whole rule had to move.

   THE RING IS DRAWN ON THE GROUND, NOT ON THE BUTTON, because of the
   3px outline-offset -- so the colour it has to beat is whatever
   surrounds the control, never the orange fill. --ink is the page's
   own text colour and follows the theme:

     --ink on --page      13.64:1 light / 14.21:1 dark
     --ink on --hero-bg   12.08:1 light / 11.98:1 dark
     --ink on --sky        8.11:1 light / 13.69:1 dark

   Every surface a primary button appears on, both themes, worst case
   8.11:1. The 3px width stays: --accent at 2px is the shared ring
   elsewhere and 2px is too fine beside a 59px button. */
.cta-btn:focus-visible,
.action-join:focus-visible,
.btn-primary:focus-visible {
  outline: 3px solid var(--ink);
  outline-offset: 3px;
}

/* ---- SECONDARY (white, --accent edge) ----------------------------
   §7.7 IS WHY THIS EXISTS AT ALL, and it reverses what the comment
   on .action-join used to say here. That comment argued a raised
   decline button "would be two primary actions" and kept the relief
   off it deliberately. Guillaume's spec of 5 Oct overrules it: §7.7
   asks for decline to be EXACTLY as prominent as join — "same size,
   weight, padding, radius AND raised edge, differing only in fill" —
   and a flat button beside a raised one is not equally prominent, it
   is the lesser option with the same measurements. The two are now
   the same control in two fills, which is what the spec and §7.7
   both describe. */
.cta-sec,
.action-decline,
.btn-secondary {
  /* --sec-fill, NOT --paper. In light mode the two are the same
     #FFFFFF and nothing changes; in dark mode --paper gives this
     button only 1.14:1 of lift over the page and it stops reading as
     a solid control. See the --sec-fill note in the token block. */
  background: var(--sec-fill);
  color: var(--ink);
  /* 2px, matching every sibling. .btn-secondary had 1px, which is why
     step 2's Back read as a lighter control than the button next to
     it even before the relief existed. */
  border: 2px solid var(--accent);
  box-shadow: 0 4px 0 var(--sec-edge), var(--sec-lift);
  transition: transform 120ms ease, box-shadow 120ms ease,
              background-color 120ms ease;
}
.cta-sec:hover,
.action-decline:hover,
.btn-secondary:hover {
  background: var(--sec-hover);
  transform: translateY(-2px);
}
.cta-sec:active,
.action-decline:active,
.btn-secondary:active {
  transform: translateY(3px);
  box-shadow: 0 1px 0 var(--sec-edge);
}
.cta-sec:focus-visible,
.action-decline:focus-visible,
.btn-secondary:focus-visible {
  outline: 3px solid var(--sec-edge);
  outline-offset: 3px;
}

/* NO MOVEMENT UNDER REDUCED MOTION, both roles.

   The transform is motion; the relief is not. Reduced-motion keeps
   every colour and the slab, and drops only the travel — the control
   still answers, it just does not move. There is no gentler version
   of "it moves" for someone who asked for none. */
@media (prefers-reduced-motion: reduce) {
  .cta-btn, .cta-btn:hover, .cta-btn:active,
  .action-join, .action-join:hover, .action-join:active,
  .btn-primary, .btn-primary:hover, .btn-primary:active,
  .cta-sec, .cta-sec:hover, .cta-sec:active,
  .action-decline, .action-decline:hover, .action-decline:active,
  .btn-secondary, .btn-secondary:hover, .btn-secondary:active {
    transform: none;
    transition: none;
  }
}

/* THE SLAB IS 4px OF EXTRA PAINT BELOW THE BORDER BOX, and a
   full-width button on a phone sits in a 20px gutter, so it cannot
   overflow sideways — the slab has no horizontal spread. The soft
   shadow does spread 22px, which IS wider than the button; measured
   at 390 after this change, document scrollWidth stayed equal to the
   viewport on every page of the flow, because a box-shadow does not
   contribute to scroll width. The numbers are in the report. */

/* THE PHONE NUDGE. Two small lifts about 1.2s after load, on the HERO
   button only, below 61rem.

   WHY AT ALL: on a phone the hero fills the screen and the button is
   the one orange rectangle on it. The nudge is the page saying
   "this one" once, then stopping. It runs twice and ends — an
   animation that loops is a distraction, and this has no
   `infinite`.

   1.2s, not immediately: before that the page is still painting and
   the viewer is reading the headline. A control that moves while you
   are reading something else is noise.

   OFF, not reduced, under prefers-reduced-motion. There is no
   gentler version of "it moves" for someone who asked for no
   motion. */
@keyframes cta-nudge {
  0%, 100%   { transform: translateY(0); }
  8%, 24%    { transform: translateY(-4px); }
  16%, 32%   { transform: translateY(0); }
}
@media (max-width: 60.9375rem) {
  .cta-hero { animation: cta-nudge 1.6s ease-in-out 1.2s 1 both; }
}
@media (prefers-reduced-motion: reduce) {
  .cta-hero { animation: none; }
}
/* THE GENERIC .action FOCUS RING, for .action elements that are
   NOT one of the two raised roles. 5 Oct 2026: this rule used to
   apply to .action-join and .action-decline too, and because it sits
   AFTER the shared treatment at the same specificity it won — so
   the homepage "Ready when you are" button and both step 3 buttons
   took a 2px brand ring where the spec asks for 3px, measured at
   1280 as `outline 2px solid rgb(102,36,201) off:2px`. The two roles
   re-declare their own rings below, after this one, so source order
   now favours the spec. */
.action:not(.action-join):not(.action-decline):focus-visible {
  outline: 2px solid var(--accent);
  outline-offset: 2px;
}

/* THE RAISED ROLES' FOCUS RINGS, RE-DECLARED LAST.

   Specificity alone does not settle this: `.action-join:focus-visible`
   and `.action:focus-visible` are both one class plus one
   pseudo-class, so whichever is written last wins. They are written
   last here, deliberately, and must stay after any `.action`
   focus rule added later. Values are identical to the ones in the
   shared treatment above — a 3px --ink ring on the primary and a 3px
   --sec-edge ring on the secondary. Both were literals until v2 and
   the reasoning for the change is at the first copy, above. */
.cta-btn:focus-visible,
.action-join:focus-visible,
.btn-primary:focus-visible {
  outline: 3px solid var(--ink);
  outline-offset: 3px;
}
.cta-sec:focus-visible,
.action-decline:focus-visible,
.btn-secondary:focus-visible {
  outline: 3px solid var(--sec-edge);
  outline-offset: 3px;
}

.actions-support {
  margin: 0 0 1.5rem;
  font-size: 0.9375rem;
  color: var(--ink-soft);
  max-width: var(--measure);
}

.cta-row { margin-block: 2rem 1rem; }
.cta-row .action { flex: 0 1 auto; display: inline-flex; }

/* FULL WIDTH ON PHONE, per HomePhone.dc.html, which gives the CTA
   `display: block` inside the hero's 20px gutters.

   Measured against the board 3 Oct 2026: the live button was
   291.4px, sized to its text by the inline-flex rule above, where
   the board's block gives 335px at a 390px viewport. Nothing else
   competes for the row on a phone, and a 44px-plus target that
   spans the column is easier to hit than one that stops short.

   Desktop keeps the inline-flex: there the button sits in a column
   beside the card and a full-width CTA would be 640px of orange. */
@media (max-width: 60.9375rem) {
  .cta-row .action { display: flex; width: 100%; }
}

/* ---- Screen 5: the decline ------------------------------------- */
.decline { padding-block: 1.5rem 3rem; max-width: var(--measure); }
.decline h1 { font-size: clamp(1.75rem, 1.3rem + 2vw, 2.5rem); }
.decline .actions { margin-top: 1.5rem; }

/* A declined outcome is a SUCCESSFUL outcome and is never styled as a
   failure: no red, no warning iconography, no apology. */

/* Used for screen-reader-only text that must stay in the accessibility
   tree — a table caption, a live-region status. Not display:none,
   which would remove it from that tree entirely. */
.visually-hidden {
  position: absolute;
  width: 1px;
  height: 1px;
  margin: -1px;
  padding: 0;
  overflow: hidden;
  clip: rect(0 0 0 0);
  clip-path: inset(50%);
  white-space: nowrap;
  border: 0;
}

/* ---- Screen 2b: identity (§3.4) -------------------------------- */

/* C29b, the basis line. Body size rather than fine print: it carries
   the promise that keeps two lawful bases apart (§7.8.1), and a promise
   set in 13px is a promise you are hoping nobody reads. */
.basis {
  margin-block: 1.25rem;
  padding: 0.875rem 1rem;
  background: var(--panel);
  border-radius: var(--radius-panel);
  font-size: 0.9375rem;
  color: var(--ink);
  max-width: var(--measure);
}
.basis a { color: var(--accent); }

/* C30, the one-click delete on the decline screen. Its own form, so it
   is spaced away from the reason buttons rather than looking like a
   third option in the same group. */
.forget {
  margin-top: 2rem;
  padding-top: 1.5rem;
  border-top: 1px solid var(--line);
  max-width: 44rem;
}
.forget-note {
  margin: 0 0 0.75rem;
  font-size: 0.9375rem;
  color: var(--ink-soft);
}
/* Not full width and not flex:1 — it is a single button rather than one
   of a pair, so it sizes to its content like the landing CTA does. */
.forget .action { flex: 0 1 auto; display: inline-flex; width: auto; }

/* F7: the Privacy Notice link, directly under the consent wording and
   before the submit button.

   Consent has to be informed, and a notice reachable only after you
   have submitted is not informing anyone. It cannot go inside the
   <label> — the settled O2 wording is stored verbatim as the consent
   record and is escaped, so it carries no markup, and a link inside a
   label is a focus-order trap (code review A5).

   Indented to line up with the label text rather than the checkbox, so
   it reads as part of the same statement. The grid column above is
   `auto 1fr` with a 0.625rem gap and a 1.1rem box. */
/* Continues the consent panel rather than floating under it: the
   label's last sentence is "See our Privacy Notice" and this is the
   link that lets you, so the two must read as one block. Indented to
   line up with the label text, past the 1.5rem box and 0.75rem gap. */
.consent-link {
  margin: 0 0 1.25rem;
  padding: 0 1.125rem 1rem calc(1.5rem + 0.75rem + 1.125rem);
  background: var(--panel);
  border-radius: 0 0 var(--radius) var(--radius);
  max-width: var(--measure);
  font-size: 0.9375rem;
}
.consent-link a { color: var(--accent); }

/* ===================================================================
   HOMEPAGE — the approved design board, rebuilt 22 Sep 2026
   ===================================================================

   Built from homepage-supplement/specs/layout.md, whose geometry table
   is followed rather than approximated. Where a number below looks
   arbitrary it came from that table; where it differs, a comment says
   why.

   THE TOKENS. The supplement ships its own tokens.css. Those values
   are adopted here under the --hero-, --banner- and --headline-accent
   names rather than by overwriting the existing --brand/--cta palette,
   which the advisory-quote flow's six other screens depend on.

   ================================================================
   THE HERO IS NO LONGER A DARK SURFACE IN BOTH THEMES. V2.
   ================================================================

   22 Oct 2026. This block used to pin --hero-bg, --hero-text and
   --headline-accent to dark values in BOTH themes, with the comment
   "they are already dark surfaces and inverting them would invert
   the board". That was exactly right of the purple design, where the
   hero was a deep purple panel at every hour of the day.

   V2 CHANGES THE PREMISE: HomeV2.dc.html draws the light hero as
   PALE SKY (--hero-bg #DCEFFA) with NAVY TEXT (--hero-text #142B49),
   and only HomeV2Dark.dc.html draws it dark. A hero pinned dark in
   both themes would now be wrong in the light one — which is the
   theme most people will see — so the pins are removed and the hero
   takes the palette's own --hero-* tokens, which are defined per
   theme at the top of this file.

   THE BANNER AND THE FOOTER KEEP THEIR PINS, and the old reasoning
   holds for them unchanged: both boards draw them as dark bands in
   BOTH themes, so inverting them would invert the board.

   Every pair below was RECOMPUTED for v2 against the surface it
   actually renders on, both themes. The hero rows now differ by
   theme because the hero does:

     --hero-text on --hero-bg     12.08:1 light / 11.98:1 dark
     --hero-text on --sky          8.11:1 light / 13.69:1 dark
     --hero-soft on --hero-bg      8.20:1 light /  9.99:1 dark
     --hero-soft on --sky          5.51:1 light / 11.42:1 dark
     --headline-accent on hero     7.54:1 light /  9.69:1 dark
     --headline-accent on --sky    5.06:1 light / 11.08:1 dark
     --banner-text on --banner-bg 17.35:1  (both themes)
     --footer-text on --footer-bg 13.64:1  (both themes)
     --tick-mark on --tick-bg      5.92:1  (both themes)

   THE --sky ROWS ARE THE WORST CASE AND THAT IS WHY THEY ARE HERE.
   The hero's ground is a gradient from --hero-bg into --sky (see
   .homehero), so a contrast figure against --hero-bg alone describes
   only the top of the panel. Every hero text colour is measured
   against BOTH ends of that ramp and the lower figure is the one
   that has to clear 4.5:1. All of them do.

   DESKTOP AND MOBILE ARE BOTH DESIGNED. The hero is two columns above
   69rem and one below; the steps are three columns above 60rem. Both
   breakpoints come from where the content stops fitting, not from a
   device width.

   EVERY IMAGE DECLARES width/height IN THE HTML, so the box is
   reserved before the bytes arrive and nothing moves when they do. */

:root {
  /* --hero-bg, --hero-text and --hero-soft are NOT set here any more.
     They come from the palette and follow the theme; see the note
     above for why the pin was removed.

     THE BANNER AND FOOTER DO KEEP THEIR PINS: both boards draw them
     as dark bands in both themes. */
  --banner-bg:        #0B1626;
  --banner-text:      #FFF9F1;
  --footer-bg:        #142B49;
  --footer-text:      #FFF9F1;

  /* THE HEADLINE ACCENT IS NOW PER-THEME, AND IT HAD TO BE.
     In light mode the hero is pale sky, and the peach #FFCB9A that
     this carried measures 1.35:1 on #DCEFFA -- invisible, not merely
     weak. The light value is the deep burnt orange the v2 boards use
     for accent text on pale ground; it measures 7.54:1 on --hero-bg
     and 5.06:1 at the bottom of the sky ramp, so it passes as text
     at both ends. Dark mode keeps the peach, which is what the dark
     boards draw and which measures 9.69:1 / 11.08:1 there. */
  --headline-accent:  #7A3606;

  /* Deliberately darker than the board's green. White on the board's
     #30B881 is 2.53:1 and fails even the 3:1 graphical minimum; this
     is 5.92:1, which passes as text. */
  --tick-bg:          #18724B;
  --tick-mark:        #FFFFFF;

  /* RENAMED IN SPIRIT, KEPT IN NAME. There is no purple any more, so
     "on-purple" describes nothing; the token is the focus ring drawn
     on the DARK BANDS (banner, footer, the contextual section), which
     are still dark in both themes. It keeps its name only because
     handler_test.go references it by that name and renaming it is a
     change to the test suite that is not this commit's business --
     see the note beside its one consumer. */
  --focus-on-purple:  #FFF9F1;
}
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) { --headline-accent: #FFCB9A; }
}
:root[data-theme="dark"] { --headline-accent: #FFCB9A; }

/* ---- The mandatory bar ------------------------------------------

   ACROSS THE TOP OF THE PAGE, above the nav, as the board shows.

   NOT sticky, NOT dismissible, NO fixed height and NO overflow
   clipping. At 360px the text wraps to roughly four lines and the bar
   grows to fit — measured at 116px there. Anything that caps this
   height breaks C1/D5, which requires it fully visible at 360px
   without scrolling. */
.topbanner {
  background: var(--banner-bg);
  color: var(--banner-text);
}
.topbanner-text {
  max-width: var(--site-max, 75rem);
  margin-inline: auto;
  /* QUIETENED 12 OCT 2026, GUILLAUME — AND NOTHING ABOUT THE
     OBLIGATION CHANGED.
     ----------------------------------------------------------------
     He asked for this sentence to be removed. It is the C1/D5 notice:
     the reason an unauthorised firm may show a price at all, and
     config.go REFUSES TO START without it when a flow is live. So the
     words are untouched and present on every page that carried them;
     only the volume came down. He chose "keep it, restyle it quieter"
     over removing it once that was explained.

     WHAT CHANGED: 16px -> 14px, and the text to 88% opacity. The bar
     stops being the loudest thing above the fold while still being
     read before the heading it qualifies.

     WHAT DID NOT, because each of these would break the thing the
     restyle was allowed to preserve:
       - the WORDS. Not shortened, not reworded. A reworded regulatory
         disclosure is a new disclosure and needs counsel, not CSS.
       - min-height 44px as a FLOOR, never a ceiling, and no overflow
         clipping. At 360px the text wraps to about four lines and the
         bar must grow to fit: C1/D5 requires it fully visible at
         360px without scrolling.
       - contrast. --banner-text on --banner-bg is 16.18:1; at 88%
         opacity over #1F0D53 it is ~12.9:1, far above the 4.5:1 AA
         minimum and still above AAA's 7:1. Measured, not assumed.
       - its position: first in document order, above the nav.

     DO NOT make it smaller still, and do not add `display: none` at
     any breakpoint. The next person to be told "remove the banner"
     should read this block and the guard at
     internal/platform/config.go:607 before touching anything. */
  min-height: 44px;
  padding: 10px clamp(1.25rem, 4vw, 2.5rem);
  margin-block: 0;
  font-size: 0.875rem;  /* 14px, was 16px */
  line-height: 1.5;     /* 21px */
  opacity: 0.88;
  text-align: center;
  text-wrap: pretty;
}

/* ---- Nav --------------------------------------------------------- */

.sitenav {
  background: var(--hero-bg);
  color: var(--hero-text);
}
.sitenav-inner {
  max-width: var(--site-max, 75rem);
  margin-inline: auto;
  padding-inline: clamp(1.25rem, 4vw, 2.5rem);
  min-height: 72px;
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  /* 32px from the wordmark to the links at desktop, 20px at 360px —
     both values are layout.md's, and the switch is at the width where
     the wordmark and both links stop sharing a line. */
  gap: 0.75rem 2rem;
  padding-block: 0.875rem;
}
@media (max-width: 30rem) {
  .sitenav-inner { gap: 0.75rem 1.25rem; }
}
.sitenav .brand {
  display: inline-flex;
  align-items: center;
  gap: 0.5rem;
  color: var(--hero-text);
  text-decoration: none;
  font-size: 1.375rem;
  font-weight: 700;
  letter-spacing: -0.02em;
}
/* ⚠️ THE FLOW HEADER NO LONGER USES THIS RULE, 23 Oct 2026.

   WHAT THIS PARAGRAPH USED TO SAY, because it was right until it was
   not: page_nav carried the bare square on nine pages, seven of them
   the quote flow, because Guillaume's instruction of 11 Oct 2026 was
   to leave the quote pages on the OLD LOGO while the homepage and
   the mission page took the ring -- "v2 is colours only, and the
   shape is not a colour".

   THAT HELD WHILE THE FLOW BOARDS DREW A SQUARE. The quote-flow v2
   boards draw Logo B, and the brief for that change says so in as
   many words, so page_nav now renders .brand-c + .brand-o and takes
   the shared ring block. The 11 Oct instruction is superseded, not
   forgotten.

   THE RULE STAYS because .portal-brandmark is not its only other
   user and a bare .brand-mark may still be rendered somewhere; see
   the note there on why the portal's square is correct and must not
   be converted. If a grep shows nothing renders .brand-mark any
   more, this rule can go -- but delete it on evidence, not on this
   comment.

   IT NO LONGER READS --headline-accent, AND IT HAD TO STOP. That
   token became per-theme in v2 and its light value is a deep burnt
   orange chosen to be readable as TEXT on pale sky; a 12px square
   drawn in it would be a dark-brown dot. --logo is the right token:
   it is the brand's orange, #E8762A light and #FFB264 dark, and it
   exists precisely for marks like this one.

   A LOGOTYPE, SO 1.4.11 DOES NOT BIND. #E8762A on --page measures
   2.84:1. WCAG 1.4.11 exempts "text that is part of a logo or brand
   name" and, by the same reasoning, the brand mark beside it; the
   wordmark's accessible name comes from the link's aria-label, not
   from this square being discriminable at a ratio. It is the most
   legible orange in the brand family, which is why it is the one
   chosen. */
.brand-mark {
  width: 0.75rem;
  height: 0.75rem;
  border-radius: 3px;
  background: var(--logo);
  flex: none;
}

/* ---- THE HOMEPAGE RING MARK — canvas v180, 11 Oct 2026 ----------

   An open orange ring, used as the letter C of the wordmark. The text
   beside it is .BrandRest — the name without its initial — so the two
   together spell the name once. See the note in index.html.

   SCOPED TO .sitenav ON PURPOSE. The rule above styles the bare
   .brand-mark, which the homepage header AND _chrome.html's page_nav
   both carry, and page_nav serves nine pages — seven of them the
   quote flow. Main.dc.html, MainPhone.dc.html and Result.dc.html all
   still draw the orange ROUNDED SQUARE plus the full name, and
   Guillaume's instruction of 11 Oct 2026 was to leave the quote pages
   on the old logo. Restyling the bare selector would have relit all
   nine in a place nobody was looking — the same trap the
   --headline-accent note further down this file records.

   THE BOARD'S OWN VALUES, verbatim from Home.dc.html:140 (desktop)
   and HomePhone.dc.html (phone):

     17px / 16px box, border-box
     3px solid #FFB264, border-right-color: transparent
     border-radius: 50%
     margin-right: -1px, and gap: 0 on the flex parent

   UPRIGHT. No rotate. The ring on /our-mission is rotated -25deg and
   drawn as a 25x28 ELLIPSE, which disagrees with its own board
   (Mission.dc.html draws 17px upright with a 3px border); that is
   reported, and mission.css is not this rule's business.

   A SQUARE BOX, SO A CIRCLE. width == height with box-sizing:
   border-box, so border-radius: 50% gives a true circle. An unequal
   box would give an ellipse, which is what went wrong on the mission
   page.

   var(--logo), NOT A LITERAL, SINCE V2 -- and that is a change of
   mechanism, not only of value. It was a literal #FFB264 because the
   v1 boards drew it in exactly that hex in a hero that was dark in
   BOTH themes, so one value served both. The v2 light hero is PALE
   SKY, and #FFB264 on #DCEFFA is 1.51:1: the ring would simply
   vanish. --logo is the per-theme brand orange the v2 palette
   defines for this job -- #E8762A light, #FFB264 dark -- and the
   dark value is byte-identical to the literal this replaces, so
   dark mode is unchanged.

   A LOGOTYPE, SO 1.4.11 DOES NOT BIND. #E8762A on the light hero
   measures 2.52:1 and #FFB264 on the dark hero 8.02:1. WCAG 1.4.11
   exempts logos and brand names from the contrast requirement, and
   this ring IS the letter C of the name. The light value is
   nonetheless the most legible orange in the brand family -- every
   lighter one was computed first -- because "exempt" is a reason not
   to fail the page, not a reason to draw something nobody can see.

   gap: 0 OVERRIDES the 0.5rem on .sitenav .brand above: the ring is
   the first letter of a word, so it sits tight to the rest of it, not
   a half-rem away. The -1px pulls out the optical gap the ring's own
   open right side leaves. */
/* BASELINE, NOT CENTRE — see the note on .sitenav .brand-mark.
   align-items:center (set on .brand above) centres the ring on the
   text box including its descender space, which floats a mark that
   has no descender. The v2 spec is explicit: "align-items: baseline".
   Set here rather than edited into the rule above so the reason
   travels with the change. */
.sitenav .brand { gap: 0; align-items: baseline; }
/* .brandrest is the name minus its initial. It REPLACED a
   <span class="wordmark">, which set font-weight: 600 and
   letter-spacing: -0.01em and so quietly overrode .brand's own 700 —
   the board draws this wordmark at 800, so the weight is set here
   explicitly rather than left to whichever span happens to win.
   Everything else inherits from .brand. */
/* THE WORDMARK IS THE SAME ON BOTH PAGES, and the numbers come from
   the boards, which agree with each other: Home.dc.html and
   Mission.dc.html both draw it `gap: 0; font-weight: 800;
   font-size: 19px` with the ring pulled -1px into the text.

   Both pages had drifted, in opposite directions — this one to 22px
   system-ui, /our-mission to 28px Manrope — so neither was at the
   board's size. Setting the family and size on .brand (the link)
   rather than on .brandrest means the ring scales with the text if
   the size ever changes. */
.sitenav .brand {
  font-family: Manrope, system-ui, sans-serif;
  font-size: 19px;
  font-weight: 800;
  letter-spacing: -0.02em;
  gap: 0;
}
.sitenav .brandrest {
  font-weight: 800;
  letter-spacing: -0.02em;
}

/* ===================================================================
   THE "o" IS ORANGE, EVERYWHERE THE WORDMARK APPEARS. 23 Oct 2026.
   ===================================================================

   Guillaume: "make sure everywhere CoRoute is mentioned except in the
   footer it is orange", after spotting it on the "CoRoute AI" heading
   and the mission chat card.

   LOGO B IS THE RING **AND** THE "o". The v2 note on the canvas says
   so — 'ring C + "o" orange (#E8762A light, #FFB264 dark), "Route" in
   text colour' — and the reviewed preview implements exactly that:

       .logo .c { border-color: var(--logo) }   the ring, empty span
       .logo .o { color: var(--logo) }          the letter o
                                                "Route" inherits

   The port drew the ring from var(--logo) but rendered "oRoute" as
   one run in the text colour, so the "o" came out white. The token
   was always right — its comment at :root already reads 'ring C and
   the "o"' — only the second half was never wired up.

   ONE SELECTOR, NOT SEVEN. The wordmark appears seven times across
   three templates (two headers, two portal/end panels, the CoRoute AI
   heading, the chat card, the context end panel). They are deliberately
   NOT given a shared class, because each sits in a different component
   with its own size and weight; what they share is this one letter.
   Styling .brand-o alone keeps that shared thing in one place, and
   var(--logo) means each context's own override (the dark sidebar and
   the end panel both set --logo: #FFB264) still applies.

   THE FOOTER IS EXCLUDED BY NOT HAVING THE SPAN. It renders
   {{ .CompanyName }} — "CoRoute Ltd" — as flat text, because that is a
   Companies Act disclosure naming a legal entity, not a logo. There is
   no .brand-o in it and there must not be: colouring a letter inside a
   registered company name would dress a legal statement as branding.
   Both homepage boards draw the footer as a plain bold wordmark with
   no ring, which is the same decision. */
/* ⚠️ ONE DECLARATION FOR EVERY PART OF EVERY WORDMARK. 23 Oct 2026.
   Guillaume: "the C and o have different ones ... in some places",
   then "can we have one variable for this so we only need to make
   the change at a single place".

   THEY HAD DIVERGED INTO FOUR ORANGES. Measured across 18 wordmarks,
   the C and the "o" disagreed in 7 of them:

       home header          C #E8762A   o #E8762A   ok
       CoRoute AI heading   C #FFAA66   o #E8762A   MISMATCH
       home end panel x2    C #FFB264   o #E8762A   MISMATCH
       mission chat header  C #FFAA66   o #E8762A   MISMATCH
       mission chat small   C #9A4A00   o #E8762A   MISMATCH

   The "o" rule below was right from the start -- it reads --logo.
   Every mismatch was a C with its own hardcoded hex, written before
   the token existed. So this is one list, not one rule per site: the
   ring's border, the typed C and the "o" all take --logo, and
   changing the logotype colour is now editing --logo and nothing
   else.

   THE TYPED Cs ARE GONE AS WELL, same day, see .brand-c below. Six
   wordmarks drew a literal "C" glyph where the header draws the
   ring. They were listed here first so they matched in COLOUR; the
   SHAPE was the second half and it is done. Nothing in the codebase
   types a C for the logotype any more, and nothing should start:
   grep for the ring class, not for the letter.

   ⚠️ CONTRAST IS WHY SOME CONTEXTS OVERRIDE --logo, AND THOSE
   OVERRIDES MUST SURVIVE. The ring is a BORDER, which WCAG 1.4.11
   treats as a non-text contrast case, but it sits inside a readable
   word and the "o" beside it is unambiguously text -- so the whole
   mark is held to text contrast here. #E8762A on a dark card
   measures 2.98:1; the mission chat card found this and moved to
   --accent-text (#9A4A00 light, 6.26:1). The fix for a contrast
   failure is to override --logo FOR THAT CONTEXT (as .p-side, .end
   and .ai-head already do in the preview), never to hardcode a hex
   on the C and leave the "o" behind -- which is exactly how the
   four oranges happened. */
.brand-o { color: var(--logo); }

/* ⚠️ THE RING IS THE C. ONE CLASS, EVERY WORDMARK. 23 Oct 2026.
   Guillaume screenshotted the "CoRoute AI" heading: the C was small,
   low, and plainly a different letterform from the header's ring.

   WHAT IT WAS. Six wordmarks typed a literal "C" glyph in whatever
   face the surrounding text used -- system-ui in the homepage
   heading and end panel, Manrope and DM Sans on /our-mission -- so
   the logotype was drawn three different ways on two pages. Only the
   header and the mission footer had the ring. They were given a
   shared colour first (above); this gives them the shared SHAPE.

   WHY A SHARED CLASS AND NOT SIX COPIES. The geometry below was
   wrong once already (17px inside 19px text, a fifth too big, read
   as an O) and was corrected at ONE site while five others kept the
   old letterform. A single class is what makes "fix the ring" one
   edit instead of six, which is the same argument as --logo above.

   EVERY NUMBER IS em, so the ring scales with its context without a
   rule per size: 48px heading, 19px chat header, ~15px label,
   ~10px drive summary, 14px end panel. The values are the v2 spec's
   -- 0.74 x 0.75em, 0.165em stroke, sized to the CAPITAL height so
   it sits level with the R beside it rather than standing taller.

   display:inline-block, NOT inline-flex, because five of the six
   sit inside inline text (.ai-label is display:inline) and a flex
   box there would not share the line box.

   THE GEOMETRY LIVES WITH .sitenav .brand-mark, below, which carries
   the derivation of every number and the forced-colors rule that
   keeps the arc open. .brand-c is a second selector on that one
   block, not a copy of it -- a copy is how the ring came to be
   corrected at one site and wrong at five. */
/* ⚠️ em, NOT px, AND THE RING IS SIZED TO THE CAPITAL HEIGHT.
   Corrected 23 Oct 2026. Guillaume, looking at the header: "the C
   seems a bit off".

   It was 17px square inside 19px text — 0.895em — against the v2
   spec's 0.74em x 0.75em. A fifth too big in both directions, so the
   ring stood taller than the capital R beside it and read as a round
   O rather than as a C sized to the letters. Measured before:

       width   0.895em   spec 0.74em    +21%
       height  0.895em   spec 0.75em    +19%
       border  0.158em   spec 0.165em
       align   center    spec baseline
       top     auto      spec .012em

   ALIGNMENT WAS HALF THE FAULT. align-items:center centres the ring
   on the text BOX, which includes the descender space, so a mark with
   no descender floats high. The letters sit on the baseline, so the
   ring must too, with .012em of optical nudge — a circle looks high
   when its geometric centre is on the line.

   WHY em AND NOT px. These are the canvas note's own numbers and they
   are all relative, so the ring tracks the wordmark at every size: the
   nav's 19px, the phone board's smaller nav, and any future change to
   .sitenav .brand's font-size. The px version needed a second rule at
   30rem to re-state the size; that rule is gone because 0.74em is
   already correct at any font-size. This matches v2-src/css/d2.css
   `.logo .c`, which is the reference implementation. */
/* .brand-c is the SAME MARK inside running text -- the homepage
   "CoRoute AI" heading, the end panel, the three mission labels. It
   shares this block rather than restating it, and adds only
   display:inline-block, because the nav's ring is a flex item while
   these five sit in a line of text. */
.sitenav .brand-mark,
.brand-c {
  box-sizing: border-box;
  width: 0.74em;
  height: 0.75em;
  border: 0.165em solid var(--logo);
  border-right-color: transparent;
  border-radius: 50%;
  background: none;
  position: relative;
  top: 0.012em;
  margin-right: -0.03em;
  flex: none;
}
/* The nav's ring is a flex item; the five in running text are not,
   and an empty inline span has no height. inline-block is the only
   difference between the two uses. */
.brand-c { display: inline-block; }

/* ⚠️ .brand-c-text IS THE LETTER "C" AS REAL TEXT, HIDDEN. 30 Oct 2026.
   ===================================================================

   THE RING IS A BORDER, SO IT IS NOT A CHARACTER, SO A CRAWLER READS
   THE WORDMARK AS "oRoute". Measured on the live site by stripping
   tags from the served HTML and concatenating the text the way an
   indexer does -- 4 bare "oRoute" on /, 4 on /our-mission, 1 on
   /get-in-touch. The worst was an H2: "oRoute AI it understands".

   WHY aria-label DID NOT ALREADY FIX IT. Every one of these wordmarks
   carries aria-label="{{ .BrandName }}" and that is correct and
   staying -- it is what makes a screen reader say "CoRoute". But
   aria-label is an ACCESSIBILITY affordance, and Google does not use
   aria-label on a NON-INTERACTIVE element as indexable text. So the
   page was announced correctly and indexed wrongly, which is why no
   accessibility check could see this.

   WHY THE "C" IS NOT PUT INSIDE .brand-c. It would render on top of
   the arc -- the ring is 0.74em of border with nothing in it, and a
   glyph in there sits over the stroke. That is the fault Guillaume
   screenshotted on 23 Oct 2026 in a different form, and
   TestNoWordmarkTypesTheLetterC still forbids a non-empty ring span.
   The "C" is a SIBLING placed immediately before the ring, so the
   text stream reads C + o + Route and the painted mark is untouched.

   WHY NOT .visually-hidden, WHICH ALREADY EXISTS, AND WHY THIS USES
   THE SAME MECHANISM ANYWAY. That class is the right technique and
   this one is deliberately a near-copy of it, differing in two
   declarations that matter at THIS call site:

     - it is INLINE-BLOCK, not the default inline. .visually-hidden is
       applied to block-level captions and status text; here the span
       sits mid-word between a border-arc and an "o", and inline-block
       is what guarantees the 1px box never participates in line
       layout around it.
     - it adds user-select: none, so a reader who selects and copies
       the wordmark gets "CoRoute" once rather than "CCoRoute". A
       table caption has no such neighbour and does not need it.

   position: absolute IS WHAT KEEPS THE PAINTED MARK EXACTLY WHERE IT
   WAS. Out of flow means zero contribution to the line box, so the
   ring and the "o" cannot shift by a pixel -- which is the one thing
   this change was not allowed to do. The text is still in the
   DOCUMENT order a crawler reads; "out of flow" is a LAYOUT property,
   not a text-stream one. Verified by measuring the wordmark's
   bounding box before and after, not by looking at it.

   display:none AND visibility:hidden ARE BOTH WRONG HERE and neither
   is a near miss: both remove the text from what a crawler extracts,
   which is the entire purpose lost. Clipping is the only family of
   technique that hides pixels while keeping text.

   THE DECLARATIONS:

     position/overflow/clip-path   hides the pixels without hiding the
                                   text
     width/height 1px              the smallest box that still counts
                                   as rendered text
     user-select: none             copying the wordmark gives
                                   "CoRoute", not "CCoRoute"

   IT IS aria-hidden AT EVERY CALL SITE. The enclosing element already
   carries aria-label with the whole name, so an announced "C" here
   would make a screen reader say "C CoRoute". The hidden text is for
   INDEXERS; the aria-label is for assistive technology. Both are
   needed and they do not overlap -- the same split static.go makes
   about MIME types and CORS. */
.brand-c-text {
  position: absolute;
  display: inline-block;
  width: 1px;
  height: 1px;
  overflow: hidden;
  clip-path: inset(50%);
  white-space: nowrap;
  user-select: none;
}

/* ⚠️ THE STROKE IS FLOORED TO WHOLE PIXELS BY THE BROWSER, and at
   the two smallest sizes that is visible. Measured 23 Oct 2026 after
   the ring conversion, in Chrome at dpr 1:

       font    0.165em wants   browser paints   effective
       10px        1.65px            1px          0.100em
       14px        2.31px            2px          0.143em
       15px        2.48px            2px          0.133em
       19px        3.14px            3px          0.158em
       48px        7.92px            7px          0.146em

   So the ring gets PROPORTIONALLY LIGHTER as the text shrinks, while
   the "o" beside it keeps its weight -- at 10px the arc is a hairline
   next to a solid letter and the two stop reading as one typeface.
   Seen in the magnified captures, not inferred: the 19px and 48px
   marks look right, the 15px is marginal and the 10px is wrong.

   max() DOES NOT HELP and neither does outline -- both were tried and
   measured. The flooring happens when the used value is resolved, so
   max(2px, 0.165em) still paints 1px at 10px... it paints 2px, but
   only because the 2px floor is itself a whole pixel. The general
   point stands: a fractional stroke cannot be painted, so the only
   lever is the font-size the ring is sized against.

   THE FIX IS PER-CONTEXT AND DELIBERATELY SMALL. Rather than scale
   the whole mark, the two undersized contexts take a slightly larger
   em basis so the floor lands on the next pixel up. This is a
   rendering correction, not a design change -- the spec's 0.165em is
   still what the ring is drawn to wherever the pixel grid allows. */
.mission .ai-drive-summary .brand-c { border-width: 0.2em; }
/* FORCED COLORS, AND THIS RULE IS NOT OPTIONAL — FOUND BY LOOKING.
   ===================================================================

   An earlier version of this block carried a comment claiming forced
   colors "keeps transparent transparent, so the ring survives as an
   open ring". IT DOES NOT. Chrome repaints EVERY border side to the
   system foreground, the transparent one included, so the ring closed
   into a solid circle and the wordmark read "OoRoute" — the ring
   stopped being a C and started being an O.

   Caught in home--desktop-1280--forced-colors.png from
   measure-layout.js --shots. No text check could have seen it: the
   markup still said BrandRest, the contrast was fine, there was no
   overflow. It was only visible in the picture, which is the whole
   argument for looking.

   `revert` is what restores transparency here: forced-colors-adjust
   is a blunt instrument and `transparent` is a keyword the UA
   stylesheet is allowed to override, so the side is reset explicitly
   inside the media query.

   THE RING IS DECORATION AND THE NAME IS STILL ANNOUNCED. Even if a
   future Chrome closes the ring again, the <a> carries
   aria-label="{{ .BrandName }}", so a screen reader says "CoRoute"
   regardless of what the border does. This rule is for people who
   SEE the page in a forced palette, not for assistive technology. */
@media (forced-colors: active) {
  .sitenav .brand-mark,
  .brand-c {
    forced-color-adjust: none;
    border-color: CanvasText;
    border-right-color: transparent;
    background: none;
  }
}

/* ---- ONE NAV LINK — canvas v180, 11 Oct 2026 ---------------------

   Guillaume, on the desktop board: "NAV, 11 Oct (Guillaume): one quiet
   link to the mission page. Keep the header this light pre-launch."

   WHAT USED TO BE HERE, because it is why this is nine lines and not
   a hundred and thirty. The nav was removed entirely on 4 Oct
   ("remove the nav from both") and about 120 lines went with it:
   .navlinks, .navtoggle, the hamburger bars, and the
   [data-nav-ready] rules that only hid the links once flow.js had
   made the toggle a real control. NONE OF THAT COMES BACK. One link
   needs no toggle, no JS and no progressive-enhancement gate; it is a
   flex item beside the wordmark at every width, including 360px.

   THE UNDERLINE IS A 2px border-bottom IN THE BOARD'S ORANGE, not
   text-decoration. The boards draw a rule sitting 2px clear of the
   baseline with a 2px gap (padding-bottom), which text-decoration
   cannot place; and it is drawn in rgba(255,178,100,0.9), the warm
   orange, while the text stays cream — a two-colour treatment that
   one text-decoration declaration cannot express either. The exact
   rgba is the boards', both of them.

   CONTRAST IS ON THE TEXT, NOT THE RULE. --hero-text on --hero-bg is
   12.08:1 light / 11.98:1 dark, well past 4.5:1, so the link passes on
   its own colour and the orange rule is decoration on top. That
   matters because the rule is the only thing distinguishing this from
   the surrounding copy, and a link identified by colour alone would
   fail 1.4.1 — here the underline is the non-colour cue and it is
   always drawn, not only on hover.

   The focus ring comes from the .sitenav a:focus-visible rule below,
   which already covered this element before it existed.

   The #approach and #how sections keep their ids; nothing links to
   them. See docs/tasks/remove-nav.md. */
.sitenav-links {
  display: flex;
  align-items: center;
  /* margin-left:auto, not justify-content on the parent: the parent
     wraps, and space-between would push the wordmark and the links to
     opposite edges of whichever line each landed on. */
  margin-left: auto;
  /* 28px, which is Contact.dc.html's nav gap. TWO LINKS SINCE 13 OCT
     2026 — "Our mission" and "Contact" — so this gap now separates
     something rather than only trailing the wordmark. */
  gap: 0.25rem 1.75rem;
  /* WRAP, AND BE ALLOWED TO SHRINK SO THE WRAP CAN HAPPEN. Both
     declarations are needed and min-width is the one that is easy to
     miss: a flex item defaults to min-width:auto, so this box refuses
     to become narrower than "Our mission Contact" on one line, and
     with margin-left:auto holding it right that minimum is a floor
     under the whole header. min-width:0 lets it narrow; flex-wrap
     then moves "Contact" onto its own line.

     MEASURED at 200% browser text on a 360px device (the widest the
     nav ever has to be relative to its space): the links box is 280px
     inside a 360px viewport, the two links sit on separate lines at
     y+3397 and y+3479, and horizontal overflow is 0. At 100% text it
     stays on ONE line at every width from 320 to 1440 — nav block
     161px, 108px clear after the wordmark at 390px, 38px at 320px.

     NOT THE CAUSE OF THE 248px OVERFLOW SEEN DURING THIS CHANGE, and
     the distinction matters because the wrong lesson is easy to draw
     here. That run reported
     "home--galaxy-s24--text200.png: page scrolls sideways by 248px",
     and the cause was a premature Go-template comment terminator in
     index.html's header
     comment, which dumped 37 lines of commentary into the page as
     36px text. These two declarations are still right — without them
     a two-link nav has no way to reflow at 200% text — but they fixed
     nothing on their own, and the overflow only cleared when the
     comment did. TestHomepageCommentsDoNotLeakIntoThePage now guards
     that, because go vet, gofmt and the whole Go suite were green on
     the broken page. */
  flex-wrap: wrap;
  min-width: 0;
}
.sitenav-links a {
  color: var(--hero-text);
  text-decoration: none;
  font-size: 1rem;
  font-weight: 600;
  border-bottom: 2px solid rgba(255, 178, 100, 0.9);
  padding-bottom: 2px;
}
/* THE CURRENT PAGE, for when this nav is reused on a page it links to.
   NOTHING ON THE HOMEPAGE MATCHES THIS: it links to /our-mission and
   /get-in-touch and is neither, so both of its links are the ordinary
   state and this rule is inert here. It exists so that the next page
   to reuse .sitenav-links has the convention waiting rather than
   inventing a second one.

   WEIGHT, NOT THE UNDERLINE, and not opacity either. Contact.dc.html
   marks its current item with font-weight 700 and dims the others to
   opacity .9. Only the WEIGHT half is taken:

     * the orange rule under every link is a WCAG 1.4.1 affordance,
       not decoration — every nav link in this header is the same
       colour as its siblings, so a link without the rule is
       identified by colour alone. It therefore cannot double as the
       current-page mark.
     * the .9 is LEFT OFF deliberately. It is safe on contrast
       (--hero-text on --hero-bg is 12.08:1 light and 11.98:1 dark,
       and dimming it to .9 leaves it far above 4.5:1 at both ends of
       the hero's gradient), but Home.dc.html and
       HomePhone.dc.html draw BOTH of their nav links at full
       strength. Dimming them here would make the page disagree with
       the board it was just reconciled against, to no benefit on a
       page where no link is current. If a page that IS one of these
       destinations ever uses this nav, add the .9 there and redraw
       its board to match.

   aria-current carries the meaning; the weight carries the look. */
.sitenav-links a[aria-current="page"] { font-weight: 700; }
/* Hover brightens the RULE to full opacity rather than moving or
   recolouring the text: the link is already underlined at rest, so
   the available answer is to strengthen what is there. No transform —
   a header link that jumps on hover is noise. */
.sitenav-links a:hover { border-bottom-color: rgb(255, 178, 100); }
/* PHONE: 15px, the phone board's size, and the 2px rule tightens to
   1px of padding as the board draws it. The links stay OUTSIDE any
   menu — HomePhone.dc.html is explicit that the mission link "stays
   visible next to the menu button", and there is no menu button on
   this page at all, so there is nowhere to hide either of them.

   THE 1rem GAP HOLDS WITH TWO LINKS. ContactPhone.dc.html draws 14px
   between its three nav items; 16px here across two is the same
   intent and was already shipped, so it is not re-tuned for one more
   label. Measured on the running page at 390px: the nav block is
   209..370 and 161px wide, ONE line, with 108px of clear space after
   the wordmark; still one line and 38px clear at 320px. */
@media (max-width: 30rem) {
  .sitenav-links { gap: 1rem; }
  .sitenav-links a { font-size: 0.9375rem; padding-bottom: 1px; }
}
/* ONE RULE BECAME TWO IN V2, AND THE SPLIT IS THE POINT.

   These three regions used to share a single cream focus ring,
   because all three were dark panels in both themes. Two of them no
   longer are: .sitenav and .homehero take the hero's ground, which
   in light mode is PALE SKY, and cream on #DCEFFA is 1.13:1 -- the
   ring would be invisible on the two regions a visitor tabs through
   first.

   .sitefooter IS STILL DARK IN BOTH THEMES (--footer-bg is pinned;
   both boards draw it that way), so it keeps the cream ring and
   keeps --focus-on-purple, which is what that token is now for.

   Measured:
     --hero-text on --hero-bg   12.08:1 light / 11.98:1 dark
     --hero-text on --sky        8.11:1 light / 13.69:1 dark
     --focus-on-purple on --footer-bg        13.64:1 both themes */
.sitenav a:focus-visible,
.homehero a:focus-visible {
  outline: 3px solid var(--hero-text);
  outline-offset: 3px;
  border-radius: 2px;
}
.sitefooter a:focus-visible {
  outline: 3px solid var(--focus-on-purple);
  outline-offset: 3px;
  border-radius: 2px;
}

/* ---- The hero panel ---------------------------------------------

   FULL-BLEED. No external margin, radius 0, min-height 560px below the
   banner on desktop, height free to grow. The background is the SOLID
   a CSS colour and all live text sits on it. */
/* THE HERO GROUND RAMPS INTO THE ARTWORK'S OWN SKY, V2, 22 Oct 2026.

   Up to v1 this was a flat `background: var(--hero-bg)` and that was
   right, because the artwork had a transparent sky and the hero
   colour WAS the sky — "the hero's ground is the sky" is how the
   comment on .hero-art-layer puts it, and it was accurate.

   THE V2 ART PAINTS ITS OWN SKY and is fully opaque (0.0%
   transparent, measured on the shipped files), so the flat ground
   and the art's sky are two different blues meeting at a line. The
   ramp dissolves that line: the hero holds --hero-bg across the top
   where the copy sits, then slides into --sky, which is sampled from
   the art itself --- #84CBFC day against the file's rgb(115,197,254),
   #021E4A night against its rgb(1,31,82).

   THE STOPS ARE THE CANVAS PREVIEW'S (v2-src/css/d2.css): hold to
   22%, ramp to 56%, flat below. The art's top fade (0->36% of the
   art box, see .hero-art-layer) finishes inside the ramp, so the two
   gradients overlap rather than abut.

   THE COPY STAYS ON THE FLAT PART. At the 600px desktop hero the
   22% hold is 132px and the ramp's midpoint is at 234px; the
   headline block ends above that and the form sits on sky that has
   only just begun to turn. Contrast was re-measured on the ramp's
   DARKEST point for the light theme, which is --sky itself: --ink on
   #84CBFC is 8.11:1, and on the dark theme --ink #E5ECF3 on #021E4A
   is 13.69:1. Both are the worst case, not the average. */
.homehero {
  position: relative;
  background: var(--hero-bg);
  background: linear-gradient(180deg,
              var(--hero-bg) 0,
              var(--hero-bg) 22%,
              var(--sky) 56%,
              var(--sky) 100%);
  color: var(--hero-text);
  border-radius: 0;
  overflow: hidden;   /* clips the artwork layer, not the content */
}
@media (min-width: 69rem) {
  /* 600px. Canvas v152, 10 Oct 2026 — the board's grid is
     `height: 600px` and Guillaume's spec names the figure: "Desktop
     hero grid height 600px".

     IT WENT 560 -> 660 -> 600 IN ONE DAY, and the middle value is why
     this comment exists rather than only the number. 660 was canvas
     v148's, raised "~130px taller so nothing is cut off" when the
     hero first gained the inline email bar. v153 brings it back to
     600 with the bar still in the hero, so the two numbers are not
     describing the same content — what came OUT between them is the
     "Today's estimate…" small print (moved to quote step 1) and the
     "Drive well, it pays" lettering (now only in the footer).

     min-height, NOT height, AND THAT IS LOAD-BEARING HERE. The board
     is a fixed 1280 canvas and sets `height: 600px`; the page is
     responsive and must not clip. 600 is the floor the composition
     needs, and the hero grows past it whenever the content is taller
     — which it IS in the state production is actually in: with
     SIGNUP_ENABLED off the bar renders its taller "Not open yet"
     box, and with the full stored CONSENT_TEXT the consent line wraps
     to several lines. Measured in both states rather than assumed;
     the numbers are in this commit's report. A literal `height` would
     have cut the consent line off in exactly the case where it
     matters most. */
  .homehero { min-height: 600px; }
}

/* THE HERO ART — v6, 8 Oct 2026 (Guillaume), hero-wide-v6.webp.

   WHAT THE BOARD DRAWS (Home.dc.html, live canvas 1791493496-f90c,
   read this session with the Artifact tool):

     position: absolute; right: -110px; bottom: -14px;
     width: 980px; height: auto; max-width: none;
     mask-image: linear-gradient(to right, transparent 0, #000 14%);

   THE SKY IS TRANSPARENT, AND THAT IS THE WHOLE IDEA. v6 is RGBA with
   a real alpha channel and the sky cut out, so THE HERO GROUND IS THE
   SKY. The rooftops, the trees and the skyline form a silhouette
   against it instead of a rectangle sitting on it. Verified on the
   shipped file rather than taken from the handover:

     1942x809, RGBA, 151 KB
     fully transparent         31.3% of sampled pixels
     alpha at top-centre       0
     alpha at top-left         0
     centre column opaque from y=359, i.e. 44% down

   ⚠️ DO NOT FLATTEN OR RE-ENCODE THIS FILE. Every earlier hero image
   here was flattened onto #4B1FAE to save the bytes the alpha cost,
   and this comment used to explain that trade. It is reversed: the
   alpha IS the design now, and flattening would paint the sky back in
   as a solid rectangle and undo v6 entirely. The supplier's note says
   the same — keep the alpha when serving or processing.

   SO THERE IS NO TOP MASK. Earlier versions carried a second
   `to bottom` ramp and a mask-composite to hide a painted edge whose
   purple sat up to 14/255 off the token. There is no top edge to
   hide: where the art would have had one, it is transparent. The
   board draws no top mask and neither does this.

   THERE ARE TWO HORIZONTAL MASKS, AND THEY DO DIFFERENT JOBS.
   Amended 9 Oct 2026, when the second was added.

   LEFT, 14%, FOR READABILITY. `transparent 0, #000 14%` fades the
   scene out from under the copy so the headline, the lede and the
   small print sit on flat hero ground. 14% of 980px is 137px. It is the
   board's value.

   RIGHT, 9%, FOR THE EDGE — and this one IS an edge mask, which is
   why the paragraph above no longer says there is none. The art's
   right edge is opaque scene content, so once the width was fixed at
   980px and the art stopped running off the window, that edge became
   a visible vertical cut. The ramp dissolves it into the hero's ground.
   The measurement and the reason for 9% rather than 14% are in the
   rule below, beside the declaration.

   WHY THE NUMBERS CHANGED FROM v5, which used 1100px / -90px / -30px
   / 16%: v5 was an opaque rectangle and its geometry was fighting the
   copy — the car sat under the lede and the small print at every
   width below about 1900, measured at 1.22:1 and 1.27:1 against a
   4.5:1 requirement, and the BOARD failed the same way. Guillaume
   resolved it in the ARTWORK rather than in CSS: v6 is composed so
   the street starts right of the text column. The smaller 980px width
   and the larger -110px overhang are part of that composition, not
   tuning, so they are used as given.

   HIS ACCEPTANCE CRITERION, and it is specific: "At 1280px the street
   should start just right of the text column, with the skyline under
   the video card." Both halves are measured in
   docs/ux/hero-art-v6-measurements.md and in the report for this
   change, because neither is something a script here asserts.

   IT IS A BACKGROUND, NOT AN <img>, AND THAT IS THE ONE DEPARTURE
   from the board's mechanism. A board has no media queries, so it can
   only draw an element. Declaring the image inside this query is what
   stops a phone fetching it: measured previously, a <picture> in the
   markup was requested at a 390px viewport even though the layer is
   display:none there, and media="" on the <source> did not help
   because the <img src> stays a valid fallback. A background-image in
   an unmatched media query is never fetched. The geometry below is
   the board's exactly; only the delivery differs.

   Hidden below 61rem: at phone widths there is no room beside the
   copy. The phone gets its own file and its own rule — see
   .hero-phone-art. (This bound was 69rem until 9 Oct 2026; it moved
   with the hero grid, so the art and the two columns appear and
   disappear together rather than 128px apart.)

   Not mirrored. The car travels on the left of the centreline, which
   is correct for a British road. */
.hero-art-layer { display: none; }

@media (min-width: 61rem) {
  .hero-art-layer {
    display: block;
    position: absolute;
    /* ⚠️ THE ART IS A FIXED COMPOSITION ANCHORED TO THE CONTENT
       COLUMN. FIXED WIDTH, FIXED OFFSET, EXACTLY LIKE THE VIDEO CARD.
       Rewritten 9 Oct 2026 (second pass). This supersedes BOTH earlier
       rules — the 8 Oct "fill to the viewport edge" and the 9 Oct
       half-rate growth — and the reason is a requirement, not a
       tuning preference.

       GUILLAUME, with a screenshot of a very wide, short window: "it
       still fucks away and becomes large", then the decisive sentence:
       "THE CAR SHOULD NOT MOVE RELATIVE TO THE TEXT AND PLAYING
       VIDEO." That rules out every width-as-a-function-of-100vw rule,
       including the half-rate one, because any such rule moves the
       art's left edge at a different rate from the copy column's.
       Slower drift is still drift; the requirement is none.

       THE VIDEO CARD WAS ALREADY RIGHT AND IS THE MODEL. It lives in
       .hero-inner, a centred column capped at --site-max, so its gap
       from the headline is a CONSTANT 140px at every width. Measured
       on the previous build, the fault stated as one table:

         vw     h1 l..r      video l..r     art l..r      h1->art  h1->video
         1104    40..580     624..1064     124..1104       -456      140
         1280    80..620     760..1200     300..1280       -320      140
         1600   240..780     920..1360     575..1600       -205      140
         1920   400..940    1080..1520     815..1920       -125      140
         2400   640..1180   1320..1760    1175..2400         -5      140
         2560   720..1260   1400..1840    1295..2560        +35      140
         3440  1160..1700   1840..2280    1955..3440       +255      140
         3840  1360..1900   2040..2480    2255..3840       +355      140

       811px of drift in the art, zero in the card. Same hero, same
       page, two anchoring models — and the card's is the one the
       board draws.

       THE BOARD'S OWN NUMBERS, READ AS A COMPOSITION. On Home.dc.html
       the art is `right: -110px; bottom: -14px; width: 980px` inside a
       `position: relative` box of `width: 1197px` — the CONTENT
       column. So in column-relative terms:

         art left  = column left + 327px    (1197 + 110 - 980)
         art width = 980px                   a literal, not a floor

       Both constants. That is the entire specification, and it is what
       is built below: the art's box is pinned to the centred column at
       a fixed offset and a fixed size, so headline, video card and car
       translate together as ONE composition when the window resizes.
       Nothing inside the hero moves relative to anything else.

       `right` is expressed from the viewport because .homehero spans
       it while .hero-inner is centred:

         column right = (100vw + min(100vw, --site-max)) / 2
         art right    = column right + 110px
         so `right`   = 100vw - art right
                      = (100vw - min(100vw, --site-max)) / 2 - 110px

       which is the page's own gutter minus the board's overhang. The
       width is now the literal 980px beside it.

       WHAT THIS COSTS, STATED RATHER THAN HIDDEN: above ~1420px of
       viewport the art no longer reaches the window's right edge, and
       bare hero ground appears to its right — 90px at 1600, 250px at 1920,
       570px at 2560, 1210px at 3840. Guillaume rejected an earlier
       version for leaving "a big blank empty area", so this is a
       REVERSAL of that constraint, made knowingly: he has now said the
       art moving relative to the text is the thing he will not have,
       and the two cannot both be satisfied by one rectangle. The
       empty-area objection is answered by the mask below rather than
       by growth.

       AND THE GROUND IS BALANCED, NOT POOLED. Because the whole
       composition is centred on --site-max, the space left over is
       split evenly either side — the same gutter the copy already
       has. The previous rule put ALL of it on the left (353px at 1920,
       1113px at 3440 of empty purple before the headline) while the
       art ran to the right edge, which is exactly the lopsided hero in
       Guillaume's screenshot. A fixed composition fixes that as a
       side effect of being fixed.

       Re-check this if --site-max, the 440px preview card or the 48px
       grid gap changes; the composition is relative to all three. */
    /* THE ART'S OWN BOX. The height follows the source's 1942x809
       ratio, so the scene is never stretched and never upscaled.

       880px, NOT THE BOARD'S 980px — A DELIBERATE DEPARTURE.
       ----------------------------------------------------------------
       Changed 12 Oct 2026. Guillaume, with a 1913px screenshot: "text
       seems to be overlapping with the graphics of the asset.. I don't
       think its like that in the board". It was not: on Home.dc.html
       the lede is capped at `max-width: 30ch` (343px rendered, four
       lines) and stops 24px clear of the art box, while the page's
       lede ran 470px and its last words sat on the blossom tree.

       TWO WAYS TO FIX THAT, AND HE CHOSE THE ART. Narrowing the text
       to the board's measure was the board-faithful fix and it was
       built first; he rejected it: "you could leave the text as is
       but make the graphics smaller", then the acceptance criterion,
       "it should barely overlap with the bars". So the copy keeps a
       comfortable measure and the picture yields instead.

       "THE BARS" ARE THE EMAIL FORM AND THE CTA, both 400px and both
       ending at copy left + 400. "Barely overlap" is the target, not
       clearance — the art should graze their right edge, not retreat
       from it. Measured at 1913 before the change: bars end at 789,
       art box began at 672, so the art crossed them by 117px.

       WHY 880 AND NOT SOME OTHER NUMBER. The art is anchored by its
       RIGHT edge, so width is the only lever on the left edge and the
       right edge does not move. 980 - 880 = 100px of left edge
       recovered, which puts the box at 772 against bars ending at
       789: a 17px graze. That is "barely overlap" read literally.
       Verified at 1280, 1440, 1680 and 1913 — the right edge stayed
       at its old value at every one and only the left edge moved.

       A FIXED px, NEVER A vw. Any width that is a function of the
       viewport moves the art's left edge at a different rate from the
       copy column's, which is the drift Guillaume rejected on 9 Oct:
       "THE CAR SHOULD NOT MOVE RELATIVE TO THE TEXT AND PLAYING
       VIDEO." The long note above is that requirement; this number
       changes the composition's size once, not its behaviour.

       SCALE: 880px of a 1942px native file is 0.45x, still a
       downscale at every width, so nothing is upscaled or softened.

       THE BOARD IS NOW BEHIND THE PAGE on this one value and that is
       recorded as a note on the canvas rather than left silent —
       board leads, so a page that departs from it says why, where
       the next reader of the board will see it. */
    width: 880px;
    right: calc((100vw - min(100vw, var(--site-max, 75rem))) / 2 - 110px);
    left: auto;
    /* HEIGHT IS STATED, NOT DERIVED.
       16 Oct 2026. 880 x 809/1942 = 366.59, which is what Chrome
       derives from width + aspect-ratio anyway, so this changes
       nothing visible in either engine.

       IT IS HERE FOR CERTAINTY, NOT AS A FIX. It was written while
       chasing a gap between the art and the copy that turned out to
       have a different cause entirely — see the note on
       .hero-art-layer's width below, which records what the fault
       actually was. `height: auto` + `aspect-ratio` on a box pinned
       by `bottom` with `top: auto` is a corner the engines have not
       always agreed on, so stating the number removes one thing that
       could ever differ. aspect-ratio stays as the statement of
       intent and the thing to recompute from if the width moves.

       RECOMPUTE THIS IF THE WIDTH CHANGES. The 90rem block below
       overrides both together for exactly that reason. */
    height: 366.59px;
    aspect-ratio: 1942 / 809;
    bottom: -14px;
    top: auto;
    pointer-events: none;

    /* ---- THE DAY/NIGHT PAIR, V2, 22 Oct 2026 ------------------
       The art is a BACKGROUND, not an <img>, so the theme swap is
       two more declarations rather than a <picture>. A background
       inside an unmatched media query is never fetched, and a
       background whose rule is overridden is never fetched either,
       so ONLY THE ACTIVE THEME'S FILE IS DOWNLOADED — verified by
       scripts/check-theme.js, which watches the network.

       THE FILES ARE THE SAME SIZE, 1942x809, both themes, so the
       swap cannot move anything. Verified byte by byte against the
       canvas assets: the day file is blob 93c69f6..., the night
       file 6af557b..., and their sha256s match the installed
       copies exactly.

       THE DEFAULT (no data-theme, no JS) IS THE DAY ART, and the
       prefers-color-scheme pair below is the no-JS fallback that
       makes a dark OS get the night art anyway. */
    background-image: url("/static/img/day/hero-home-wide.webp");
    background-repeat: no-repeat;
    background-position: center;
    background-size: 100% 100%;
    /* TWO MASKED EDGES, AND THE SECOND ONE IS NEW.

       LEFT, 14% — the board's value, and it is for READABILITY, not
       for an edge: it fades the scene out from under the copy so the
       headline, the lede and the small print sit on flat hero ground. 14%
       of 980px is 137px.

       RIGHT, 9% — added 9 Oct 2026 with the fixed width, because a
       fixed width is what first brings the art's right edge INTO the
       window. Measured on hero-wide-v6.webp, the rightmost column is
       ~93% opaque (727 of 809 rows fully opaque: the postbox, the
       hedge, the kerb). So without a fade there is a hard vertical
       cut — sampled across the seam on the 8 Oct build at 2560, y=600:

         x=1974  rgb(218,207,187)   the pavement
         x=1975  rgb( 75, 31,174)   the hero token, dead flat

       With the 9% ramp the last column composites to a max per-channel
       delta of 0 from #4B1FAE — the scene ends IN the purple rather
       than against it, so wide viewports read as a composition on a
       purple field instead of an image that ran out.

       WHY 9% AND NOT THE LEFT'S 14%. The subject matter decides it.
       Measured on the source: the red postbox's body occupies columns
       1736..1772, i.e. 89.4%..91.2% of the width, and the woman and
       lamppost sit just inside it. Only 8.8% of the art lies beyond
       the postbox. A 14% fade starts at 86% and dissolves the postbox;
       a wider, softer ramp (22%, tried and rendered) washes out the
       postbox AND the tree. 9% starts at 91% — the first column past
       the postbox — so it consumes only hedge, wall and pavement.
       Both were composited over #4B1FAE and LOOKED AT before choosing;
       the images are in the report for this change.

       mask-composite is still not needed: one gradient carries both
       stops, because they are on the same axis. */
    /* MASK GEOMETRY IS STATED EXPLICITLY.
       16 Oct 2026. Only mask-image was set; the spec defaults are
       `mask-size: auto` and `mask-repeat: repeat`, which Chrome
       happens to render as a single box-sized copy but which are not
       what this rule means. Pinning one copy across the box makes
       "14%" mean 14% of the art's width in any engine.

       LIKE THE HEIGHT ABOVE, THIS WAS NOT THE FIX IT WAS WRITTEN AS.
       It went in while chasing the gap recorded on the width rule
       below, on a guess about Firefox that proved wrong. It is kept
       because the declaration was genuinely incomplete, not because
       it moved anything. */
    /* ---- A SECOND MASK AXIS, V2, 22 Oct 2026 -------------------
       A TOP FADE IS NEW AND IT IS REQUIRED BY THE NEW ART, not a
       preference. Every hero image up to v6 had a TRANSPARENT SKY —
       31.3% of hero-wide-v6.webp is alpha 0, and the comment at the
       top of this block still says, correctly of that file, "where
       an edge would be, the file is transparent. SO THERE IS NO
       EDGE MASK AT ALL." That was true and is no longer: measured
       on the shipped v2 files, day/hero-home-wide.webp is 0.0%
       transparent and its top row is a painted sky,
       rgb(129,201,252) at the left and rgb(115,197,254) at the
       centre.

       WITHOUT A TOP FADE THE RESULT IS A HARD HORIZONTAL SEAM where
       the bright painted sky meets the hero ground — looked at, not
       inferred, in the first render after the file was swapped. The
       hero's own background now ramps into --sky (see .homehero), so
       the art's top has to dissolve INTO that ramp rather than start
       against it.

       0 -> 36%, which is the canvas preview's value for the same
       art (v2-src/css/d2.css fades 0->36% on the home hero). 36% of
       the 366.59px box is 132px of fade — long enough that the
       transition is invisible at a glance and short enough to leave
       the skyline, the trees and the car untouched: the art's
       horizon sits at ~44% of its height, so the fade finishes
       before the city does.

       mask-composite: intersect (and -webkit-mask-composite:
       source-in, which is the same operation under the old syntax)
       IS NOW NEEDED, where the one-axis version did not need it.
       Two gradients on two different axes must be multiplied
       together; listing them without a composite mode makes the
       second simply replace the first in some engines and union in
       others, and "the top fade cancels the left fade" is how the
       copy stops being readable. */
    -webkit-mask-image:
      linear-gradient(to right, transparent 0, #000 14%, #000 91%, transparent 100%),
      linear-gradient(to bottom, transparent 0, #000 36%);
            mask-image:
      linear-gradient(to right, transparent 0, #000 14%, #000 91%, transparent 100%),
      linear-gradient(to bottom, transparent 0, #000 36%);
    -webkit-mask-composite: source-in;
            mask-composite: intersect;
    -webkit-mask-repeat: no-repeat;
            mask-repeat: no-repeat;
    -webkit-mask-size: 100% 100%;
            mask-size: 100% 100%;
    -webkit-mask-position: 0 0;
            mask-position: 0 0;
  }

  /* THE NIGHT ART. Same box, same size, same mask — only the file
     changes, so nothing can move. Both selectors for the usual
     reason: the media query is the no-JS fallback for a dark OS,
     the attribute is the explicit Dark choice.

     THE TWO MUST SHARE A BREAKPOINT. They did not until 23 Oct 2026,
     and this comment asserted a coverage the stylesheet did not have;
     see the band measured at the fallback rule below. */
  :root[data-theme="dark"] .hero-art-layer {
    background-image: url("/static/img/night/hero-home-wide.webp");
  }
}
/* ⚠️ 61rem, AND IT MUST MATCH THE DAY RULE'S BREAKPOINT EXACTLY.
   This said 69rem until 23 Oct 2026, and the mismatch was a visible
   fault with JavaScript OFF on a dark OS.

   .hero-art-layer becomes display:block at 61rem (976px) and the DAY
   art is set in that same block. The no-JS dark fallback was gated at
   69rem (1104px), so from 976px to 1103px nothing overrode the day
   file: the page painted the dark navy background (#101D30) with the
   SUNLIT daytime street pasted into it. Measured with script
   execution disabled and prefers-color-scheme: dark —

       976px  body #101D30   art day/hero-home-wide.webp   WRONG
      1040px  body #101D30   art day/hero-home-wide.webp   WRONG
      1103px  body #101D30   art day/hero-home-wide.webp   WRONG
      1104px  body #101D30   art night/hero-home-wide.webp ok

   WHY IT HID. With JavaScript on, theme-init.js sets data-theme="dark"
   and the attribute rule above — which IS at 61rem — covers the band,
   so every scripted check passes at every width. Only the scriptless
   path reaches this rule, and that is precisely the path the v2 brief
   requires to work ("JS-off falls back to prefers-color-scheme").

   THE 69rem WAS COPIED FROM THE OLD BOUND. .hero-art-layer was
   display:none below 69rem until 9 Oct 2026, when it moved to 61rem to
   meet .hero-phone-art. The display bound moved; this one did not.
   If the display bound moves again, MOVE THIS WITH IT. */
@media (min-width: 61rem) and (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .hero-art-layer {
    background-image: url("/static/img/night/hero-home-wide.webp");
  }
}

/* ===================================================================
   THE ART'S LEFT EDGE IS PINNED THROUGH THE 976-1103 BAND.
   Added 9 Oct 2026 with the breakpoint move, and it is a CONTRAST
   fix, measured, not a tuning preference.
   ===================================================================

   THE RULE ABOVE ANCHORS THE ART TO THE CONTENT COLUMN'S RIGHT EDGE
   (-110px). Above 1200px the column is capped at --site-max and
   centred, so the art holds still relative to the copy - which is
   Guillaume's requirement, "THE CAR SHOULD NOT MOVE RELATIVE TO THE
   TEXT AND PLAYING VIDEO". BELOW 1200px the column IS the viewport,
   so the anchor drags the art left as the window narrows:

     vw     art left edge     clarify text box     art over text
     1280        370              80..458              88px
     1200        330              40..418              88px
     1104        234              40..418             184px
     1024        154              40..418             264px
      976        106              40..418             312px

   AND THAT IS A READABILITY FAULT, NOT ONLY AN AESTHETIC ONE. The
   left mask ramp (14%, see above) exists precisely "for READABILITY,
   not for an edge: it fades the scene out from under the copy". 14%
   of 980px is 137px of fade, which clears the copy at 1280 and does
   not clear it at 976. Sampled from the rendered PNG - the fraction
   of background pixels inside the clarify text's own box measuring
   below 4.5:1 against its ink:

     vw      unpinned      pinned (this rule)
     1280      6.8%            6.8%   (unchanged - outside the band)
     1104     13.6%           13.6%   (unchanged - outside the band)
     1088     15.1%           13.6%
     1024     22.7%           13.6%
      976     23.5%           13.3%

   13.6% at 1104 is the SIGNED-OFF figure: that is what the shipped
   build does at the width Guillaume approved. Pinning holds the whole
   band at that same figure instead of letting it degrade to 23.5%.
   The decorative tagline improves more, from 40.5% to 18.2% at 976.

   WHAT THE PIN IS. Through the band the art's left edge is held at
   x=234 - exactly where it sits at 1104px, the band's top. So the
   composition a reader sees at 1104 is the composition they keep as
   they shrink to 976; the car stops moving relative to the text, for
   the same reason and by the same principle as the fixed 980px width.
   The art still bleeds off the right edge, as it does at 1104.

   WHY NOT WIDEN THE MASK INSTEAD. Tried and measured first: a
   width-stepped ramp (33% at 976 falling to 19% at 1104) only reached
   16.8% at 976, and 33% of 980px is 323px of fade - far enough in to
   dissolve the front of the car, which is the subject. Rejected on
   the measurement, not on taste.

   WHY NOT CLAMP THE ANCHOR WITH min(). Also tried: a min() that
   stopped the left edge at the copy's right edge bit ABOVE the band
   too, moving h1->art from -250 to -190 at 1280 and -150 at 1200.
   That changes the signed-off wide composition, which must not move.
   A band-scoped override cannot reach outside the band.

   THE UPPER BOUND IS 68.9375rem AND IT IS LOAD-BEARING. At 69rem and
   above the unpinned anchor must win, or the wide composition moves.
   Verified after this change: 1104, 1200, 1280, 1920, 2560, 3440 and
   3840 all measure exactly as they did on dcb01ab. */
@media (min-width: 61rem) and (max-width: 68.9375rem) {
  .hero-art-layer {
    /* 234px is the art's left edge at 1104px: the viewport width
       (1104) minus the 980px art plus the 110px overhang, i.e.
       1104 - 980 + 110 = 234. Written as the right offset the rule
       above uses, so the two are the same kind of value. */
    right: calc(100vw - 234px - 980px);
  }
}

.hero-inner {
  position: relative;   /* above the artwork layer */
  z-index: 1;
  max-width: var(--site-max, 75rem);
  margin-inline: auto;
  padding-inline: clamp(1.25rem, 4vw, 2.5rem);
  /* 24px top / 32px bottom at 360px, 32px / 48px at desktop. */
  padding-block: 1.5rem 2rem;
  display: grid;
  gap: 1.5rem;
}

/* TWO COLUMNS: flexible copy + a 440px preview, 40px gap, per
   layout.md.

   THE BREAKPOINT IS 61rem (976px), LOWERED FROM 69rem ON 9 Oct 2026,
   and the old bound's reasoning is worth stating because it was not
   wrong, it was untested. It read: "the breakpoint is 69rem (1104px),
   just above the 1099px where layout.md says to stack — below that
   the 440px card plus a readable copy measure no longer fit."

   MEASURED, THEY DO FIT. At 976: gutters 2x39, copy column 418px,
   gap 40px, card 440px — 977px of content in a 976px box with zero
   horizontal overflow at every width tested from 976 to 3840. The
   copy column reaches its minimum of 418px at the breakpoint itself
   and grows from there; the headline's 540px cap simply stops
   binding below 1103px and the text reflows.

   "A readable copy measure" was the real question and the answer is
   that 418px is one. The headline is clamp()-sized, so at 976 it
   renders at 44px rather than 60px, and 418px at 44px is about 24
   characters a line — tighter than the 540px desktop measure and
   well inside what the phone board already accepts at 350px.

   See "THE 976–1103 BAND" below for why this moved and what the
   band looked like before. */
@media (min-width: 61rem) {
  .hero-inner {
    grid-template-columns: minmax(0, 1fr) 440px;
    gap: 40px;
    padding-block: 32px 48px;
    align-items: start;
  }
}
/* THE CARD GROWS WITH THE CONTAINER, 16 Oct 2026. Widening --site-max
   to 1440 above 90rem moved the margins in but left this track at a
   literal 440px, so the card's share of the row FELL from 36.7% to
   30.6% — the opposite of the complaint. 530px restores the board's
   proportion in the wider box (530/1440 = 36.8%, against 440/1200 =
   36.7%).

   Stated as a second literal rather than a percentage because the
   copy column is the flexible track: a % here would fight minmax(0,
   1fr) for the remainder and the two would re-solve at every width.
   One number per container width is easier to reason about and is
   what the board specifies. */
@media (min-width: 90rem) {
  .hero-inner { grid-template-columns: minmax(0, 1fr) 530px; }
}

/* THE ART GROWS WITH THE CONTAINER TOO, AND THIS WAS A BUG I SHIPPED.
   16 Oct 2026. Guillaume, with a Firefox screenshot: "the spacing
   between text and art on the right is too wide".

   THE CAUSE WAS THE CONTAINER WIDENING EARLIER TODAY, not a browser
   difference — I looked for one twice and was wrong both times. The
   art's right offset is calc'd FROM --site-max:

     right: (100vw - site-max)/2 - 110px

   Widening site-max 1200 -> 1440 makes that offset 120px SMALLER,
   which slides the whole box right; and since the width was a fixed
   880px, the LEFT edge slid with it. Measured as the art's left edge
   relative to the container's left edge:

     site-max 1200, art 880   artL - containerL = +430   (was always this)
     site-max 1440, art 880   artL - containerL = +670   (240px too far right)

   That 240px IS the gap. It shows in Firefox and hides in Chrome only
   because the 14% fade ramp blurs the boundary differently; the box
   is in the wrong place in both.

   1120 = 880 + 240, the same amount the container grew, which puts
   artL - containerL back to +430 at 1600, 1920 and 2552 — verified at
   all three. Height follows the 1942x809 ratio: 1120 x 809/1942 =
   466.57, stated rather than derived for the reason on the base rule.

   THE OVERHANG AND THE MASK ARE UNCHANGED. -110px still hangs the
   composition off the container's right edge, and the 14%/91% ramp is
   a PERCENTAGE of the box, so a wider box fades over proportionally
   more pixels and the scene still ends in the purple. */
@media (min-width: 90rem) {
  .hero-art-layer {
    width: 1120px;
    height: 466.57px;
  }
}
/* ===================================================================
   THE HONESTY CHIPS, on any video frame
   ===================================================================

   Restored 3 Oct 2026. Both homepage boards draw them and the
   canvas's red note calls them load-bearing; the live page had
   neither, because they went out with the rest of the card's text
   when it was stripped on 30 Sep.

   OPAQUE, NOT TRANSLUCENT, and the reason is measurable. These sit
   over video, so what is behind them changes 30 times a second —
   bright sky, dark tarmac, a white van. A translucent chip has no
   single contrast ratio and therefore cannot be shown to pass
   anything. The fill is fixed, so the only ratio that matters is
   text on chip:

     #FFF9F1 on #1F0D53 (dark chip) = 15.37:1
     #291449 on #FFFFFF (paper chip)  = 15.23:1

   Both are the same in light and dark mode, because a chip on a
   photograph is part of the picture and the picture does not
   invert. */
.vid-chip {
  position: absolute;
  left: 12px;
  bottom: 10px;
  z-index: 2;
  margin: 0;
  padding: 4px 8px;
  border-radius: 4px;
  background: #0B1626;
  color: #FFF9F1;
  font-size: 0.625rem;
  line-height: 1.4;
  font-weight: 500;
  letter-spacing: 0.005em;
  pointer-events: none;
}
/* ===================================================================
   THE 91-SECOND EXPLAINER
   ===================================================================

   A third child of .hero-inner, so one `order` can move it on phone
   without a second copy of the markup. See index.html for why it
   could not live further down the page.

   ON DESKTOP IT IS DELIBERATELY UNDECIDED. Guillaume has not chosen
   whether the explainer or the dashcam loop leads above 69rem, so
   this spans both columns below the pair — the band Home.dc.html
   draws — and nothing here presumes the answer. */
.explainer {
  margin: 0;
  /* The SOFT ink too: --ink-soft flips to #C9B8E8 in dark mode and
     measured 1.58:1 on this pinned fill. Pinning one ink and not
     the other leaves the quiet text invisible. */
  --ink-soft: #40566B;
}
/* The band sits AFTER the hero, on the panel ground, which is what
   Home.dc.html draws.

   IT WAS BRIEFLY A CHILD OF .hero-inner, to let one `order` lift it
   above the card on phone. Measured at 1280 that grew the hero from
   634px to 1136px, and the street artwork is bottom-anchored, so it
   slid down with it and painted from y662 to y1252 — below the
   headline, behind this poster.

   THAT ORDERING PROBLEM NO LONGER EXISTS BELOW 69rem, because the
   card it was ordering against is gone there (5 Oct 2026). The
   display:contents reorder at the stack point is kept all the same:
   it still numbers the remaining sections, and it is what puts the
   hero copy above this band rather than below the footer. */
/* IT FOLLOWS THE THEME SINCE V2 — the same change, and the same
   reason, as .portal. See the long note there.

   WHAT THIS SAID AND WHY IT WAS RIGHT AT THE TIME: "PINNED LIGHT...
   Both bands are drawn on the panel ground by a board that defines
   NO dark theme (grep prefers-color-scheme over either board returns
   0), and both hold artwork that is light in every theme: this one a
   cream poster, the portal a white dashboard. Letting the band
   follow --panel put light artwork on a dark ground in neither
   theme's design." Guillaume saw that fault on his phone in dark
   appearance while DevTools emulated light.

   HomeV2Dark.dc.html answers it now. Sampled off the rendered board,
   the how-it-works band is #101D30 — the board draws it DARK, with
   the film card's own light poster sitting on it as a card, which is
   what a card is for. So the band follows --panel and the ink
   follows --ink, and the two move together.

   THE POSTER IS STILL LIGHT IN BOTH THEMES and that is deliberate:
   it is a real first frame of a real film, not a theme element, and
   the brief forbids swapping it (see theme.js on syncPosters). A
   light poster inside a card on a dark band is the board's own
   composition — the thing Guillaume objected to was a light
   dashboard with NO card, bleeding into a dark band. */
/* --page, NOT --panel, AND THE BOARDS AGREE WITH EACH OTHER.
   Sampled down the left gutter of both renders: HomeV2 draws this
   band #FFF9F1 and HomeV2Phone draws it #FFF9F1 -- the page ground,
   not the pale-sky panel. The page had --panel, which put a second
   blue band directly under the blue hero and made the two read as
   one long panel rather than as the hero and then a section.

   The portal band below keeps --section-alt (#EAF4FA), which is
   what the boards draw there, so the page still alternates: hero
   sky, explainer cream, contextual navy, portal pale, closing
   cream. */
.explainer {
  background: var(--page);
  border-block: 1px solid var(--line);
  padding-block: 2rem;
}
@media (min-width: 61rem) {
  .explainer { padding-block: 3.5rem; }
  /* 760px TO MATCH THE CONTEXTUAL FILM, Guillaume 16 Oct 2026: "make
     the how it works video width match that of the Contextual AI
     video". Measured before: explainer 680, ctxfilm 760, both centred
     on the same axis, so the two cards stepped in and out by 40px a
     side down the page.

     THIS DEVIATES FROM Home.dc.html, WHICH DRAWS 680. The board also
     predates the contextual section becoming a single film in a
     760px frame, so the two were never drawn against each other.
     Flagged rather than silently built: the canvas should be updated
     to match, and until it is, this rule and the board disagree on
     purpose.

     margin-inline is what centres it; see the reset below, which must
     not take the inline axis back. */
  .explainer-figure { max-width: 760px; margin-inline: auto; }
}
/* .explainer-h IS BACK, LATER THE SAME DAY. This comment used to say
   it and .explainer-sub had been removed on Guillaume's "the film
   speaks for itself" — true for a few hours, and false as of item 2
   of his 4 Oct spec, which puts a plain "How it works" heading above
   the card on both boards. Its rule is further down with the rest of
   the card.

   .explainer-sub is still gone and is NOT coming back: neither board
   draws a subtitle under the heading. The chip carries the duration
   and the sound warning, which is what the old subtitle said.

   The figure's own margin-inline: auto is what centres the frame. */

/* <figure> CARRIES A UA margin-inline: 40px, and clearing only
   margin-BLOCK leaves it. Above 61rem that default is overridden by
   `margin-inline: auto` in the block above, so it never showed there;
   below 61rem nothing was overriding it until the full-screen phone
   rule did, and that rule is now deleted. Measured on an iPhone 15
   render: the card came out 313px wide at x=40 in a 393px viewport
   instead of 350 at x=20.

   THE RESET IS INSIDE A max-width QUERY, AND THAT IS THE WHOLE
   POINT. Writing it as an unconditional `.explainer-figure {
   margin-inline: 0 }` here is what I did first, and it silently
   un-centred the card on desktop: this rule sits AFTER the 61rem
   block, a media query adds no specificity, so between two equally
   specific rules source order decides and `0` beat `auto`. Measured
   at 1280 before the fix: the card's centre was 212px LEFT of the
   viewport centre while its heading was dead centre — the exact
   "explainer jammed left in a centred section" fault recorded
   against 3 Oct in the design-canvas skill, reintroduced by the same
   mechanism. Scoping it to the width that needs it is what keeps the
   two rules from fighting at all. */
.explainer-figure { margin-block: 0; }
@media (max-width: 60.9375rem) {
  .explainer-figure { margin-inline: 0; }
}


/* The play control. Same pattern as the dashcam card: a real
   <button>, because opening a video is an action and an action has
   to work from a keyboard. All the UA button styling is undone, and
   `display: block` matters — inline-block adds a baseline gap under
   the poster and makes the box taller than 16:9. */
/* THE STAGE holds the film and the overlay in one 16:9 box, so the
   two occupy exactly the same bounds and the overlay can be removed
   without the box changing size. It is what replaced the old
   reveal-then-play sequence on 5 Oct 2026: the <video> is in the
   layout from the first paint and the OVERLAY is what disappears.

   NO LAYOUT SHIFT, which is the reason the ratio is on the stage
   rather than on either child. The box is 16:9 before the video has
   any metadata, while the poster draws, and after the overlay goes —
   three states, one height. The video's width/height attributes stay
   for the same job on a page with no CSS. */
.explainer-stage {
  position: relative;
  width: 100%;
  aspect-ratio: 16 / 9;
  border-radius: 12px;
  overflow: hidden;
  background: var(--hero-bg, #142B49);
}
/* AN OVERLAY, NOT A SIBLING THE VIDEO REPLACES. It covers the stage,
   so the badge and the Watch pill sit over the video's own poster
   frame and a tap anywhere on the picture still activates it — which
   is what the old full-bleed button gave and what the boards draw.
   Transparent background: the picture behind it IS the poster now. */
.explainer-play {
  position: absolute;
  inset: 0;
  /* z-index, and without it the feature does not work at all.
     Guillaume, 5 Oct 2026: "the how it works does not play on the
     localhost".

     Both this overlay and the <video> are positioned children of the
     stage, so painting order decides which receives the click — and
     the video comes LATER in the DOM, so it won. Measured:
     elementFromPoint at the button's own centre returned
     VIDEO.explainer-video, and a trusted click never reached the
     handler. A JS .click() did, which is exactly why this looked
     fixed under test and was broken in a browser.

     This is a consequence of removing the reveal-then-play step. The
     video used to be display:none until the click, so nothing could
     be on top of the overlay; now it is always in the layout. */
  z-index: 2;
  display: block;
  width: 100%;
  height: 100%;
  border-radius: inherit;
  overflow: hidden;
  background: transparent;
  padding: 0;
  border: 0;
  margin: 0;
  font: inherit;
  color: inherit;
  text-align: inherit;
  cursor: pointer;
  -webkit-appearance: none;
  appearance: none;
}
.explainer-play:focus-visible {
  outline: 3px solid var(--accent);
  outline-offset: 3px;
}
/* .explainer-poster IS GONE — 5 Oct 2026. It styled an <img> inside
   the button that served the same explainer-poster.jpg the <video>
   already names in its `poster` attribute. With the film visible from
   the first paint, the video's own poster draws that frame and the
   second element was one request for the same bytes. The rule went
   with the element rather than being left as a dead selector. */
/* THE CENTRED PLAY CUE IS GONE — 4 Oct 2026, with the full-screen
   phone treatment above.

   .explainer-cue was a 72px orange circle in the middle of the
   poster, carrying a 3px --ink ring. The ring was added earlier the
   same day for a measured reason worth keeping on record: the poster
   had been re-cut to a frame where the character stands in an ORANGE
   SUNBURST, and --cta against that sunburst measured 1.03:1, so the
   circle vanished and only its dark triangle read as a control.

   THE BOARD'S ANSWER WAS TO MOVE IT, NOT TO RING IT. Item 2 replaced
   this with a dark "Watch" pill bottom-right, and the board's note
   gives the same observation as the reason: a round orange button in
   the centre "vanished into the orange circle behind the character
   and covered her". A dark pill clear of her face solved the contrast
   problem and the covering problem at once.

   ⚠ AND ON 8 NOV 2026 THE CENTRE BUTTON CAME BACK, so this note is
   history rather than current policy and the pill it points at no
   longer exists. Guillaume chose the "player look" (option B): a 92px
   centre button whose RING is what makes it safe on the sunburst. The
   1.03:1 figure above is why that ring is load-bearing and not
   decoration, and it was re-measured against the actual poster before
   shipping — the numbers, and the one place the build departs from
   the canvas note, are on .vp-big. See .vp-big, not .explainer-watch,
   which is deleted. */
/* "91 seconds · sound on" — the duration and the fact that it is
   NOT silent, both before the click. A narrated video that starts
   talking unannounced is the thing people hate most about autoplay,
   and saying so first is the cheap fix.

   IT MOVED FROM TOP-RIGHT TO TOP-LEFT, 4 Oct, because that is where
   both boards draw it, and the move is what frees the bottom-right
   corner for the Watch pill. The two must not share a corner: the
   chip is the fact, the pill is the action.

   OPAQUE WHITE, not the old rgba(255,249,241,0.94). The frame behind
   it is a real video still whose top-left is bright sky in this cut
   and could be anything in the next one, so the chip carries its own
   solid ground rather than tinting whatever is underneath.
   #291449 on #FFFFFF measures 16.27:1.

   LITERALS, NOT --paper AND --ink, AND THAT IS THE POINT. I wrote
   this as `background: var(--paper); color: #291449` first and it
   was invisible in dark mode: --paper flips to #291449, which is
   exactly the ink colour, so the chip rendered as an empty dark
   rectangle. Caught on the iphone-15 dark shot from
   measure-layout.js --shots, not by reading the rule.

   A THEME TOKEN IS WRONG HERE IN PRINCIPLE, not just in that pairing.
   This chip sits on a photograph, and the photograph does not change
   with the theme — the same frame, the same bright sky, in light and
   dark. A surface that tracks the UI theme while the thing behind it
   does not is how a label ends up unreadable in one mode. The two
   Contextual AI chips and the dashcam card's are literal for the same
   reason; see .vid-chip. */
.explainer-badge {
  position: absolute;
  left: 12px;
  top: 12px;
  z-index: 2;
  padding: 5px 9px;
  border-radius: 6px;
  background: #FFFFFF;
  color: #142B49;
  font-size: 0.6875rem;
  font-weight: 700;
  line-height: 1.3;
  pointer-events: none;
}

/* The video is hidden until the button is pressed; flow.js adds
   .is-playing. Both states occupy the same 16:9 box, so the band
   does not jump when it swaps. */
/* THE BLACK BARS, fixed 4 Oct. Guillaume: the video played with
   roughly a third of the box black, above and below.

   TWO CAUSES, and the obvious one was not the whole story.

   1. NO object-fit. A <video> defaults to `object-fit: contain`,
      which fits the picture inside the box and paints the remainder
      with the element's background — #000 here, hence bars rather
      than a gap.

   2. height="720" AS AN ATTRIBUTE BEAT THE aspect-ratio. Measured
      before the fix the box was 680x720, ratio 0.944 — nearly
      square, not the 680x382 a 16:9 box gives. `aspect-ratio` only
      derives the height when the height is `auto`, and the
      presentational attribute supplies a used height of 720px. So a
      16:9 picture sat in a near-square box and contain letterboxed
      it heavily.

   `height: 100%` is what the dashcam card already had, which is why
   that one was never affected — see .driving-preview__video.

   WHY cover AND NOT contain. Both are identical today: the encode is
   1280x720 and the box is 16/9. They differ only if a future re-cut
   changes ratio, and then cover crops while contain letterboxes.
   Cropping a few pixels off a marketing clip is better than
   reintroducing the black bars this comment exists about, and the
   dashcam card made the same choice for the same reason. The
   width/height ATTRIBUTES stay: they reserve the right box before
   the metadata loads, which is what stops the page jumping, and with
   an explicit CSS height they no longer fight the box. */
/* display:none IS GONE FROM THIS RULE — 5 Oct 2026, the whole point
   of the stage above. The film is in the layout and visible from the
   first paint, drawing its own poster until playback starts.

   The ratio now lives on .explainer-stage, so this element simply
   fills it: a second aspect-ratio here would be a second opinion
   about the same box. object-fit: cover and an explicit height are
   kept for the reason the comment above this block gives — a <video>
   defaults to `contain`, and a height ATTRIBUTE beats aspect-ratio,
   which is what put black bars on this film on 4 Oct.
   TestVideosFillTheirBoxes guards it. */
.explainer-video {
  display: block;
  width: 100%;
  height: 100%;
  object-fit: cover;
  border-radius: inherit;
  background: #000;
}
/* THE OVERLAY IS HIDDEN UNTIL THE SCRIPT SAYS IT WORKS.
   ---------------------------------------------------------------
   .explainer-play is a button whose only job is to call play(). With
   no JS — or before flow.js has run — pressing it does nothing, and
   a dead control over a video is worse than no control: the viewer
   taps, nothing happens, and there is no second affordance.

   So the DEFAULT is overlay hidden, and flow.js adds .is-ready to
   the stage once it has wired the handler. A no-JS visitor gets the
   <video> itself, showing its own poster, with the browser's own
   keyboard and context-menu controls — a worse player than the pill,
   and a reachable one. Before 5 Oct they got a dead button over a
   display:none element and could not watch the film at all.

   A <noscript><style> WOULD HAVE BEEN SIMPLER AND IS BLOCKED. The
   CSP is `style-src 'self'`, so an inline <style> is dropped; it was
   tried, served and silently ignored, which is the worst of the
   three outcomes because the markup looks right. */
.explainer-play { display: none; }
.explainer-stage.is-ready .explainer-play { display: block; }

/* ON PLAY, THE OVERLAY GOES AND THE FILM STAYS WHERE IT WAS. This
   pair used to read `.explainer-play { display: none }` AND
   `.explainer-video { display: block }` — the second half is what
   made the film hidden until the tap, and it is gone: the video is
   never display:none now, so there is nothing to reveal. Only the
   overlay is toggled, and the stage's height never changes. */
.explainer-figure.is-playing .explainer-play { display: none; }

/* ===================================================================
   THE FULL-SCREEN PHONE EXPLAINER IS GONE — 4 Oct 2026
   ===================================================================

   A 100svh block lived here for a few hours today: the band's gutter
   and padding cancelled, the poster cropped to a portrait with
   object-fit, a 260px scrim at the bottom, and the cue, badge and
   honesty chip floated up off the bottom edge. It is deleted, not
   disabled, and this note is what remains of it so the next reader
   does not reinstate it as a "fix".

   WHY IT WENT. Guillaume, having looked at both: the bare desktop
   band and the full-screen phone crop both "read as odd". Item 2 of
   his 4 Oct spec replaces BOTH with one treatment at every width —
   heading, then the poster whole in a white framed card, matching the
   hero card and the two Contextual AI cards. The two are not layered:
   a framed card inside a full-screen crop is neither thing.

   WHAT THE DELETION FIXES FOR FREE. That block's own comment recorded
   that the portrait crop removed the burned-in "Planned service /
   Illustrative preview" words altogether, which is why an HTML copy
   of them had to be added below 61rem. Showing the frame whole at
   16:9 brings the burned-in words back, so .explainer-chip and the
   61rem rule that hid it on desktop are both gone with it — there is
   no width now at which the poster is cropped.

   NO NEW BREAKPOINT, and one fewer rule: the card is the same at
   every width bar its padding and type size.

   =================================================================== */

/* THE WHITE FRAME. The same component as the hero card and the two
   Contextual AI cards: paper ground, 12px of padding as the visible
   frame, a soft shadow, and the 16:9 picture inset inside it with its
   own smaller radius. The boards draw 16px/12px radii and a 12px pad
   on desktop, 14px/8px and 10px on phone.

   680px IS KEPT AND IS NOT NEW. Home.dc.html draws the card at
   max-width 760px INCLUDING its 12px of padding each side, so the
   picture inside is 736px. The page has drawn this frame at 680px
   since 3 Oct and that is the figure .explainer-figure already
   carried; it is the narrower of the two and I have not widened it,
   because widening the only element on the band is a composition
   change and composition belongs on the board. REPORTED, not fixed:
   board 760px, page 680px. */
/* THE FIGURE IS THE CARD. No wrapper element was added for it: the
   <figure> was already the max-width container and the thing that
   centres, so giving it the paper ground and the padding makes one
   box do one job. A nested .explainer-card div would have been a
   second box with the same bounds. */
.explainer-figure {
  background: var(--paper);
  border-radius: 14px;
  padding: 0.625rem;
  box-shadow: 0 18px 40px rgba(11, 22, 38, 0.18);
}
@media (min-width: 61rem) {
  .explainer-figure {
    border-radius: 16px;
    padding: 0.75rem;
  }
}
/* The picture inside the frame takes the smaller radius, so the paper
   reads as a mount around it rather than as a border on it. */
/* ON THE STAGE, not on the button: the stage is the box with the
   picture in it now, and both the film and the overlay inherit from
   it. Before 5 Oct the button WAS that box. */
.explainer-stage { border-radius: 8px; }
@media (min-width: 61rem) {
  .explainer-stage { border-radius: 10px; }
}

/* "How it works". Centred over the card, like the portal's and the
   Contextual AI section's headings.

   ONE SECTION-HEADING SIZE, 5 Oct 2026 (Guillaume). Both boards carry
   the note "ALL section headings one size: 48px, 800, -0.02em,
   line-height 1.1. Hero h1 stays larger." This rule used to set
   34px/26px of its own, which is what the boards drew BEFORE that
   normalisation — it is now --section-h, the same token .portal-h and
   .ctxai-h use, so the five section headings cannot drift apart again.
   line-height is 1.1 with them, not the 1.15 this had. */
.explainer-h {
  margin: 0 0 1.125rem;
  text-align: center;
  font-size: var(--section-h);
  font-weight: 800;
  letter-spacing: -0.02em;
  line-height: 1.1;
  /* INHERIT, don't re-assert --ink. The band pins its own ink because
     its fill is pinned light (see .explainer); naming --ink here
     overrode that pin and put cream on #F3ECFD in dark mode —
     1.10:1, which is what Guillaume could not read on 6 Oct 2026. */
  color: inherit;
}
@media (min-width: 61rem) {
  .explainer-h { margin-bottom: 1.5rem; }
}

/* ===================================================================
   THE "PLAYER LOOK" VIDEO COVER — BOTH FILMS, ONE SET OF RULES
   ===================================================================
   8 Nov 2026. Guillaume chose option B of three: "people didn't
   realise the cover was a video." Built from the canvas note
   `howitworks`, whose values are copied here rather than
   approximated — 92/68px button, 7px ring, 96/72px fade,
   rgba(12,24,42,.1) wash rising to .18 on hover.

   THE WATCH PILL IS GONE AND ITS RULES WITH IT. .explainer-watch and
   .explainer-watch-cue lived here and are deleted, not disabled. The
   pill was the 4 Oct answer to a measured problem — see .vp-big below,
   which records the measurement and why a centre button is safe again.
   Two controls advertising one action is one too many.

   .badge-secs IS GONE TOO. It hid the duration from the contextual
   film's chip below 37.5rem, because "70 seconds · sound on · planned
   capability" measured 248px inside a 326px video. The duration now
   lives in the strip for both films, so the chip is short at every
   width and the override has nothing left to do. */

/* THE HIT AREA IS THE WHOLE COVER. It is .explainer-play that
   positions and sizes this (inset:0, 100%x100%, z-index:2 — and that
   z-index is load-bearing, see its own note: the <video> comes later
   in the DOM and wins the click without it). .vp-hit only adds what
   the player look needs on top, which is the focus ring and the
   colour the strip inherits. */
.vp-hit {
  color: #FFFFFF;
}
/* 3px #FFAA70, from the note. NOT var(--accent): this ring is drawn
   against a photograph that does not change with the theme, which is
   the same rule .explainer-badge carries in full. --accent differs
   between light and dark and would make the ring's visibility a
   function of a theme the poster does not follow.

   outline-offset puts it OUTSIDE the cover's rounded edge so it is
   never confused with the fade at the bottom. */
.vp-hit:focus-visible {
  outline: 3px solid #FFAA70;
  outline-offset: 3px;
}

/* THE WASH. 10% navy over the poster, 18% on hover — enough to seat
   the white furniture on a bright frame without dulling the picture.
   It is a child rather than a background on the button because the
   button's background must stay transparent: the poster behind it is
   the <video>'s own `poster` attribute. */
.vp-dim {
  position: absolute;
  inset: 0;
  background: rgba(12, 24, 42, 0.1);
  transition: background 0.15s;
}
.vp-hit:hover .vp-dim { background: rgba(12, 24, 42, 0.18); }

/* ⚠️ THE CENTRE PLAY BUTTON, AND THE RING IS LOAD-BEARING.
   ---------------------------------------------------------------
   A 72px orange circle stood here until 4 Oct 2026 and was REMOVED
   because it vanished. The poster had been re-cut to a frame where
   the character stands in an ORANGE SUNBURST, and --cta against that
   sunburst measured 1.03:1. The board's answer then was to move the
   control to a dark pill bottom-right rather than to ring it.

   OPTION B BRINGS IT BACK, so the contrast question is live again and
   has been re-measured against the ACTUAL poster files rather than
   assumed. Method: sample the frame at 2° intervals around the ring's
   outer edge, composite the 10% wash over it as it renders, then
   composite each ring layer over that and compute WCAG contrast.
   WCAG 1.4.11 asks 3.0:1 for a non-text control boundary.

     bare orange circle, no ring      1.07:1 worst  (the 4 Oct failure,
                                                     reproduced)
     + white 7px ring (the spec)      1.28:1 worst, mean 8.50:1
     + white ring AND a 2px dark edge 3.71:1 worst, 0/360 points under 3

   THE WHITE RING ALONE IS NOT ENOUGH, and this is the one place the
   build departs from the canvas note. The white ring fixes the case it
   was drawn for — orange-on-orange — and creates the mirror of it:
   against the pale sky at the bottom of the explainer poster,
   rgb(230,225,203), a 90% white ring measures 1.28:1. On the
   contextual film's overcast road, rgb(196,203,214), 57% of the ring's
   circumference sat under 3.0:1.

   SO THE RING IS TWO-TONE: the note's white 7px ring, unchanged, with
   a thin dark edge outside it. One of the two always separates — white
   carries the dark frames, the dark edge carries the pale ones — and
   measured across 360 points on four poster/width combinations the
   worst case is 3.67:1 with nothing below 3.0. The white ring is NOT
   reduced to make room for it; the dark edge is added outside.

   Both rings are box-shadows on one element, so they cost no extra
   DOM and scale together. The drop shadow is the note's third layer,
   unchanged. */
.vp-big {
  position: absolute;
  left: 50%;
  top: 50%;
  transform: translate(-50%, -50%);
  width: 92px;
  height: 92px;
  border-radius: 50%;
  background: var(--cta);
  display: grid;
  place-items: center;
  box-shadow:
    0 0 0 7px rgba(255, 255, 255, 0.9),
    0 0 0 9px rgba(12, 24, 42, 0.85),
    0 14px 30px rgba(20, 43, 73, 0.35);
  transition: transform 0.15s ease;
}
.vp-hit:hover .vp-big,
.vp-hit:focus-visible .vp-big {
  transform: translate(-50%, -50%) scale(1.08);
}
/* The triangle, from borders — one shape, no request, and the CSP
   allows no CDN anyway. #142B49 on --cta measures 7.61:1. margin-left
   nudges it so the SHAPE reads centred rather than its bounding box,
   the same trick the old watch cue used. */
.vp-tri {
  width: 0;
  height: 0;
  border-left: 30px solid #142B49;
  border-top: 19px solid transparent;
  border-bottom: 19px solid transparent;
  margin-left: 8px;
}

/* THE BOTTOM FADE. transparent → rgba(12,24,42,.75) over 96px, so the
   strip has something to sit on whatever the frame does, and so the
   burned-in honesty line bottom-left is not left bright against it.
   .vp-label is drawn on top of this for exactly that reason. */
.vp-fade {
  position: absolute;
  left: 0;
  right: 0;
  bottom: 0;
  height: 96px;
  background: linear-gradient(transparent, rgba(12, 24, 42, 0.75));
}

/* THE PLAYER STRIP. A VISUAL CUE, NOT A PLAYER.
   ---------------------------------------------------------------
   play icon · 0:00 · timeline · total · audio icon. Nothing here is
   wired: the elapsed figure is a literal 0:00 and the track is a flat
   bar, because the film is stopped whenever this is on screen. On play
   flow.js hands over to the native control bar — the handover both
   films already had — and the overlay goes.

   tabular-nums so that 0:00 and the total do not shift the flex row as
   digits change. */
.vp-bar {
  position: absolute;
  left: 14px;
  right: 14px;
  bottom: 36px;
  display: flex;
  align-items: center;
  gap: 10px;
  font-size: 0.75rem;
  font-weight: 600;
  font-variant-numeric: tabular-nums;
  color: #FFFFFF;
  text-shadow: 0 1px 2px rgba(12, 24, 42, 0.6);
}
.vp-icon {
  width: 16px;
  height: 16px;
  flex: none;
}
.vp-track {
  flex: 1;
  height: 4px;
  border-radius: 2px;
  background: rgba(255, 255, 255, 0.35);
}

/* ⚠️ THE TOTAL LENGTH IS EMPTY UNTIL flow.js FILLS IT, AND THE
   min-width IS WHAT STOPS THAT MOVING ANYTHING.
   ---------------------------------------------------------------
   The number is read from the film's own metadata, never typed — the
   canvas note is explicit: "Length from the live film: 86 s (the
   boards said 91)." A literal in the markup is a number that goes
   stale on the next re-cut, silently, in the one place a viewer is
   told what they are committing to.

   preload="none" MUST STAY (27MB of encodes, see the <video>), so the
   duration genuinely is not known on first paint. The strip therefore
   starts with the slot EMPTY and flow.js writes into it when metadata
   arrives — which is after the first play, not before. 4ch reserves
   the width "1:26" will need, so writing it causes NO LAYOUT SHIFT:
   the flex row's geometry is identical empty and filled. CLS is a hard
   rule here and an empty box that grows is the usual way it is broken.

   text-align: right so the digits land against the audio icon rather
   than floating in the middle of the reserved slot. */
.vp-total {
  flex: none;
  min-width: 4ch;
  text-align: right;
}

/* "Planned service / Illustrative preview", REDRAWN OVER THE FADE.
   It is burned into the How it works poster bottom-left and the fade
   now covers that corner, so without this our own gradient would dim
   the honesty label. The canvas note calls for it explicitly.

   LITERALS, NOT --paper/--ink. Same rule as .explainer-badge, whose
   note has the full account: this sits on a photograph, the
   photograph does not change with the theme, and a surface that
   tracks the UI theme while its background does not is how a label
   ends up unreadable in one mode. #FFF9F1 on rgba(12,24,42,.92)
   measures 15.9:1 against the flat colour it is 92% of. */
.vp-label {
  position: absolute;
  left: 10px;
  bottom: 8px;
  background: rgba(12, 24, 42, 0.92);
  color: #FFF9F1;
  font-size: 0.625rem;
  font-weight: 500;
  padding: 4px 8px;
  border-radius: 4px;
}

/* PHONE, from HomeV2Phone: 68px button, 5px white ring, 72px fade,
   and the strip tightened. The dark outer edge scales with it.
   37.5rem rather than one of the four contract breakpoints because
   this is the same width the retired .badge-secs rule used and it is
   where the 16:9 cover gets too small for the 92px button — the
   contract's R1 governs the QUOTE FLOW's shell, and this is a
   homepage component with its own geometry. */
@media (max-width: 37.5rem) {
  .vp-big {
    width: 68px;
    height: 68px;
    box-shadow:
      0 0 0 5px rgba(255, 255, 255, 0.9),
      0 0 0 6.5px rgba(12, 24, 42, 0.85),
      0 10px 22px rgba(20, 43, 73, 0.35);
  }
  .vp-tri {
    border-left-width: 22px;
    border-top-width: 14px;
    border-bottom-width: 14px;
    margin-left: 6px;
  }
  .vp-fade { height: 72px; }
  /* The three sizes are TOKENS because the cqw caps further down this
     file are declared after this block and would otherwise win. See
     the note on .vp-label there. */
  .vp-bar {
    left: 10px;
    right: 10px;
    bottom: 28px;
    gap: 8px;
    --vp-bar-size: 0.6875rem;
  }
  .vp-icon { --vp-icon-size: 14px; }
  .vp-label { left: 8px; bottom: 6px; --vp-label-size: 0.5625rem; padding: 3px 6px; }
  .vp-hit:hover .vp-big,
  .vp-hit:focus-visible .vp-big {
    transform: translate(-50%, -50%) scale(1.06);
  }
}

/* ⚠️ NO SCALING UNDER prefers-reduced-motion, AND THAT MEANS FOCUS
   TOO — WHICH THE BOARD'S OWN CSS MISSES.
   ---------------------------------------------------------------
   The note says "grows slightly on hover/focus" and "Respect reduced
   motion (no hover scaling)". The board's rule kills the transition
   and the HOVER transform, and leaves :focus-visible scaling — so a
   keyboard user who has asked for no motion still gets a jump on Tab.
   That is the same instruction applied to one of its two triggers, so
   both are covered here.

   The transform is reset to the centring translate rather than to
   `none`: `none` would drop translate(-50%,-50%) and throw the button
   off centre, which is a worse bug than the one being fixed. The wash
   keeps its colour change — a background that is not animated is not
   motion. */
@media (prefers-reduced-motion: reduce) {
  .vp-big,
  .vp-dim { transition: none; }
  .vp-hit:hover .vp-big,
  .vp-hit:focus-visible .vp-big { transform: translate(-50%, -50%); }
}

/* ⚠ AT 200% TEXT THE OVERLAY FURNITURE SWALLOWED THE PICTURE.
   ---------------------------------------------------------------
   CAUGHT BY LOOKING at home--galaxy-s24--text200.png from
   measure-layout.js --shots, 8 Nov 2026 — the same pass that caught
   the forced-colors triangle below, and neither was caught by a
   number.

   WHAT IT LOOKED LIKE at 360px with the browser's text set to 200%:
   the "Planned service / Illustrative preview" label wrapped onto two
   lines and grew into a dark slab across the MIDDLE of the cover,
   hiding the play button, the timeline and the elapsed time. The
   "Sound on" chip grew into the play circle. The picture behind was
   almost entirely covered by its own labels.

   THE OVERFLOW CHECKER PASSED THE WHOLE TIME, and correctly: nothing
   left the viewport. The cover is a fixed 16:9 BOX, and everything
   inside it was sized in rem, which scales with the user's text
   setting while the box does not. So the furniture grew and the
   stage did not, which produces no overflow and no misalignment —
   just a label eating the film.

   THE FIX IS cqw, NOT A SMALLER rem. Shrinking the labels would
   punish every reader to protect one setting, and capping them in px
   would ignore the text preference entirely — which is the thing
   WCAG 1.4.4 asks us not to do. A container query unit ties the
   furniture to THE COVER, which is the box it has to fit inside:
   `min()` keeps the rem value (so the preference is honoured and the
   text still grows) until it would be too large for the cover, and
   from there the cover's own width governs.

   The stage declares `container-type: inline-size` so cqw resolves
   against the cover rather than the viewport — a phone cover is
   330px inside a 390px viewport, and the difference matters.

   WHY NOT HIDE THE LABEL AT 200%. It is the honesty line. It is
   burned into the How it works poster and redrawn here precisely so
   it stays readable; hiding it at a text size chosen by people who
   most need to read it would be the worst possible place to drop it.
   It gets smaller relative to the cover, and it stays. */
.explainer-stage,
.ctxfilm-stage {
  container-type: inline-size;
}
/* THE CAP IS WRITTEN AS A TOKEN SO THE PHONE BLOCK KEEPS ITS OWN
   SIZES. These rules sit AFTER the 37.5rem block, so a flat
   `font-size: min(0.75rem, 4cqw)` here would override the phone's
   smaller 0.6875rem and quietly undo it — measured: the strip came
   back as 12px at 390 where the board draws 11px. Each element
   therefore caps against ITS OWN size via a custom property that the
   phone block re-points, which is the same mechanism the spacing
   tokens use in the quote contract. */
.vp-label {
  font-size: min(var(--vp-label-size, 0.625rem), 3.4cqw);
  padding: min(4px, 1.2cqw) min(8px, 2.4cqw);
  max-width: calc(100% - 20px);
}
.vp-bar {
  font-size: min(var(--vp-bar-size, 0.75rem), 4cqw);
}
.vp-icon {
  width: min(var(--vp-icon-size, 16px), 4.4cqw);
  height: min(var(--vp-icon-size, 16px), 4.4cqw);
}
/* The chip is .explainer-badge, which is shared with nothing else on
   the page now that both covers use it, so it is capped here rather
   than at its own rule — the cap is a property of sitting on a cover,
   not of the chip. */
.vp-hit .explainer-badge {
  font-size: min(0.6875rem, 3.6cqw);
  padding: min(5px, 1.4cqw) min(9px, 2.6cqw);
  max-width: calc(100% - 24px);
}

/* ⚠ FORCED COLORS TURNS THE BORDER TRIANGLE INTO A BLACK SQUARE,
   AND THIS PROJECT HAS ALREADY BEEN BITTEN BY EXACTLY THIS.
   ---------------------------------------------------------------
   CAUGHT BY LOOKING at home--desktop-1280--forced-colors.png from
   measure-layout.js --shots, 8 Nov 2026. The play cue rendered as a
   SOLID BLACK SQUARE in the middle of the white ring — not a
   triangle, not a play symbol, just a block.

   THE CAUSE IS THE ONE .brand-c RECORDS IN FULL further up this file:
   Chrome repaints EVERY border side to the system foreground in
   forced-colors mode, THE TRANSPARENT ONES INCLUDED. .vp-tri is a
   zero-size box whose shape is made entirely of one opaque border and
   two transparent ones, so when all three are painted the same colour
   the negative space that made it a triangle fills in and the element
   becomes its own bounding box — 30x38px of solid CanvasText.

   NO TEXT CHECK COULD SEE THIS. The markup is unchanged, there is no
   overflow, contrast is fine (it is a solid block on white, which
   measures beautifully), and compare-board.js deliberately does not
   look at pixels. It was visible only in the picture.

   THE FIX, same as .brand-c: opt the element out of the forced
   palette with forced-color-adjust: none, then restate BOTH
   transparent sides explicitly — `transparent` is a keyword the UA
   stylesheet is allowed to override, so it has to be re-declared
   inside the query rather than relied on from the base rule.

   THE CIRCLE KEEPS ITS FORCED COLOURS and is given an explicit
   border, because opting the triangle out leaves it drawn in --cta
   on whatever Canvas the user has chosen; a 1px CanvasText edge keeps
   the control's outline findable when the fill is not.

   THE CONTROL IS STILL REACHABLE EVEN IF THIS RULE IS WRONG. The
   whole cover is the <button> and its accessible name says "Play How
   it works, 86 seconds, with sound", so this is for people who SEE
   the page in a forced palette, not for assistive technology — the
   same distinction .brand-c draws. */
@media (forced-colors: active) {
  .vp-tri {
    forced-color-adjust: none;
    border-left-color: #142B49;
    border-top-color: transparent;
    border-bottom-color: transparent;
  }
  .vp-big {
    forced-color-adjust: none;
    background: #FFAA70;
    border: 1px solid CanvasText;
  }
}

/* ON PLAY THE WHOLE OVERLAY GOES. .explainer-play already carries
   `display:none` under .is-playing (explainer) and .is-started
   (contextual film), and every piece of the player look is a CHILD of
   that button — so the handover to the native controls needs no new
   rule. That is why the strip is inside the button rather than a
   sibling of it. */


/* ON PHONE THE EXPLAINER IS SECOND. Guillaume, 3 Oct: "the whole
   idea should be the second thing to appear on the phone."

   THIS SURVIVED A ROUND TRIP TODAY AND IS UNCHANGED. For part of
   10 Oct the canvas carried Option A (v152), whose phone board put a
   sign-up band between the hero and the explainer; this block was
   renumbered to suit, and then renumbered back when Guillaume
   replaced Option A with Option B (v153). Option B has no band — the
   waitlist bar is inside the hero copy — so the phone sequence is
   what it was: hero copy, explainer, then the rest. Recorded because
   the numbers below were edited twice in one evening and the history
   would otherwise look like drift.

   THE BAND IS NOT MOVED — THE CARD IS. The explainer is a sibling
   after the hero and cannot be ordered above a hero child; `order`
   does not cross a formatting context. But the dashcam card IS a
   hero child, so moving it to the end of the hero puts the band
   between the copy and it:

     hero-copy, [hero ends], explainer band, ... and the card last.

   Except the card would then still be INSIDE the hero and so still
   above the band. So it is lifted out of the hero's flow entirely
   at this width and placed after the band with `order` on a flex
   body — see .homehero + .explainer below.

   NOTHING IS DUPLICATED. One element, one DOM position; the
   sequence a screen reader hears on phone is the sequence shown. */
@media (max-width: 60.9375rem) {
  /* display: contents IS THE WHOLE TRICK, and it is here because
     nothing simpler works.

     The card (.preview) is a child of .hero-inner, inside .homehero.
     The band is a sibling of .homehero. `order` cannot reorder across
     that boundary — it only sorts siblings — so no amount of ordering
     on either one puts the band between the copy and the card.

     display: contents removes a box from the layout tree while
     keeping its children. Applied to .homehero and .hero-inner, their
     children are promoted to be children of <main>, where the band
     already lives. Now all four are siblings of one flex column and
     one `order` each does the job:

       1 hero copy
       2 the explainer band
       3 the dashcam card

     THE HERO'S OWN GROUND IS THE COST. A box removed with
     display:contents paints nothing, so .homehero's background goes
     with it; it is restored on .hero-copy, which is the part that
     needs to sit on the hero. The artwork layer is already display:none
     below 69rem, so it is unaffected.

     This is a phone-only rule. Above 61rem the hero is an ordinary
     two-column grid and none of it applies. */
  main { display: flex; flex-direction: column; }
  .homehero,
  .hero-inner { display: contents; }

  /* EVERY CHILD NEEDS AN ORDER, not just the three being moved.
     `order` defaults to 0, so the first version — which numbered only
     the hero copy, the band and the card 1/2/3 — sorted all three
     BELOW the steps, the approach band and the signup, which were
     still 0. Measured at 390px: the hero copy started at y2019, under
     the footer sections. The page read bottom-up.

     So the whole column is numbered. 0 would work for the tail, but
     writing them out is what stops the next section added here
     landing in the middle of the hero by default. */
  .hero-copy {
    order: 1;
    background: var(--hero-bg);
    color: var(--hero-text);
    padding: 1.5rem clamp(1.25rem, 4vw, 2.5rem) 2rem;
  }
  /* …BUT NOT INSIDE THE CONFIRMATION PANEL. 28 Oct 2026.

     This block is the HOMEPAGE's narrow-width hero treatment, and
     .hero-copy is a shared class. The confirmation page's panel uses
     it too, so below this breakpoint the copy column was painted
     with --hero-bg INSIDE the cream panel: a blue block sitting in a
     cream card, the inverse of the design, with its own padding
     doubling the panel's.

     Guillaume spotted it as "a bit strange where the green tick is"
     — the tick was the visible symptom of the column it sits in
     having a background it should not have. */
  .confirmsky .hero-copy {
    background: none;
    color: inherit;
    padding: 0;
  }
  .explainer { order: 2; }
  /* .preview IS NOT ORDERED HERE ANY MORE — it is not rendered at
     all below 69rem. See the rule immediately after this block. The
     order number it used to carry (3) is deliberately left unused
     rather than renumbering 4-6 down: those numbers appear in the
     comment above and in commit history, and a gap is cheaper to
     read than a shift. */
  /* .steps became .portal + .ctxai on 4 Oct. Both need a number:
     `order` defaults to 0, so an unnumbered section sorts ABOVE
     everything numbered 1-3 — that is how the hero copy ended up at
     y2019 under the footer the first time this rule was written. */
  /* REORDERED AGAIN 14 Oct 2026, and this REVERSES the 11 Oct v180
     move recorded below. Guillaume's order, from Home.dc.html and
     HomePhone.dc.html, which both already drew it this way:

       hero 1, explainer 2, ctxai 3, portal 4, differentiator 5

     so the page ends on the marital-status band again and the
     "Read our mission" button is the last thing before the footer.

     WHAT THIS GIVES UP, STATED PLAINLY: v180 moved this band to third
     precisely so more visitors would reach that button — "the last
     section on a long page is not reached". That argument was not
     refuted, it was overruled: the boards lead the page, both boards
     draw ctxai third, and Guillaume confirmed the reversal when asked
     directly on 14 Oct. If reach turns out to matter more than order,
     the fix is a second mission link higher up, NOT another silent
     renumber here.

     THE DOM MOVE ALONE STILL WOULD NOT DO IT, which is the durable
     half of the note below and the reason this block exists: below
     61rem <main> is the flex column set up here, and a flex item's
     `order` beats document order. index.html was reordered in the
     same commit; leaving 3/4/5 on the old elements would have given
     desktop the new order and every phone the old one — desktop
     right, phone wrong, the exact shape of failure the design-canvas
     skill records for the phone step indicator.

     The numbers stay contiguous 1-5 and match the DOM, so a reader
     comparing the two sees one order rather than two. */
  .ctxai { order: 3; }
  .portal { order: 4; }
  .differentiator-band { order: 5; }
  /* 6 IS FREE, and has been since 11 Oct 2026. It was
     .differentiator-band back when that band was last on the page.
     The 14 Oct reversal above puts the band last again but numbers it
     5, not 6, because 1-5 contiguous is what matches the DOM — so 6
     stays retired rather than being revived for its old element.
     Left unused for the reason the note on 7 below gives at length: a
     number outliving the markup it was written for is a trap, and the
     hazard here is a section added at the foot of the page silently
     inheriting a position in the middle. */
  /* 7 WAS .signup-band AND IS NOW FREE, 10 Oct 2026. It numbered the
     "Ready when you are" section at the foot of the page, whose
     markup was removed earlier today when the waitlist moved into the
     hero. Option B has no section of that name at all, so nothing
     takes the number. It is left unused rather than reassigned, so a
     reader who finds `order: 7` in the history is not sent to the
     wrong element.

     IT IS ALSO A WORKED EXAMPLE OF THIS BLOCK'S OWN WARNING, by a
     route the warning did not anticipate. During Option A a new band
     reused the class name .signup-band and SILENTLY INHERITED this
     7, landing at the bottom of the phone page — the hero link that
     pointed at it scrolled past six sections to reach it. Measured at
     390: band top y4915.55 against hero copy ending at y545. The
     hazard is not only an UNnumbered child; it is a number outliving
     the markup it was written for. */
}

/* ===================================================================
   THE DASHCAM LOOP CARD IS NOT RENDERED BELOW 61rem — 5 Oct 2026,
   REBOUNDED FROM 69rem TO 61rem ON 9 Oct 2026.
   ===================================================================

   HomePhone.dc.html stopped drawing it. Guillaume's note on the
   board: "not sure what it brings" — directly under the narrated
   explainer it was a second, unexplained video, and it is the loop
   that should not play on mobile data. Real footage is still shown
   WITH its context in the Contextual AI section further down, where
   both honesty chips sit on every frame.

   THE BOUND IS 61rem, NOT 69rem, AND THAT IS THE 9 Oct CHANGE.
   The card is a PHONE-REGIME removal, so it belongs on the phone
   regime's own boundary. At 69rem it was removing the card from
   976–1103px as well — a 128px band of desktop-width windows with
   no video card, which is not what this board decision says. See
   "THE 976–1103 BAND" note further down for the measurements and
   for Guillaume's report of it.

   `display: none` IS THE RIGHT TOOL HERE BUT IT IS NOT THE
   MECHANISM THAT SAVES THE BYTES, and conflating the two is the
   mistake this comment exists to prevent. Measured on this very
   element on 3 Oct: with only `display: none` on the video, Chrome
   still fetched all 12MB and reported readyState 4. The stylesheet
   suppresses the PICTURE; it does not suppress the REQUEST.

   What suppresses the request is in index.html: both <source>
   elements carry (min-width: 61rem), and a <source> whose media
   query does not match is never selected. Below 61rem the <video>
   has no playable source and asks the network for nothing. THE TWO
   MUST STAY ON THE SAME WIDTH or one of them is lying — if you move
   this rule, move the two <source> media attributes with it, and
   re-measure that a 390px viewport still requests no video bytes.

   61rem adds no breakpoint: it is the phone regime's own boundary,
   already used by the hero reorder, the phone art, the full-width
   CTA and the hero grid. */
@media (max-width: 60.9375rem) {
  .preview { display: none; }
}

/* ===================================================================
   THE 976–1103 BAND — THE ONE-COLUMN RULE THAT USED TO LIVE HERE IS
   GONE, 9 Oct 2026. The desktop hero now starts at 61rem, not 69rem.
   ===================================================================

   GUILLAUME, WITH TWO SCREENSHOTS TAKEN WHILE SHRINKING HIS WINDOW:
   "there is also three stats.. it should either switch to mobile or
   stay desktop view and I gradually shrink chrome's browser window".
   THERE MUST BE TWO TREATMENTS, NOT THREE. The third one existed
   only in a 128px-wide window and was neither of the real two.

   WHAT THE BAND DID BEFORE, measured on dcb01ab:

     vw     treatment     art w x h    video card   hero height
     1024   PHONE img     1024x620      ** NO **       1234
     1103   PHONE img     1088x620      ** NO **       1234
     1104   desktop       980x408         yes           647

   One pixel of window took the hero from 1234px to 647px and made
   the video card appear. In the band the copy sat in the left 45%
   with flat hero ground beside it, no card, and the street art below as
   a photo taller than the text block above it. Two unrelated things
   stacked, not a composition.

   WHY DESKTOP AND NOT PHONE. Three reasons, in order of weight.

   1. HomePhone.dc.html DELIBERATELY HAS NO VIDEO CARD. Guillaume,
      5 Oct: "not sure what it brings", and the board's own note ends
      "Desktop keeps it beside the hero." So extending the PHONE
      treatment up would have to keep the card hidden, which leaves
      the card missing across a band of desktop-width windows — the
      content regression, not just a layout one. The phone board is
      not evidence for what a 1024px window should show.

   2. 976–1103px IS A DESKTOP WIDTH. It is a laptop, or a maximised
      window on a small screen, or a desktop user who has zoomed in
      one step. 993 and 1024 are both in measure-layout.js's width
      list for exactly that reason.

   3. THE TWO COLUMNS FIT. Measured at 976 after this change: copy
      column 418px, card 440px, gap 40px, gutters 2x39 — no overflow
      at any width, and the headline reflows from three lines at
      540px to three lines at 418px. 418px is a comfortable measure
      for a 44px headline; it is reflow, not cramping.

   WHAT THIS BLOCK USED TO SAY, because the reasoning still holds for
   its own case and should not be lost: it set one column here on the
   argument that "the preview is gone at these widths, so a second
   track would reserve 380px plus a 32px gap for nothing". That was
   correct GIVEN the card was hidden. The card is no longer hidden
   here, so the premise is gone and the block with it. The desktop
   two-column rule above now covers 61rem upward. */

.hero-copy { min-width: 0; }

/* HEADLINE: 60px / 1.02 / 700 with a 540px measure at desktop, 38px /
   1.08 at 360px — layout.md's exact numbers at both ends. clamp()
   between them so the awkward widths are designed too rather than
   snapping at a breakpoint.

   The line-height is the tighter 1.02 only once the type is big
   enough to carry it; below the desktop breakpoint it relaxes to
   1.08, which is what layout.md asks for at 360px and what a 38px
   headline needs to stay readable.

   The mockup's tiny text is not a font-size spec. Every size here is
   rem- or clamp-based with nothing px-locked, so the type SCALES at
   200% — but scaling is only half of what 200% needs, and the
   overflow-wrap note on the rule below is the other half. */
.hero-h1 {
  /* 20px below the headline, canvas v153 — the board's own
     margin-bottom on this element. It was 1rem (16px), which came
     from .hero-inner's grid gap rather than from the board; the pair
     of gaps around the paragraph are now both the board's, 20 above
     and 30 below. See .hero-lede. */
  margin: 0 0 20px;
  font-size: clamp(2.375rem, 1.55rem + 3.6vw, 3.75rem);
  line-height: 1.08;
  font-weight: 700;
  letter-spacing: -0.03em;
  /* 540px is layout.md's specified measure and it governs where the
     headline wraps. It is deliberately NOT narrowed to the longest
     line: width:fit-content was tried and does nothing here, because
     the lead span wraps to two lines and its fit-content width
     resolves back to the 540px cap that caused the wrap. Narrowing
     the cap itself would rebreak the headline, which is a design
     decision and not a layout tidy-up — see the note on the preview
     card's shift for what the 540px box does and does not overlap. */
  max-width: 540px;
  text-wrap: balance;
  /* WCAG 1.4.10 REFLOW. Without this the page scrolled sideways by
     33px at 200% text on a 360px phone, and it had done since before
     the portal work.

     Measured at Page.setFontSizes standard:32, viewport 360: the h1
     renders at 76px, its column is 280px, and the single word
     "insurance" is 353px — 73px wider than the box it must fit. With
     the initial `overflow-wrap: normal` a word that cannot fit does
     not break; it juts out, and after the hero gutter that is 33px
     past the viewport edge.

     Both properties act ONLY when a word cannot fit on a line of its
     own, so at every width this project tests the headline breaks
     exactly where it always did.

     HYPHENS FIRST, break-word AS THE BACKSTOP — and be honest about
     what each one is doing here.

     break-word is what actually closes the overflow, and on its own
     it renders "insuran / ce": compliant and ugly. `hyphens: auto`
     is there to turn that into "insur- / ance" where the engine can,
     and the arithmetic says it fits — at 76px "insurance" is 353px
     against a 280px box, but "insur-" is 206px.

     IT DOES NOT FIRE IN THE HEADLESS CHROME THIS PROJECT MEASURES
     WITH. Probed directly at 76px/700 with lang en, en-GB and unset:
     scrollWidth 374 in all three, so the build has no hyphenation
     dictionary it will use at this size and weight. The screenshots
     therefore still show "insuran / ce".

     It is kept because real browsers on real devices do hyphenate,
     <html lang="en-GB"> is already there to drive it, and it costs
     nothing when it is declined. Treat the shipped look at 200% as
     the break-word one until someone confirms otherwise on a real
     device — do not delete break-word on the strength of the hyphens
     line, or the overflow comes back.

     The claim above that rem- and clamp-based sizing "supports 200%
     zoom" was half true: it makes the TYPE scale, which is necessary,
     but scaling type is what creates this overflow rather than what
     prevents it. Reflow needs the text to be allowed to break too. */
  hyphens: auto;
  overflow-wrap: break-word;
}
@media (min-width: 69rem) {
  /* 1.02 once the headline is at its 60px desktop size. */
  .hero-h1 { line-height: 1.02; }
}
/* The two tones are two spans on one line. Block display keeps the
   board's line break without a <br>, which a screen reader would not
   pause at differently anyway — the heading reads as one sentence. */
.h1-lead,
.h1-accent { display: block; }
.h1-lead   { color: var(--hero-text); }
/* THE ACCENT IS THE BOARD'S OWN TREATMENT, AND IT IS ONE RULE.

   HomeV2.dc.html and HomeV2Dark.dc.html both style this span
   identically:

     .accent { color: var(--accent-text);
               text-decoration: underline;
               text-decoration-color: var(--mark);
               text-decoration-thickness: .12em;
               text-underline-offset: .14em }

   and the two themes differ only in what the TOKENS resolve to.
   That is the clever part and it is worth spelling out, because it
   reads as two designs and is one:

     light   --accent-text navy #142B49, --mark orange #FF9E52
             -> navy words with an orange rule under them
     dark    --accent-text orange #FFAA70, --mark TRANSPARENT
             -> orange words, and the underline simply disappears

   So "navy with an orange underline" and "orange with no underline"
   are the same declaration seen twice. Nothing is conditional.

   IT REPLACES A TWO-TONE SPLIT THAT WAS NEVER ACCESSIBLE IN LIGHT.
   The v1 page drew this span in a solid accent colour on a dark
   hero. On the v2 LIGHT hero that forced the accent down to a deep
   burnt orange to clear 4.5:1 on pale sky -- #FFAA66 is 1.59:1
   there, #C8682A 3.25:1, #B4541A 4.21:1, and only #7A3606 passed.
   The result was legible and read as brown, not as the brand's
   orange. The board's answer avoids the problem rather than solving
   it: the WORDS carry the ink colour and the ORANGE carries the
   emphasis, so the orange never has to be legible as text at all.

   MEASURED:
     light  --accent-text #142B49 on --hero-bg     12.08:1
            --accent-text on --sky (ramp bottom)    8.11:1
     dark   --accent-text #FFAA70 on --hero-bg      7.64:1
            --accent-text on --sky                  8.73:1

   --mark IS A DECORATION AND 1.4.11 DOES NOT BIND ON IT. The rule
   under the words adds emphasis to text that is already at 12:1; it
   carries nothing the words do not. It is also why it can be
   `transparent` in dark mode without anything being lost.

   --headline-accent IS NOW UNUSED BY THIS RULE and is kept for the
   brand mark and the mission page's tagline, which are text and do
   need a legible accent. */
.h1-accent {
  color: var(--accent-text);
  text-decoration: underline;
  text-decoration-color: var(--mark);
  text-decoration-thickness: 0.12em;
  text-underline-offset: 0.14em;
}

/* ===================================================================
   THE "COMING 2027" PILL — canvas v153, 10 Oct 2026
   ===================================================================

   Home.dc.html (16px type, 10px dot) and HomePhone.dc.html (15px
   type, 10px dot). It sits above the h1 and carries the pre-launch
   message now that the dark banner has gone from this page.

   IT IS A LABEL, NOT A LINK, SINCE TODAY. It was a solid peach <a>
   reading "Coming 2027 | Register now" at 20px/800 with a peach halo
   and a drop shadow — the loudest element in the hero. The pill stops
   competing with the hero's real actions, so: the
   "Register now" half, the hairline, the halo, the shadow, the hover
   and the focus ring all go with the <a>, and what is left is a quiet
   outlined chip stating a fact.

   THE BOARD'S OWN VALUES, verbatim from the v153 markup (identical
   in v152, which carried the same pill):
     background  rgba(255,203,154,0.16)
     border      1px solid rgba(255,203,154,0.45)
     color       #FFCB9A
     padding     8px 16px 8px 12px, radius 999px, 16px/700
     dot         10px, #FF6B35, the same 1.8s pulse

   V2 TAKES THEM FROM TOKENS INSTEAD, AND THE BOARD'S LITERALS ARE
   THE DARK VALUES. The pill above was three hardcoded peach values
   that were right on a hero which was dark in BOTH themes. The v2
   light hero is PALE SKY: #FFCB9A text on #DCEFFA measures 1.35:1,
   so the pill would have been a ghost. --pill-bg/--pill-line/
   --pill-ink are the palette's own per-theme set for exactly this
   component, and their DARK values are the board's peach — so dark
   mode renders what the dark boards draw and only the light theme
   is new.

   COLOURS, MEASURED ON THE RENDERED PILL IN BOTH THEMES:
     light  --pill-ink #7A3606 on --pill-bg #FFE6D1   7.42:1 (text)
            --pill-ink on the bare hero                7.54:1
            --pill-line against the fill               1.26:1 (DECORATIVE)
            the dot against the fill                   2.36:1 (DECORATIVE)
     dark   --pill-ink #FFCB9A on the composited fill  8.50:1 (text)
            --pill-ink on the bare hero                9.69:1

   THE BORDER AND THE DOT BOTH FAIL 3:1 AND ARE BOTH ALLOWED TO, for
   the same reason and it is not "the board says so". WCAG 1.4.11
   applies to what CONVEYS INFORMATION and to the boundaries of
   CONTROLS. This is no longer a control — that is the change made
   today — and neither mark carries anything the words do not: the
   border encloses text that reads identically without it, and the dot
   is a liveness flourish beside "Aiming for 2027", which says the
   same thing in words at 5.11:1. Both are aria-hidden or non-semantic in
   the HTML. If the pill ever becomes a link again, the border becomes
   a control boundary and has to reach 3:1.

   THE PULSE IS AN EXPANDING box-shadow, not a transform or an opacity
   fade. A ring that grows out of the dot and disappears costs no
   layout (box-shadow is painted, never reflowed) and cannot nudge the
   text beside it, which a scaling transform on an inline-flex child
   can. 1.8s, as the board's keyframes say.

   OFF UNDER prefers-reduced-motion, which is the board's rule too and
   is not negotiable: an indefinitely repeating animation is exactly
   what that preference exists to stop. There is no slower version
   offered — the dot simply rests. */
.soon-row { margin: 0 0 1.375rem; }   /* the board's 22px */
.soon-pill {
  display: inline-flex;
  align-items: center;
  gap: 0.625rem;                      /* the board's 10px */
  padding: 0.5rem 1rem 0.5rem 0.75rem;/* the board's 8px 16px 8px 12px */
  border-radius: 999px;
  background: var(--pill-bg);
  border: 1px solid var(--pill-line);
  color: var(--pill-ink);
  font-size: 1rem;                    /* the board's 16px */
  font-weight: 700;
  line-height: 1.2;
  /* IT MUST WRAP AT LARGE TEXT, AND THIS WAS MEASURED THE HARD WAY ON
     THE PILL THIS REPLACES. Without these three lines the pill is one
     unbreakable row: at a 360px viewport with a 32px base font the
     old one came out 433px wide and the page scrolled sideways by
     113px. measure-layout.js caught it on the galaxy-s24 text200
     pass, which is exactly the case that makes an inline-flex row of
     fixed-gap children overflow — the text doubles, the padding is
     rem so it doubles too, and nothing is allowed to break.

     This pill is shorter and lighter, so it is further from the edge
     than the old one was, but the mechanism is unchanged and so is
     the guard. max-width stops the box exceeding the column;
     flex-wrap lets the label drop under the dot if it ever has to;
     and border-radius falls back to a gentler corner once it is two
     lines, because a 999px radius on a tall box bulges into a
     lozenge. At normal sizes a one-line pill's height makes 999px and
     1.25rem identical, so the fallback is invisible until it is
     needed. */
  flex-wrap: wrap;
  max-width: 100%;
  border-radius: 1.25rem;
}
.soon-dot {
  flex: none;
  width: 0.625rem;                    /* the board's 10px */
  height: 0.625rem;
  border-radius: 50%;
  /* The dot keeps a literal and keeps it in BOTH themes. It is the
     one mark here that is pure liveness -- it carries nothing the
     words beside it do not say -- and #FF6B35 is a saturated orange
     that reads against both the warm light fill and the dark one.
     2.36:1 light, 4.42:1 dark, and 1.4.11 does not bind on it for
     the reason set out above. */
  background: #FF6B35;
  animation: soon-pulse 1.8s ease-out infinite;
}
@keyframes soon-pulse {
  0%   { box-shadow: 0 0 0 0    rgba(255, 107, 53, 0.7); }
  70%  { box-shadow: 0 0 0 10px rgba(255, 107, 53, 0); }
  100% { box-shadow: 0 0 0 0    rgba(255, 107, 53, 0); }
}
@media (prefers-reduced-motion: reduce) {
  .soon-dot { animation: none; }
}
/* FORCED COLORS. The pill is built entirely out of a translucent
   background and a translucent border, and forced-colors drops
   backgrounds and flattens colours — leaving "Aiming for 2027" as
   bare text with no chip around it. That is not an operability failure
   (it is not a control and the words still read), but the pill is a
   deliberate visual unit carrying the page's pre-launch disclosure,
   so it keeps its edge the same way every other bordered thing on
   this page does. */
@media (forced-colors: active) {
  .soon-pill { border: 1px solid CanvasText; }
}
/* HomePhone.dc.html runs the pill a size smaller: 15px type and
   tighter row spacing, same 10px dot. Same 61rem the rest of the
   phone hero uses — NOT a new breakpoint. */
@media (max-width: 60.9375rem) {
  .soon-row { margin-bottom: 1rem; }  /* the board's 16px */
  .soon-pill { font-size: 0.9375rem; }/* the board's 15px */
}

/* Body 18px / 27px, 470px measure — layout.md, both at desktop and at
   360px. 9.53:1 on the hero's ground. */
.hero-lede {
  /* 30px BELOW THE PARAGRAPH, canvas v153 — the board sets
     `margin-bottom: 30px` on this element and Guillaume's spec names
     the figure: "paragraph (30px below on desktop)".

     IT IS THE GAP UNDER THE PARAGRAPH, NOT UNDER THE HEADLINE, and
     that was read off the board rather than inferred from the
     sentence. The board's own values in source order are: h1
     margin-bottom 20px, this paragraph margin-bottom 30px, then the
     bar. So the headline-to-paragraph gap is 20px and the
     paragraph-to-bar gap is 30px. Measured on the built page before
     this rule existed, the second gap was 16px, inherited from
     .hero-inner's grid `gap` — which is not a number anybody chose
     for this pair.

     margin: 0 on top is kept deliberately: the gap above comes from
     .hero-h1's own margin-bottom, so the two are never added
     together. */
  margin: 0 0 30px;
  font-size: 1.125rem;
  line-height: 1.5;
  color: var(--hero-text);
  /* 400px — THE SAME MEASURE AS THE EMAIL BAR AND THE CTA.
     ---------------------------------------------------------------
     Changed 12 Oct 2026. Guillaume, with a 1913px screenshot: "text
     seems to be overlapping with the graphics of the asset.. I don't
     think its like that in the board". He was right. The board caps
     this paragraph at `max-width: 30ch` (Home.dc.html, canvas
     v1791835703-df12) — 343px rendered, four lines, stopping 24px
     clear of the art box. The page ran 470px and its last words sat
     on the blossom tree: measured at 1280, 1440, 1680 and 1913, the
     right edge landed 51px past the point where the art reaches full
     opacity, at every width. "every" ending line 1 and "month."
     ending line 2 were both on the picture.

     470px WAS NOT BEING OVERRIDDEN, which is worth recording because
     it was the first suspicion. The computed value was 470px and the
     element measured 470px; the 640px figure that prompted the
     search was .hero-copy, the PARENT. One rule matches this element
     and it is this one.

     WHY 400px AND NOT THE BOARD'S 30ch (309px at this page's 18px).
     The board-faithful fix was built first and Guillaume rejected
     it: "you could leave the text as is but make the graphics
     smaller", then "it should barely overlap with the bars". The art
     shrank from 980px to 880px to meet that (see .hero-art-layer),
     and the text keeps a comfortable measure rather than coming all
     the way in to 309px.

     400px IS THE BARS' OWN WIDTH — the email form and the CTA are
     both 400px — so the paragraph, the bar and the button now share
     one right edge at copy left + 400. That is a stronger reason
     than any contrast figure: it is the alignment the eye reads down
     the column, and it is why this is a px literal matching a px
     literal rather than a ch measure that would drift away from
     them. If the bars change width this should change with them.

     It is ~37 characters a line at 18px, inside the 45-75 band for
     body text and close to the four-line wrap the board draws. */
  max-width: 400px;
}

/* "or get launch updates" AND ITS .hero-alt / .hero-alt-link RULES
   WERE BUILT AND THEN REMOVED THE SAME EVENING, 10 Oct 2026.
   ------------------------------------------------------------------
   They belonged to Option A (canvas v152), where the hero had ONE
   action — the quote button — and this quiet underlined link jumped
   to a separate sign-up band below the hero. Guillaume replaced
   Option A with Option B (v153) before it shipped: the waitlist bar
   is back in the hero column, so there is no band to link to and
   nothing for the link to say.

   The markup is gone from index.html and these rules are deleted
   rather than left behind. Nothing on the page used them; checked
   with a tree grep after removing the markup. */

/* THE SMALL PRINT MOVED TO QUOTE STEP 1, 10 Oct 2026 (item 4 of
   Guillaume's spec, identical in Options A and B), and .hero-clarify
   went with it.
   ------------------------------------------------------------------
   It styled "Today's estimate uses your age, postcode and car (and
   licence years, for newer drivers). Driving analysis comes later."
   That sentence is now the last line of the advisory-quote infonote
   in quote_about_you.html, where it describes something the reader is
   about to do. It was a MOVE, not a deletion.

   The rule is deleted rather than left to rot: an unused rule for an
   element that no longer exists is the stale-claim failure this
   project keeps paying for. Nothing else used the class — checked
   with a tree grep, it appeared in this file and index.html only. */

/* THE AVAILABILITY DISCLAIMER WAS REMOVED FROM THE HERO, 10 Oct 2026.
   ------------------------------------------------------------------
   Guillaume, looking at the rendered hero: "spacing is a bit tight..
   remove 'Illustrative estimate only—not an offer of insurance...'".
   The <strong class="hero-disclaim"> it styled is gone from
   index.html, so this rule is deleted rather than left to rot — an
   unused rule for an element that no longer exists is the stale-claim
   failure this project keeps paying for.

   This reverses gate item 4. It does NOT touch the C1/D5 startup
   guard, which keys on HOLDING_PAGE_ONLY and the flow flags, not on
   template text. See the comment at the removal site in index.html. */

/* ===================================================================
   THE PHONE HERO ART — the car, below the hint, phone only
   ===================================================================

   Guillaume, 4 Oct 2026, on his own phone: "the car is missing on my
   phone". Built to the HomePhone board, which has drawn this block
   since 3 Oct.

   v6, 8 Oct 2026: hero-phone-v6.webp, 1024x768, 90 KB, RGBA with a
   TRANSPARENT SKY — the same idea as the desktop art and for the same
   reason. SEPARATE ARTWORK DRAWN WHOLE, not a crop of the desktop
   file: that was established 4 Oct and still holds. The desktop art's
   left side is composed to sit behind a desktop headline, so at 390px
   it reads as a block beside a tiny car.

   WHAT THE BOARD DRAWS (HomePhone.dc.html, live canvas
   1791493496-f90c, read this session):

     display: block; width: calc(100% + 40px); height: auto;
     margin: 0 -20px -32px;

   THERE IS NO MASK, AND ITS ABSENCE IS THE POINT. Every earlier
   version of this rule ended in a `to bottom` ramp — 30% for the 4:3
   blended art, 22% for the 2.40:1 v4 — because those files were
   opaque and their top edge met the hero's ground as a visible line.
   This file's sky is cut out, so there is no line to hide and the
   board draws no mask. Verified on the shipped file:

     1024x768, RGBA
     fully transparent     33.1% of sampled pixels
     alpha at top-centre   0
     centre column opaque from y=311, i.e. 40% down

   ⚠️ DO NOT FLATTEN OR RE-ENCODE IT, for the same reason as the
   desktop art: flattening paints the sky back as a rectangle and
   undoes the design.

   THE TOP MARGIN IS 0, NOT 8px. The board changed it along with the
   mask, and the two go together: 8px of space made sense under a
   rectangle whose top was being faded out, and with a silhouette the
   art can start immediately. The bottom -32px is unchanged.

   THE RATIO IS 4:3 AGAIN, which closes a question raised on the v4
   art. That file was 1200x500 (2.40:1) and rendered a 162.5px band at
   390px where the 4:3 art renders 292.5px — flagged at the time as
   possibly desktop-ratio art landing in the phone slot. Guillaume has
   supplied separately composed 4:3 art, so it was.

   FULL-BLEED BY NEGATIVE INLINE MARGIN, AND IT MUST BE THE SAME
   EXPRESSION AS THE GUTTER, NOT -20px. .hero-inner's padding-inline is
   clamp(1.25rem, 4vw, 2.5rem). At the board's 390px that resolves to
   the 1.25rem floor (4vw = 15.6px is below it), so the board's literal
   -20px is right THERE and wrong everywhere else: 4vw overtakes the
   floor at 500px, so by 768 the gutter is 30.7px and a fixed -20px
   leaves 10.7px of purple down each side. Measured before this was
   first changed — at 1088 the block sat at x=20 with width 636 inside
   a 1088 viewport, i.e. not bleeding at all.

   The board is drawn at 390 and cannot say what should happen at 900,
   but a block whose whole purpose is to reach both edges must not
   quietly stop reaching them. So the margin repeats the clamp: it
   cancels whatever the gutter actually is, at every width, and stays
   correct if the gutter is ever re-tuned. The board's `width:
   calc(100% + 40px)` says the same thing in its own coordinate
   system — 100% plus the two 20px gutters — so the negative margin
   alone expresses it here and the width stays 100% of the bled box.

   NOTHING MOVES AS IT DECODES. With `height: auto` the <img>'s own
   width and height attributes (1024x768) give it an intrinsic aspect
   ratio, so the box is reserved from the ratio before the bytes
   arrive. Those attributes must track the file.

   NOT .hero-art-layer AT A WIDER BREAKPOINT. That layer is absolutely
   positioned behind the headline, and its display:none at 69rem is
   correct for the reason its own comment gives — art under text at
   narrow widths is what the contrast measurement rules out. This is a
   different element: normal flow, below the hint, nothing laid over
   it. No text, no contrast question.

   SCOPED TO BELOW 61rem, WHICH IS THE PHONE REGIME, NOT BELOW 69rem.
   There are two layouts under 69rem and they hold this element
   differently:

     < 61rem   .hero-inner is display:contents and .hero-copy carries
               the gutter as its own padding. This block is a child of
               .hero-copy, so cancelling that padding bleeds it to both
               edges. This is the phone, and it is what the board draws.

     61-69rem  .hero-inner is a real grid with ordinary gutters and
               .hero-copy is a grid child rather than the padded box
               it is on phone, so cancelling padding it does not have
               would not bleed the art — it would misplace it.

   An earlier version of this rule had no upper bound but 69rem and was
   measured at 1088 showing a half-width block jammed in the left
   column. 61rem is NOT a new breakpoint: it is the one the phone
   reorder above already uses, and it is the exact width at which the
   hero stops being a single column.

   OFF BY DEFAULT, ON ONLY INSIDE A QUERY — the same shape as
   .hero-art-layer. The two are never both visible at any width.

   THIS IS AGAIN THE ONLY BLOCK THAT TURNS IT ON, as of 9 Oct 2026.
   A second block briefly showed this same file in the 61–69rem band;
   it was removed the same day when that band became the desktop
   treatment instead. The two art treatments now meet exactly at
   61rem: .hero-phone-art below it, .hero-art-layer at and above it,
   no gap and no overlap. See the note where that block used to be,
   below this one. */
.hero-phone-art { display: none; }

@media (max-width: 60.9375rem) {
  .hero-phone-art {
    display: block;
    position: relative;
    overflow: hidden;
    /* Board: margin 0 -20px -32px, measured at 390 where the gutter is
       20px. The inline value repeats .hero-inner's own clamp so it
       cancels the gutter at EVERY width, not only where the clamp is on
       its floor — see the note above. The -2rem bottom pulls the next
       section up under the art, which is what makes it read as bleeding
       into the page rather than sitting in a slot. The top is 0: the
       board removed the old 8px when the sky became transparent. */
    margin-block: 0 -2rem;
    margin-inline: calc(-1 * clamp(1.25rem, 4vw, 2.5rem));
  }

  /* THE SELECTOR IS :not(.art-night), AND THAT IS NOT TIDINESS.
     This rule used to be a bare `.hero-phone-art img { display:
     block }`, which was correct while there was exactly one image
     here. V2 puts TWO in the box, one per theme, and
     `.hero-phone-art img` (0,1,1) outranks `.art-night` (0,1,0) --
     so the bare version would have un-hidden the night art and
     rendered both pictures stacked, in both themes, and downloaded
     both files. Excluding the hidden one from the rule that sets
     display is the smallest fix that cannot be undone by accident;
     an !important on .art-night would have worked and would have
     left a trap for the next person. */
  .hero-phone-art img:not(.art-night) {
    display: block;
  }
  :root[data-theme="dark"] .hero-phone-art img.art-night {
    display: block;
  }
  /* The no-JS dark case. :root:not([data-theme]) applies only when
     theme-init.js has NOT run, so it cannot fight the attribute. */
  @media (prefers-color-scheme: dark) {
    :root:not([data-theme]) .hero-phone-art img.art-night {
      display: block;
    }
  }
  .hero-phone-art img {
    /* WHOLE, NOT CROPPED. width 100% of the already-bled box and
       height auto, which is the board exactly. No object-fit: there
       is no box to fit the image into.

       THE MASK CAME BACK IN V2, and the comment that stood here --
       "NO MASK -- the sky is transparent, so there is no top edge to
       fade" -- was true of hero-phone-v6.webp and is false of the
       file this now loads. Measured on the shipped v2 bytes,
       day/hero-home-phone.webp is 0.0% transparent with a painted
       sky, where v6 was 33.1% transparent. The fade is applied below,
       beside the desktop art's, for the same reason and at the same
       stop. */
    width: 100%;
    height: auto;
    -webkit-mask-image: linear-gradient(to bottom, transparent 0, #000 22%);
            mask-image: linear-gradient(to bottom, transparent 0, #000 22%);
    -webkit-mask-repeat: no-repeat;
            mask-repeat: no-repeat;
    -webkit-mask-size: 100% 100%;
            mask-size: 100% 100%;
  }
}   /* end @media (max-width: 60.9375rem) — the phone hero art */

/* THE PHONE ART'S HEIGHT IS CAPPED ON WIDE PHONE-REGIME WINDOWS.
   Added 9 Oct 2026, with the band fix, because it is the same fault
   one regime lower down.

   `height: auto` is the board exactly and is right where the board
   speaks — at 390px the art renders 390x293, about 32% of a phone
   screen, a band under the copy. The file is 4:3 and nothing bounds
   it, so as the window widens inside the phone regime the band
   becomes the page. Measured on dcb01ab:

     vw     art w x h      art height as % of a 900px window
     390     390x293                  33%
     768     768x576                  64%
     896     896x672                  75%
     960     960x720                  80%
     975     975x731                  81%

   A 731px-tall image above a 900px viewport is the artwork alone
   filling four fifths of the screen, with the copy pushed off it.

   620px IS THE SAME CAP THE 61-69rem BAND BLOCK USED before it was
   removed, and it was chosen by measurement there: on
   hero-phone-v6.webp the top 106 source px are fully transparent,
   the 5th-percentile skyline is 112 and the median is 290, so a
   620px cap with the crop taken off the TOP trims the tallest
   spires and leaves the median rooftop, the street, the car and the
   road untouched. A 440px cap was modelled and rejected there: it
   crops past the median and beheads the terrace. The measurement
   outlives the block it was made in, so the cap is kept here.

   FROM 48rem UP, NOT FROM 390. Below 768px the art is already 64%
   or less and `height: auto` is the board's own treatment; capping
   there would crop a phone for no reason. 48rem is an existing
   breakpoint (the hero script and swoosh are hidden below it), so
   this adds none.

   IT IS A CEILING, NOT A HEIGHT, AND THAT MATTERS AT 768 EXACTLY.
   Written first as `height: 620px` and measured: at 767px the art
   was 575px tall and at 768px it was 620px, so widening the window
   by ONE PIXEL made the artwork TALLER and the hero 111px longer.
   An image that grows as its container grows is the fault this cap
   exists to stop, so reintroducing it at the cap's own edge is not
   acceptable. `max-height` leaves `height: auto` in charge until the
   natural height actually reaches 620px — measured, that is at
   827px of viewport — and only bites above it. Below that width
   nothing changes at all.

   object-fit: cover with object-position bottom, so the crop comes
   off the TOP (sky) once the ceiling does bite. The <img> keeps its
   own intrinsic box from the 1024x768 attributes, so the height is
   still reserved from the ratio before decode and nothing shifts as
   the bytes arrive. */
@media (min-width: 48rem) and (max-width: 60.9375rem) {
  .hero-phone-art img {
    max-height: min(620px, 72vh);
    object-fit: cover;
    object-position: 50% 100%;
  }
}

/* THE 61-69rem PHONE-ART BAND BLOCK IS GONE, 9 Oct 2026 (same day
   it was added). IT WAS THE RIGHT FIX FOR THE WRONG FAULT.

   WHAT IT WAS FOR. .hero-art-layer was display:none below 69rem and
   .hero-phone-art display:none above 61rem, so from 976 to 1103px the
   hero had NEITHER artwork. That gap was real, and this block closed
   it by showing the PHONE image, height-capped to 620px, in that band.

   WHY IT IS REMOVED. It treated a missing IMAGE as the fault. The
   actual fault, which Guillaume found the same day by dragging his
   window, was that the whole band was a THIRD LAYOUT: no video card,
   a 1234px hero against the desktop's 647px, and the copy in the left
   45% of flat hero ground. Filling the empty half with a very large photo
   made that third layout more complete, not less of a third layout.
   His words: "it should either switch to mobile or stay desktop view".

   WHAT REPLACED IT. The band is now the DESKTOP treatment — two
   columns, the video card, and .hero-art-layer, whose bound moved from
   69rem to 61rem to meet .hero-phone-art's. The two art treatments now
   meet exactly at 61rem with no gap and no overlap, which is what this
   block was reaching for.

   THE CONTRAST FINDING THIS BLOCK RECORDED IS STILL TRUE AND STILL
   MATTERS, so it is not lost with the block. It said that lowering
   .hero-art-layer to 61rem put art under the copy and measured the
   hint band at 1.00:1. Re-measured on the dcb01ab composition, which
   is a FIXED 980px layer rather than the viewport-filling one that
   finding came from, the figure is far milder - 13.3% of background
   pixels below 4.5:1 at 976 against 13.6% at the signed-off 1104 -
   but only once the layer's left edge is PINNED through the band. See
   "THE ART'S LEFT EDGE IS PINNED THROUGH THE 976-1103 BAND" above for
   the measurements, including what the figure is without the pin
   (23.5%) and why the mask alone could not fix it. */

/* THE "Drive well, it pays" HERO LETTERING WAS REMOVED, 10 Oct 2026
   (Guillaume, canvas v153): "it stays in the footer".
   ------------------------------------------------------------------
   .hero-script and .hero-swoosh styled two decorative images —
   drive-well-script.svg and orange-swoosh.svg, both aria-hidden with
   empty alt, both hidden below 48rem. The markup is gone from
   index.html, so the rules go with it rather than being left to rot.

   THE CONTRAST FINDING THEY CARRIED IS RECORDED HERE BECAUSE IT IS
   STILL TRUE OF THE ARTWORK, not of the lettering. The swoosh's right
   end sat over the hero art's faded left ramp and measured 4.57:1 at
   1104px — the worst figure anywhere in the hero and still above
   4.5:1. With the lettering gone, the worst remaining hero text
   figure is the pill's 5.11:1. If the art's mask ramp is ever moved
   earlier, the thing to re-measure is now the headline and the lede,
   not this pair.

   NEITHER FILE IS DELETED, AND ONLY ONE OF THEM IS NOW UNUSED.
   orange-swoosh.svg IS STILL SERVED — _chrome.html:257 draws it as
   .scriptmark-swoosh on the quote-flow pages, which is a different
   element with a different job. Checked with a tree grep rather than
   assumed; an earlier draft of this note called it unused and was
   wrong. drive-well-script.svg is the one nothing references now; it
   is kept because it is the asset a reworded version of the tagline
   would be redrawn from (see the note at the removal site in
   index.html), and because deleting a hand-drawn asset to save 1.5 KB
   is not a trade worth making. */

/* THE CTA, with the board's arrow.

   Orange fill with --on-cta dark ink: 8.67:1. White on this orange is
   1.8:1 and must never be used. Against the hero's ground the fill's own
   edge is 5.31:1, well past the 3:1 a control boundary needs, so no
   extra border is required here — unlike on cream, where the same
   orange is 1.78:1 and the border below supplies the edge. */
.cta-arrow {
  display: inline-flex;
  align-items: center;
  gap: 0.625rem;
}
.cta-arrow-icon { flex: none; }
/* The arrow nudges on hover, and not at all for anyone who asked for
   reduced motion (the global rule at the top disables transitions). */
.cta-arrow:hover .cta-arrow-icon { transform: translateX(3px); }
.cta-arrow-icon { transition: transform 120ms ease; }

/* ---- The planned product preview, beside the hero copy -----------

   ZERO OVERLAP with the hero's bottom edge, deliberately. layout.md
   measured the board: the preview's bottom is about y470 and the hero
   ends around y506, so it does not cross. There is no negative margin
   here and adding one would be inventing an effect the board does not
   have.

   440px at desktop / radius 20 / padding 16, 100% at 360px with radius
   16 / padding 12.

   NO VERTICAL OFFSET. There was a margin-top: -24px here, with a
   margin-bottom: 24px compensating, and the comment beside it said it
   lifted the card "as the board shows". IT DID NOT: Home.dc.html lays
   the hero out with `align-items: start` and no offset of any kind, so
   the card's top sits level with the copy column, not 24px above the
   headline.

   The paragraph above this one, in the same block, said "There is no
   negative margin here and adding one would be inventing an effect the
   board does not have." Both sentences cannot be true. Read against
   the live canvas on 3 Oct 2026, the first one is the accurate one and
   the offset was the invention — measured at 1280/1440/1920/2560, the
   card's top was 24px above the h1's top at every width.

   Removed, board leads. */
.preview {
  background: var(--paper);
  color: var(--ink);
  border-radius: 20px;
  padding: 16px;
  margin: 0;
}


/* THE CARD SITS OUTSIDE THE 1200px CONTAINER ON WIDE SCREENS, so the
   car gets a wider slot between the headline and the card.
   ==================================================================

   THE PROBLEM, measured before the change at 1280/1600/1920/2560:

     vw    h1 box ends   car        card       h1box→car  car→card
     1280  620           545-735    760-1200   -75        +25
     1920  940           865-1055   1080-1520  -75        +25
     2560  1160          1185-1375  1400-1840  -75        +25

   The car's visible slot is ~200px and constant, because copy, car
   and card are all inside one centred 1200px box: widening the
   viewport moves all three together and gives the car nothing. The
   h1's BOX overruns the car's left edge by 75px at every width (the
   longest glyph run ends 26px clear of it, so nothing is visually
   struck through, but the box is the layout's own claim about where
   the copy ends and it should not overlap).

   THE FIX: push the card right by up to 120px, into the gutter
   outside the container. The copy column keeps its full width, the
   grid track keeps its 440px, and everything the card gains the car
   gains too.

   THE OFFSET IS CAPPED AND THAT CAP IS THE WHOLE POINT. It is
   min(120px, half the space outside the container) — so it is 0 in
   the 1104–1200px band where there is no gutter to move into, ramps
   up as the gutter opens, and STOPS at 120px however wide the
   viewport gets. An uncapped offset is the bug that was fixed in
   e6fcb93: a length defined against the viewport while the content
   box is a fixed 1200px means the two drift apart forever, and at
   2560 the card would be halfway to the screen edge.

   IT CANNOT WIDEN THE PAGE. translate does not affect layout size,
   and the offset never exceeds half the free space outside the
   container, so the card's right edge stays inside the viewport by
   construction. Verified at 320..2560: zero horizontal overflow.

   Below 69rem this does not apply at all: the hero is one column,
   the card is full width and the artwork layer is display:none. */
/* REMOVED 3 Oct 2026, because the problem it fixed is gone.

   The table above is a real measurement and the fix was sound WHEN
   THE ART WAS 1048px WIDE: copy, car and card all sat inside one
   centred 1200px box, the car had a constant ~200px slot, and the
   h1's box overran the car's left edge by 75px at every width.

   THE WIDE ART CHANGED THE PREMISE. hero-purple-wide.webp is 2600px
   drawn at 1630px, so the scene continues past the container instead
   of ending inside it, and the car is no longer squeezed between the
   copy and the card. Re-measured on the live page after the swap:

     vw     h1 box ends   card starts   gap
     1280   612.5         792.5         +180
     1440   692.5         952.5         +260
     1920   932.5         1192.5        +260

   A positive gap at every width, where the old table recorded -75.
   There is nothing left for the shift to buy, and it costs the thing
   the board is explicit about: Home.dc.html puts the card in the
   hero's 440px track inside a 1197px box, and the transform put it
   80px OUTSIDE the container at 1440 and above.

   STILL TRUE AFTER EVERY ART CHANGE SINCE, re-measured rather than
   assumed. The art has been replaced four times since this table was
   written — most recently by hero-wide-v6.webp, drawn 980px wide and
   RIGHT-anchored at -110px — but THIS TABLE DOES NOT DEPEND ON THE
   ART. It measures the h1 against the preview card, and both are laid
   out by the grid, not by the artwork. Re-measured 8 Oct against v6:
   the h1 still ends at 613 at 1280 and the card still starts at 753,
   so the numbers below stand unchanged. Measured on the live page, h1
   box right edge to preview card left edge:

     vw     h1 box ends   card starts   gap
     1104   569           609           +40
     1280   613           753           +140
     1440   693           833           +140
     1920   933           1073          +140
     2560   1253          1393          +140

   Still positive at every width, where the table above recorded -75,
   so the conclusion holds: there is nothing left for the shift to
   buy. The gap is now a constant +140 above 1280 rather than growing,
   because the card's offset is capped at 120px and the h1's measure
   is capped at 540px — both fixed, so their distance stops changing.
   At 1104 it narrows to +40, which is the tightest case and still
   clear.

   Board leads. If the car ever needs a wider slot again, widen the
   art or change the board first — do not reintroduce a transform
   that moves the card out of the layout the board draws. */
/* A transform on a card containing a focusable link would move the
   focus ring with it, which is correct, but it also creates a
   containing block for fixed-position descendants. There are none
   here and there should not be: if one is ever added, it will be
   positioned against this card rather than the viewport. */
@media (max-width: 60.9375rem) {
  .preview {
    border-radius: 16px;
    padding: 12px;
    /* 320px content at 360px: the wrap's 20px gutters already do this,
       so the card is simply full width of its column. */
    width: 100%;
  }
}

.preview-figure { margin: 0; }

/* THE FOOTAGE BOX. Replaced the poster/overlay pair on 29 Sep 2026.

   THE ASPECT RATIO CHANGED AND THE BOX HAD TO BE TOLD. The still was
   1536x1024 — 3:2 — and the old rule let the image's own intrinsic
   ratio set the height (width:100%, height:auto). The video is
   1280x720, which is 16:9, so the same rule would have silently made
   the card SHORTER: at the 407px the card occupies on desktop, 3:2
   gives 271px of picture and 16:9 gives 229px.

   aspect-ratio states it instead of inheriting it, so the box has one
   declared shape and the video fills it. object-fit: cover means a
   future re-cut at a different ratio crops rather than letterboxes —
   a black bar would read as a broken player.

   There is no second layer any more, so the old warning about keeping
   two layers aligned no longer applies: the annotation is burned into
   the footage and cannot drift from it. */
/* IT IS A <button> SINCE 29 SEP 2026. Clicking it OPENED FULLSCREEN
   until 5 Oct 2026; it now turns the browser's own control bar on
   (.is-chromed below). Either way it is a control, and a control has
   to work without a mouse. That means undoing what
   a UA stylesheet gives a button: padding, border, background, the
   inherited-font exception and the text-align. Miss one and the card
   gains a hairline or a 2px inset that nothing explains.

   `display: block` matters as much as the rest: a button is
   inline-block by default, which adds a baseline gap under the video
   and makes the box a few pixels taller than the ratio asks for. */
.driving-preview {
  position: relative;
  display: block;
  width: 100%;
  aspect-ratio: 16 / 9;
  isolation: isolate;
  border-radius: 10px;
  overflow: hidden;
  background: #FFF9F1;
  padding: 0;
  border: 0;
  margin: 0;
  font: inherit;
  color: inherit;
  text-align: inherit;
  cursor: pointer;
  -webkit-appearance: none;
  appearance: none;
}
/* THE FOCUS RING IS NOT OPTIONAL. This is now the card's only
   control, and the only way a keyboard user can tell they have
   reached it — the video looks identical either way.

   :focus-visible, not :focus, so a mouse click does not leave a ring
   behind. The offset puts it outside the rounded corner rather than
   on top of the footage, where a dark frame would vanish into dark
   frames of video. --accent is the same ring the quote flow uses;
   this card does not get its own. */
.driving-preview:focus-visible {
  outline: 3px solid var(--accent);
  outline-offset: 3px;
}
/* THE MOBILE FULL-FILM RULES WERE HERE AND ARE GONE, 5 Oct 2026.

   .driving-preview__full was an empty <video> that the phone's tap
   swapped in for the 207KB teaser loop. Two rules drove it: a
   default `display: none`, and a (max-width: 47.9375rem) block that
   revealed it on .is-full.

   Both went because the ELEMENT went, and the element went because
   the card went: HomePhone.dc.html no longer draws the dashcam card
   (Guillaume: "not sure what it brings"), so .preview is removed
   below 69rem — see the rule in the phone block above. A selector
   matching nothing is not harmful, but it is a claim that a feature
   exists, and this repository's recorded failure is descriptions
   outliving the behaviour they describe.

   flow.js's removal note carries the full account. */

.driving-preview__video,
.driving-preview__stillfallback {
  display: block;
  width: 100%;
  height: 100%;
  object-fit: cover;
  /* The button is the control; the video must not swallow its
     clicks or become a second tab stop inside it. */
  pointer-events: none;
}

/* THE :fullscreen RULES ARE GONE — 5 Oct 2026. Two of them lived
   here, setting object-fit: contain and a black ground so the
   letterbox bars were deliberate rather than a crop. They described a
   state this page can no longer reach: all custom fullscreen was
   removed after Guillaume was trapped in it on his phone, and
   flow.js's top block holds the bug and the measurement.

   They are deleted rather than kept "in case", because a rule for a
   state that cannot occur is how a stylesheet stops describing the
   page — the next reader cannot tell it from a rule that matters.

   IF THE NATIVE BAR'S OWN FULLSCREEN BUTTON IS USED, the browser
   fullscreens the <video> itself and applies its OWN presentation to
   it, which is `contain` on a black ground: exactly what these rules
   were reproducing by hand. Nothing is lost.

   THE HANDOVER TO THAT BAR. flow.js adds .is-chromed on the first tap
   and sets `controls` on the video; from then on the bar owns every
   tap, so the video needs its pointer events back. Inline it has
   pointer-events: none so the button owns the first tap — after that
   tap the button's job is finished. The button keeps its own box so
   nothing moves. */
.driving-preview.is-chromed .driving-preview__video {
  pointer-events: auto;
}
/* The card is no longer a control once the bar is showing, so it must
   not keep offering a pointer. */
.driving-preview.is-chromed { cursor: default; }

/* REDUCED MOTION. A 98-second autoplaying loop is the case this query
   was written for, and honouring it is not optional here.

   The fallback is hidden by default rather than the video being
   hidden by default, so the majority path needs no override. The
   poster carries the same frame, so the card looks intentional
   rather than broken.

   THIS RULE HIDES THE MOTION. IT DOES NOT STOP THE DOWNLOAD, and an
   earlier version of this comment claimed it did. Measured on the
   running page: with reduce set and only `display: none` here,
   Chrome still fetched all 12MB and reported readyState 4.

   Stopping the fetch needs markup, so it lives in index.html — both
   <source> elements carry media="(prefers-reduced-motion:
   no-preference)", and a source whose media query does not match is
   never selected. The two halves are separate on purpose: this one
   guarantees nothing MOVES, that one guarantees nothing is
   DOWNLOADED, and neither does the other's job. */
.driving-preview__stillfallback { display: none; }

@media (prefers-reduced-motion: reduce) {
  .driving-preview__video { display: none; }
  .driving-preview__stillfallback { display: block; }
}

/* The annotation labels, the two tiles and the hedge line were all
   removed on 29 Sep 2026 — Guillaume wanted the card to be the video
   and nothing else. Their rules went with them rather than being left
   as dead selectors:

     .anno-list / .anno / .anno--vehicle / .anno--centreline
     .preview-cards / .preview-card / .preview-card-title
     .preview-card-body / .score-line / .score-figure
     .score-qualifier / .preview-note
     .preview-head / .preview h2 / .preview-sub

   Two of those carried decisions rather than styling, and the
   decisions are recorded in docs/design/preview-decisions.md rather
   than lost with the CSS: the score figure and its qualifier were
   deliberately the same size, weight and colour (a single shared rule
   was what made that structural), and the annotation labels were
   deliberately real text rather than baked into the image. Both are
   moot now that neither element exists, and both would have to be
   re-argued, not just re-added, if the tiles ever come back. */

/* ---- The portal, and the contextual-AI argument -----------------

   Both built 4 Oct 2026 from the design board, replacing the three
   "Feedback you can use" step cards whose rules used to live here.

   THE PORTAL IS A PICTURE OF SOFTWARE and is styled like one: its own
   greys and radii rather than the site's panel tokens, because it is
   meant to read as a screenshot of a different surface. Its text sits
   on white at the sizes an app uses, not the page's reading sizes.

   It is aria-hidden in the markup, so none of this needs to survive a
   screen reader — but it DOES need to survive 200% text and a narrow
   screen, which is why nothing here is a fixed pixel height. */

/* THE BAND IS PINNED LIGHT, like the frame it holds. Guillaume,
   6 Oct 2026, comparing his phone to DevTools: "on the ip address the
   background colour of the panel 'Drive better - Pay less' is dark
   purple, on my desktop ... its light purple (which is correct)".
   His phone is in dark appearance; DevTools emulates light.

   var(--panel) resolves to #F3ECFD in light and #382455 in dark, so
   the band followed the theme while .portal-frame inside it is
   pinned #FFFFFF in both — a white dashboard floating on dark purple,
   which is neither theme's design.

   THAT PREMISE EXPIRED ON 22 OCT 2026 AND THIS RULE CHANGED WITH IT.
   The paragraph that stood here read: "NEITHER BOARD DEFINES A DARK
   THEME: grep prefers-color-scheme over Home.dc.html and
   HomePhone.dc.html returns 0. So there is no board answer here and
   the page was improvising." That was true, and the improvisation it
   licensed -- pin the band light, because it is a PICTURE OF A LIGHT
   APPLICATION -- was a reasonable call with no board to consult.

   THERE IS A BOARD NOW. HomeV2Dark.dc.html and HomeV2PhoneDark.dc.html
   are fixed-theme dark snapshots of the whole page, and they answer
   the question directly. Sampled off the rendered board rather than
   read out of its CSS:

     how-it-works band   #101D30
     contextual band     #0B1626
     drive-better band   #142640
     portal sidebar      #0B1626
     portal cards        #182B43

   The board draws the dashboard itself as a DARK APPLICATION -- not a
   light dashboard floating on a dark band, which is the thing the
   improvisation was avoiding and was right to avoid. So the band and
   the frame both follow the theme now, through the --ui-* tokens,
   which the v2 palette defines in full for both themes precisely
   because a product mock-up needs its own surface set.

   GUILLAUME'S 6 OCT REPORT IS STILL HONOURED, and that is the test
   this change had to pass. He saw "a white dashboard floating on dark
   purple, which is neither theme's design". It still is not: the
   frame and the band move TOGETHER, both to dark, so there is no
   light-on-dark seam in either theme. What changed is which pair they
   move to, not whether they move together.

   THE DESIGN SYSTEM IS NOT SILENT, THOUGH, and the sentence above is
   easy to over-read. Checked against the live canvas 13 Oct 2026,
   version 1791875629-0db4: project/ds/advisoryquote/tokens.json
   carries a FULL dark theme — every token has a light and a dark
   value, declared themes "light" and "dark". What has no dark
   version is the BOARDS, which are drawn at one width in one theme.
   So "the canvas does not say" is true of the picture and false of
   the tokens; look in tokens.json before concluding dark mode is
   unspecified.

   That distinction matters for THIS band specifically: ok-ink's dark
   value #8FE0AD is documented there as "9.10:1 dark", which it is —
   on the dark ok-bg it is paired with, and NOT on the pinned white
   tiles in this band, where it measured 1.56:1 before .portal-frame
   re-pointed --ok-ink. The token was right; the surface it would
   have landed on was not. A dark value in the system is not a
   licence to use it anywhere dark mode is active.

   The line colour is pinned with it; --line goes dark and would draw
   a near-invisible border against the pinned fill. */
.portal {
  background: var(--section-alt);
  border-block: 1px solid var(--line);
  padding-block: 4.5rem;
  /* THE INK PINS ARE GONE, AND THEY HAD TO GO TOGETHER WITH THE FILL.
     This rule used to re-point --ink and --ink-soft at light values,
     because the fill above was pinned light while those tokens
     followed the theme -- cream ink on a pale band measured 1.10:1.
     Now that the fill follows the theme too, the tokens are correct
     on their own and re-pinning them would reintroduce exactly the
     fault they were added to fix, in the opposite direction: dark ink
     on a dark band.

     THE HISTORY IS WORTH KEEPING because the mechanism is a trap.
     The --ink-soft pin was first written into a `.portal { order: 4 }`
     rule INSIDE a max-width media query, by a script that inserted at
     the first `}` after the first `.portal {` it found. It passed on
     a phone and failed above 61rem -- 7.56:1 at 393px, 1.58:1 at
     1280 -- and swallowed `order: 4` into a comment on the way. A pin
     inside a breakpoint is not a pin, and a surface that pins its
     fill must pin EVERY ink token it inherits or none. */
  /* NO COLOUR DECLARATION HERE SINCE V2, AND THAT IS THE WHOLE POINT
     OF THE RULE ABOVE IT.

     This used to be `color: #101D30`, pinned to match a pinned fill.
     Guillaume found the bug it fixes on his phone, 6 Oct 2026:
     "something is not correct with the background and the text.. can
     bearly read it" -- the band's fill was pinned light and --ink was
     not, so "Drive better - Pay less" rendered cream on pale at
     1.10:1 against the 3:1 a 32px heading needs.

     THE FIX AND THE FAULT ARE THE SAME SHAPE AND THAT IS THE LESSON.
     A pinned background without a pinned colour is always this bug;
     so is a THEMED background with a pinned colour, which is what
     leaving this line in would now be. Measured on the running page
     after the fill was moved to var(--section-alt) and before this
     line came out: .portal-h computed rgb(16,29,48) on rgb(20,38,64)
     -- the same heading, the same invisibility, the opposite
     direction. Found by reading computed styles off the rendered
     dark page, which is the only thing that catches it.

     So: the fill follows the theme and the ink follows the theme, and
     the rule is that they move TOGETHER. --ink is 12.76:1 on
     --section-alt in dark and 12.79:1 in light. */
}
/* "Drive better – Pay less" — both boards, 5 Oct 2026.

   THE SIZE IS THE BOARDS' SHARED SECTION-HEADING SIZE: 48px at 1280
   and 32px at 390. Canvas version 1791232647-f669 normalised EVERY
   section heading to one size, with a note saying so: "ALL section
   headings one size (Guillaume): 48px, 800, -0.02em, line-height
   1.1. Hero h1 stays larger."

   An earlier build of this today used 48/31 from version
   ...-4f97, which had this heading at 31px on phone and the
   Contextual AI title at 52px. Both were folded into the one size
   an hour later. --section-h is that size, shared with .ctxai-h.

   Solved so the ends land on the boards' numbers rather than near
   them: at 390 (24.375rem) and 1280 (80rem),
     slope = (3 - 2) / (80 - 24.375) = 0.017978rem per rem = 1.798vw
     intercept = 2 - 0.017978 * 24.375 = 1.5618rem
   Verified in the measurement table in the report. */
:root {
  --section-h: clamp(2rem, 1.5618rem + 1.798vw, 3rem);
}
.portal-h {
  margin: 0 0 0.5rem;
  text-align: center;
  font-size: var(--section-h);
  line-height: 1.1;
  font-weight: 800;
  letter-spacing: -0.02em;
}

/* THE HIGHLIGHTER STROKE behind "Pay less.".

   It is a background GRADIENT with hard colour stops, not a
   border-bottom and not a <mark>. The board draws it from 58% to
   92% of the line box — through the x-height and under the
   baseline, the way a felt-tip crosses a word — where a
   border-bottom would sit below the descenders and read as an
   underline.

   The stops are the board's, copied rather than approximated:
     transparent 0 -> 58%, #FFCB9A 58% -> 92%, transparent 92%.

   ⚠️ THE BAND DESCRIBED ABOVE NO LONGER SHIPS. Read the rule that
   follows this comment, not this comment: the gradient was removed
   on 6 Oct 2026 and the span is `color: var(--brand)` and nothing
   else. The paragraph is kept for the geometry it records, in case
   the stroke is ever wanted back, and its contrast figures are about
   a surface that is not painted.

   WHAT ACTUALLY SHIPS, measured for v2: --brand on the band it sits
   on, which is --section-alt. 12.79:1 light, 12.76:1 dark. The text
   follows --brand in both themes and needs no override.

   white-space: nowrap IS REQUIRED, NOT TIDINESS. Both boards keep
   "Pay less." on one line, and a highlighter broken across two
   lines paints two disconnected stripes with a ragged end — it
   stops reading as one mark. The phone board sets exactly this. The
   words are short enough that it cannot force overflow at 320px;
   measured in the report's overflow rows. */
/* COLOUR ONLY, NO BAND. Guillaume, 6 Oct 2026: "the band made the
   headings hard to read when they wrap on mobile." The live board
   (1791310599-dcfe) agrees — it styles this span `color:
   var(--brand)` and nothing else, and greps 0 for the gradient.

   The band was drawn at 58%–92% of the line box, which is fine on one
   line and wrong on two: a wrapped heading gets a stripe through the
   middle of each line at a height chosen for a single one.

   white-space: nowrap STAYS. It is not part of the band — it is what
   keeps "Pay less" off two lines, which the board also requires. */
.portal-h-mark {
  color: var(--brand);
  white-space: nowrap;
}

.portal-lede {
  margin: 0 auto 2.25rem;
  max-width: 52ch;
  text-align: center;
  font-size: 1.0625rem;
  line-height: 1.5;
  color: var(--ink-soft);
}

/* ===================================================================
   THE SAVING PILL — "▼ £142 → £92 a month | 35% less in a year"
   ===================================================================

   New 5 Oct 2026, REBUILT 9 Oct 2026 from the 1791498167-c644
   boards, which changed its figures, its colours and its phone
   treatment all at once. Both boards draw it between the lede and
   the dashboard frame.

   WHAT WAS DELETED HERE, so nobody looks for it. The previous build
   carried ~90 lines explaining a one-span-two-presentations design
   for an "example driver" attribution: a ::before middot on desktop,
   ::first-letter uppercasing on phone, and a rule that moved the
   PAINTED PILL BOUNDARY in one level at 61rem so the attribution
   could fall out of the flex row onto its own line.

   THE BOARDS DROPPED THE ATTRIBUTION ENTIRELY, so all of that is
   gone rather than left in place describing markup that no longer
   exists. "example driver" appears on neither board. It was never an
   honesty chip — the page's honesty markers are the Contextual AI
   and dashcam chips, which still render — so removing it removed no
   claim the page needed to make.

   THE PHONE NOW STACKS THE PILL'S OWN TWO HALVES, which is how the
   phone board solves the 390px width problem instead. That is a much
   simpler mechanism than the one it replaces: the pill keeps painting
   itself at both widths, and only its flex-direction and corner
   radius change.

   GREEN IS NOW THE SURFACE, NOT THE INK, and it inverted on both
   boards: #17643A background with #FFFFFF text, where it used to be
   #17643A text on #E7F6EC. White on #17643A measures 7.19:1, so it
   passes AA at any size — better than the 6.43:1 the old pale pill
   had. Fixed in both themes, like the highlighter above: the pill is
   its own surface rather than a tint of --panel, so inverting it for
   dark mode would mean inventing a green nobody has specified and
   would stop it matching the board.

   #BFF0CF ON #17643A, the percentage half, measures 5.68:1 — also AA
   at any size. It is the board's own colour and it is checked rather
   than assumed, because a light mint on a mid green is exactly the
   pairing that fails. */
.portal-saving {
  margin: 0 auto 2.25rem;
  text-align: center;
}
.portal-saving-pill {
  display: inline-flex;
  align-items: center;
  gap: 1rem;
  padding: 1rem 1.875rem;
  border-radius: 999px;
  background: #17643A;
  color: #FFFFFF;
  font-weight: 800;
  box-shadow: 0 10px 26px rgba(23, 100, 58, 0.35);
  font-variant-numeric: tabular-nums;
  /* The board is a fixed 1280 canvas; the page has to hold at 200%
     text, where the two halves plus the divider no longer fit one
     line. Allowing a wrap keeps the surface on-screen instead of
     overflowing it (WCAG 1.4.10 reflow). At every normal text size
     this is one line, identical to the board. */
  flex-wrap: wrap;
  justify-content: center;
  max-width: 100%;
}
.portal-saving-arrow {
  font-size: 1.625rem;
  line-height: 1;
}
.portal-saving-fig {
  font-size: 1.625rem;
  letter-spacing: -0.01em;
}
/* THE STRIKETHROUGH IS DECORATION. The accessible reading is supplied
   as visually-hidden text in the markup, because text-decoration is
   invisible to assistive tech. */
.portal-saving-was {
  text-decoration: line-through;
  text-decoration-thickness: 2px;
  opacity: 0.75;
  font-weight: 700;
}
/* The board's divider: a 1px hairline stretched to the pill's inner
   height, not a border on a sibling — a border would hug one half's
   text box and come out short. It is an empty aria-hidden span
   because it is a rule, not content. */
.portal-saving-rule {
  width: 1px;
  align-self: stretch;
  background: rgba(255, 255, 255, 0.35);
}
.portal-saving-pct {
  font-size: 1.625rem;
  color: #BFF0CF;
}
/* FORCED COLORS drops `background`, which would leave white-on-white
   and an invisible divider. The system pair plus a border keeps the
   pill a distinct object, and the divider becomes a real border. */
@media (forced-colors: active) {
  .portal-saving-pill { border: 1px solid CanvasText; }
  .portal-saving-rule { background: CanvasText; forced-color-adjust: none; }
  .portal-saving-pct { color: inherit; }
}

/* PHONE, from HomePhone.dc.html: ONE SOLID GREEN CARD, TWO LINES.

   The desktop pill's two halves sit side by side with a divider
   between them; the phone board stacks them and drops the divider.
   Nothing is hidden and nothing is duplicated — the same two facts
   at both widths.

   THE ORDER REVERSES, AND THAT IS THE BOARD'S OWN EMPHASIS. Desktop
   leads with the price drop and follows with the percentage; the
   phone board leads with "35% less in a year" in #BFF0CF and puts
   "£142 → £92 a month" under it in white. `order` does it, so the
   DOM order — and therefore the reading order for a screen reader
   and for anyone tabbing past it — stays the desktop one at both
   widths. That is deliberate: the figures are the substance and the
   percentage is the summary, so the money reading first is the
   better reading order even where the board draws it second.

   IT IS A CARD, NOT A PILL, BELOW 61rem: border-radius drops from
   999px to 20px, because a 999px radius on a two-line box turns the
   short lines into a lozenge. The board draws 20px.

   Sizes are the phone board's: 16px/18px padding, 26px percentage
   with a 22px arrow, 20px price line 4px below it. */
@media (max-width: 60.9375rem) {
  .portal-saving-pill {
    display: flex;
    flex-direction: column;
    align-items: center;
    gap: 0.25rem;
    padding: 1rem 1.125rem;
    border-radius: 20px;
    box-shadow: 0 10px 24px rgba(23, 100, 58, 0.32);
  }
  /* The percentage leads, and the arrow belongs to it — so the arrow
     joins it on the same line via ::before and the standalone arrow
     element stands down. `order` reorders siblings; it cannot
     reparent, which is why the arrow is redrawn rather than moved.
     Decorative at both widths, so no accessible content moves. */
  .portal-saving-arrow { display: none; }
  .portal-saving-rule { display: none; }
  .portal-saving-pct {
    order: -1;
    font-size: 1.625rem;
    font-weight: 800;
  }
  .portal-saving-pct::before {
    content: "\25BC\00A0";
    font-size: 1.375rem;
    line-height: 1;
  }
  .portal-saving-fig {
    font-size: 1.25rem;
    font-weight: 700;
    letter-spacing: normal;
    /* Allowed to wrap, balanced, for the same reflow reason as the
       desktop pill. The board sets nowrap, which is free on a fixed
       390px canvas at one text size; at 360px with 200% text it
       scrolls the page sideways. Two balanced short lines centred in
       the card is the honest way to hold the board's intent at text
       sizes the board cannot show. */
    text-wrap: balance;
  }
}

/* THE FRAME SPANS THE FULL 1200, not 1200 minus the wrap's gutter.
   On the board the portal's wrapper is `max-width: 1200px; margin: 0
   auto` with NO padding, so the frame is 1200px wide. Inside .wrap it
   was 1104px -- the 48px gutter each side -- and those 96px came
   straight out of the drives table's flexible note column, which fell
   to 55px and wrapped the AI note to twelve lines. The heading and
   lede above it keep the gutter; only the frame breaks out.
   Measured 4 Oct: board note column 140.8px / 4 lines. */
.portal-frame {
  --gut: clamp(1.25rem, 4vw, 3rem);  /* THE SAME EXPRESSION AS .wrap:156.
                                        A constant -48px here would shove
                                        the frame off-screen below 1200,
                                        where the gutter clamps down. */
  width: calc(100% + (2 * var(--gut)));
  margin-inline: calc(-1 * var(--gut));
}
/* The shadow is what sells it as a separate surface. */
/* THE FRAME FOLLOWS THE THEME SINCE V2, THROUGH --ui-*.

   It was `background: #FFFFFF` plus three pinned ink tokens, on the
   reasoning that the mock-up is a PICTURE OF A LIGHT APPLICATION and
   paints its own surfaces in both themes, the way the banner and
   footer stay dark in both. With no dark board to consult that was a
   defensible call.

   HomeV2Dark.dc.html consults differently. Sampled off the rendered
   board: the sidebar is #0B1626 and the cards are #182B43 -- it
   draws the dashboard as a DARK APPLICATION. So the frame takes
   --ui-card / --ui-ink, which the v2 palette defines per theme for
   exactly this: a product mock-up needs its own surface set, kept
   away from the marketing palette, and now has one in both themes.

   THE THREE INK PINS GO WITH THE FILL, and that is the rule this
   surface has been teaching since 6 Oct: a pinned surface needs
   EVERY ink token it inherits pinned with it, or none. They are
   re-pointed at --ui-* rather than deleted, because the descendants
   still inherit and the --ui-* set is what is correct on --ui-card
   in both themes.

   THE HISTORY IS KEPT BECAUSE THE TRAP IS REAL. Before any pin
   existed, .portal-tile-figure and .portal-card-title rendered
   rgb(255,249,241) on rgb(255,255,255) -- about 1.03:1. The tile
   numbers 89 / 42 / 612 and the "Recent drives" and "Premium"
   headings were invisible in dark mode, found in a phone screenshot
   and not by any check. Then --ink alone was pinned and the quiet
   text kept failing at 1.56:1 and 1.38:1 because --ink-soft flips
   independently; then --ok-ink was found the same way, third.

   MEASURED FOR V2, light / dark, on --ui-card:
     --ui-ink   14.27:1 / 12.03:1
     --ui-soft   5.67:1 /  7.06:1
     --ok-ink    6.52:1 /  9.10:1  (on --ok-bg, which moves with it) */
.portal-frame {
  background: var(--ui-card);
  color: var(--ui-ink);
  --ink-soft: var(--ui-soft);
  border: 1px solid var(--ui-line);
  border-radius: 14px;
  overflow: hidden;
  box-shadow: 0 1px 0 rgba(0, 0, 0, 0.04), 0 30px 60px rgba(20, 43, 73, 0.18);
}
.portal-chrome {
  display: flex;
  align-items: center;
  gap: 8px;
  padding: 10px 14px;
  background: var(--ui-bar);
  border-bottom: 1px solid var(--ui-line);
}
.portal-dot { width: 11px; height: 11px; border-radius: 50%; flex: none; }
.portal-dot--r { background: #FF5F57; }
.portal-dot--y { background: #FEBC2E; }
.portal-dot--g { background: #28C840; }
.portal-url {
  margin-inline: auto;
  display: flex;
  align-items: center;
  gap: 6px;
  min-width: 0;
  max-width: 380px;
  flex: 1 1 auto;
  background: var(--ui-card);
  border: 1px solid var(--ui-line);
  border-radius: 6px;
  padding: 5px 10px;
  font-size: 12px;
  color: var(--ui-soft);
}
/* THE ADDRESS TEXT IS BLURRED — item 3 of Guillaume's 4 Oct spec, and
   `filter: blur(4px)` is what Home.dc.html draws on it.

   4px AT A 12px FONT is a heavy blur on purpose: the point is that no
   particular address can be read off a mockup of a product that does
   not exist yet, while the shape of one still says "this is a
   browser". At 2px the words were still legible when I rendered it.

   THE BLUR IS THE ONLY THING MAKING IT UNREADABLE, so it must not be
   the kind of effect a browser can decline. filter is not animated,
   not transformed and not conditional on a media query — in
   particular it is NOT dropped under prefers-reduced-motion, because
   a static blur is not motion and dropping it there would reveal the
   text to exactly the users who asked for less movement.

   user-select: none so the string cannot be selected and pasted as a
   URL that resolves to nothing; aria-hidden in the HTML so it is
   never read aloud. Both are in addition to the frame's own
   aria-hidden, not instead of it.

   FORCED COLORS: a blur survives forced-colors mode, which is the
   right outcome here — the text stays unreadable rather than being
   repainted crisply in system colours. Checked with
   measure-layout.js's forced-colors pass. */
.portal-url-text {
  filter: blur(4px);
  -webkit-user-select: none;
  user-select: none;
  /* It is a blurred smudge, so it must not be allowed to wrap to two
     lines and grow the chrome row. */
  white-space: nowrap;
  overflow: hidden;
}
/* .portal-chip IS GONE — 4 Oct 2026, Guillaume's decision.

   It was the "Example data" chip, the one place on this page that
   said the portal's figures are invented, and it appeared twice: in
   the browser chrome on desktop and in the dark header card on phone.
   Both boards now draw the portal without it.

   ITS OWN COMMENT HERE USED TO SAY the disclaimer was safe to state
   once "because the mandatory banner already carries the bigger
   disclaimer". That sentence went stale the same day: the banner was
   removed from the homepage in 1b848d4 and now runs only on
   /quote/questions and /privacy. He was shown that before deciding
   and removed the chip anyway.

   The rule is deleted rather than left matching nothing, and this
   note is what stands in its place. The full argument, and the reason
   not to quietly reinstate it, is on the portal section in
   templates/index.html; homepage_test.go asserts the absence so a
   silent return is caught too. */

.portal-body { display: grid; grid-template-columns: 216px minmax(0, 1fr); }
/* DARK NAVY, per the board. It was light here until 4 Oct, which
   made the whole mockup read as one flat surface — the board's
   contrast between a dark rail and a light workspace is what makes
   it look like an application rather than a section of this page. */
/* THE SIDEBAR IS DARK IN BOTH THEMES, so its ink must be LIGHT in
   both -- and that is not --ui-card. --ui-card is the CARD surface:
   white in light mode and #182B43 in dark, so using it as ink here
   gave the rail dark text on dark navy in dark mode. Found by
   looking at the rendered dark page, where the wordmark and the
   user block vanished. #FFF9F1 is 14.63:1 on --ui-side in light and
   17.35:1 in dark; it is a literal because this surface does not
   change with the theme and neither should its ink. */
.portal-side {
  background: var(--ui-side);
  color: #FFF9F1;
  padding: 16px 12px;
  display: flex;
  flex-direction: column;
  gap: 14px;
}
.portal-brand {
  margin: 0 0 4px;
  padding-inline: 4px;
  color: #FFF9F1;
  display: flex;
  align-items: center;
  gap: 8px;
  font-weight: 700;
  font-size: 15px;
}
/* ⚠️ A SQUARE, AND IT IS SUPPOSED TO BE. Not an unconverted ring.
   23 Oct 2026, checked while converting the six typed Cs: this mark
   and the ring both sit beside the word "CoRoute", so it looks like
   one of them was missed. It was not.

   This is the APP TILE inside the portal mock-up -- a simulated
   product UI, the way a favicon appears in browser chrome -- and
   both homepage boards draw it as a rounded square at both widths.
   .ctxend-badge--t and .agent-avatar are the same kind of mark. The
   RING is the logotype, used where the page speaks as CoRoute; the
   SQUARE is an icon standing for the app inside a picture of the
   app.

   THE SENTENCE THAT WAS HERE SAID "The quote flow keeps a square
   mark too, by Guillaume's instruction of 11 Oct." TRUE WHEN
   WRITTEN, FALSE SINCE 23 OCT 2026: the v2 restyle gave the flow
   header the ring (.brand-c in .flownav-brand), and the note at
   .flownav's contrast warning says so in passing -- "the flow pages
   only acquired it on 23 Oct because the square mark they used
   before had no text in it at all".

   It cost a real decision. On 28 Oct the square mark was carried to
   Guillaume as an open question, because this line reads as current
   policy; he asked to see it, and the screenshots showed both
   headers already drawing the identical ring. The claim survived
   because the paragraph AROUND it is still correct, which is the
   most durable kind of stale comment. The rule below is the part
   that still binds.

   So: do not convert this to .brand-c. If it ever should change, it
   changes on the board first. */
.portal-brandmark {
  width: 10px; height: 10px; border-radius: 3px; background: var(--cta);
}
.portal-nav { list-style: none; margin: 0; padding: 0; display: grid; gap: 2px; }
.portal-nav-item {
  display: flex;
  align-items: center;
  gap: 10px;
  padding: 9px 12px;
  border-radius: 8px;
  font-size: 13px;
  color: var(--ui-side-ink);
}
.portal-nav-item.is-current {
  background: rgba(255, 255, 255, 0.1);
  color: #FFF9F1;
  font-weight: 600;
}
/* The user sits at the BOTTOM of the rail as a footer block, which
   is where the board puts it and where an app puts it. margin-top:
   auto is what pins it there however many nav items there are. */
.portal-user {
  margin: auto 0 0;
  display: flex;
  align-items: center;
  gap: 10px;
  margin-top: auto;
  padding: 12px;
  border-top: 1px solid rgba(255, 255, 255, 0.12);
}
.portal-avatar {
  flex: none;
  width: 32px; height: 32px;
  border-radius: 50%;
  /* Orange on navy, as the board has it: 7.9:1 for the initials. */
  background: #FFAA66;
  color: #142B49;
  font-size: 13px;
  font-weight: 700;
  display: flex;
  align-items: center;
  justify-content: center;
}
.portal-user-name { display: block; font-size: 13px; font-weight: 600; color: #FFF9F1; }
.portal-user-car { display: block; font-size: 11px; color: var(--ui-side-ink); }

/* A CONTAINER, so the split turns on the space it actually has
   rather than on the viewport. The sidebar is 216px and the frame
   breaks out of the gutter, so viewport width does not predict the
   main pane's width -- at 1024 the viewport looks wide while this
   pane is 745px, which is where the drives table collapsed. */
.portal-main { background: var(--ui-bg); padding: 20px 24px; display: grid; gap: 16px; min-width: 0;
               container: portal-main / inline-size; }
.portal-top {
  display: flex;
  flex-wrap: wrap;
  gap: 12px;
  align-items: flex-start;
  justify-content: space-between;
}
.portal-greet { margin: 0; font-size: 19px; font-weight: 700; }
.portal-sync { margin: 2px 0 0; font-size: 13px; color: var(--ui-soft); }
.portal-controls { margin: 0; display: flex; gap: 8px; }
.portal-search,
.portal-range {
  border: 1px solid var(--ui-line);
  border-radius: 8px;
  padding: 7px 11px;
  font-size: 13px;
  color: var(--ui-soft);
  white-space: nowrap;
}

/* FOUR EXPLICIT TRACKS, which is what Home.dc.html draws:
   `repeat(4, minmax(0,1fr))` with a 14px gap.

   It was `repeat(auto-fit, minmax(min(10rem, 100%), 1fr))`, and
   measured at 1280 that produced FIVE tracks inside the 934px row --
   four tiles at 224.5px and an empty 0px fifth, because auto-fit fills
   the line with as many 160px-floored tracks as fit and then collapses
   the ones with no item. Harmless to look at, but it means the tiles
   were never sharing the row the way the board divides it, and a fifth
   tile would silently land in a track nobody designed.

   The phone board draws three explicit tracks; that override is in the
   56rem block. Between the two, four tiles across 896px is 209px each,
   which still holds "Monthly premium" on one line. */
.portal-tiles {
  list-style: none;
  margin: 0;
  padding: 0;
  display: grid;
  gap: 12px;
  grid-template-columns: repeat(4, minmax(0, 1fr));
}
.portal-tile {
  border: 1px solid var(--ui-line);
  border-radius: 10px;
  padding: 13px 14px;
  background: var(--ui-card);
}
.portal-tile-label { margin: 0 0 6px; font-size: 12px; color: var(--ui-soft); }
.portal-tile-figure {
  margin: 0;
  font-size: 26px;
  font-weight: 700;
  line-height: 1.1;
  font-variant-numeric: tabular-nums;
}
.portal-tile-unit { font-size: 14px; font-weight: 600; color: var(--ui-soft); }
/* The arrow carries the direction and the words repeat it, so neither
   colour nor glyph is doing the job alone. */
.portal-tile-delta { margin: 6px 0 0; font-size: 12px; color: var(--ui-soft); }
/* THE FALLBACK IS DEAD CODE AND THE OLD NOTE PREDICTED THIS EXACTLY.

   What stood here read: "`var(--ok-ink, #17643A)` LOOKS like a pin
   and is not one: the fallback in a var() applies only when the
   token is UNDEFINED... It is safe ONLY because .portal-frame above
   re-points --ok-ink. So do not read the fallback as the defence --
   the defence is the frame's pin, and IF THAT PIN IS EVER REMOVED
   these three rules go back to 1.56:1 and 1.38:1 with no warning."

   V2 REMOVED THE PIN, because the frame now follows the theme, and
   the warning came true: scripts/check-contrast.js found
   .portal-badge at 1.38:1 in dark mode, which is the precise figure
   the old comment names. It was the only failure left on the page,
   and it was found by walking computed styles, not by reading CSS --
   exactly as predicted, "with no warning".

   THE FIX IS THAT THE PAIR MOVES TOGETHER. --ok-ink and --ok-bg are
   a per-theme PAIR (#17643A on #EAF7EE light, #8FE0AD on #16301F
   dark) and they are correct together in both themes: 6.52:1 and
   9.10:1. The failure was never the ink -- it was an ink from one
   theme on a background literal from the other. Every rule below
   that uses --ok-ink now names --ok-bg beside it, so the two cannot
   be separated again by editing one of them.

   The var() fallbacks are dropped with the pin: a fallback that can
   never fire is a comment pretending to be code. */
.portal-tile-delta--up { color: var(--ok-ink); }
.portal-tile-delta--down { color: var(--ok-ink); }

.portal-card {
  border: 1px solid var(--ui-line);
  border-radius: 10px;
  padding: 14px;
  background: var(--ui-card);
}
.portal-card-head {
  margin: 0 0 10px;
  display: flex;
  align-items: center;
  justify-content: space-between;
  gap: 12px;
}
.portal-card-title { font-size: 14px; font-weight: 700; }
.portal-card-link { font-size: 12px; color: var(--brand); }
.portal-badge {
  font-size: 11px;
  font-weight: 700;
  /* THE INK AND THE FILL ARE ONE PAIR AND BOTH ARE TOKENS. The fill
     was a literal #E6F4EC while the ink followed --ok-ink, so in dark
     mode this was the pale mint on a light green: 1.38:1, the single
     worst pair left on the page. 6.52:1 light / 9.10:1 dark now.
     See the long note at .portal-tile-delta for how it got there. */
  color: var(--ok-ink);
  background: var(--ok-bg);
  border-radius: 999px;
  padding: 3px 9px;
}

/* TWO COLUMNS, 60/40 — the board's minmax(0,1.5fr) minmax(0,1fr).
   Drives on the left; Premium above Your policy on the right.

   It was one stacked column until 4 Oct, which is why the live
   portal ran to roughly twice the board's height: the same content,
   no frame, scrolling instead of composing. align-items: start
   stops the shorter column stretching to match the taller. */
/* ONE COLUMN UNTIL 64rem, TWO ABOVE IT. The board draws this split at
   1280, where the frame is 1200 and the left card gets 550.8px. At
   1024 the frame is 1009, the 216px sidebar is still showing, and the
   left card falls to ~437px -- which leaves 29px for the flexible
   note column after the row's fixed 270+48+16 and its three 14px
   gaps. Measured before this rule: note 0px, wrapped to 12 lines,
   rows 241px tall. The stacked single column is correct there; the
   drives table simply needs ~550px before it can sit beside anything.
   64rem is an existing breakpoint (quote-layout R1), not a new one. */
.portal-split {
  display: grid;
  grid-template-columns: minmax(0, 1fr);
  gap: 16px;
  align-items: start;
}
/* NO FIFTH BREAKPOINT. quote-layout R1 allows four widths and the
   split needs to turn at ~1180, which is none of them. So it turns on
   its OWN width instead of the viewport's: the left card must reach
   the board's 550.8px before anything sits beside it, and 550.8 +
   16 gap + its 1fr sibling means the split needs ~918. Below that
   the two tracks wrap and stack, which is what `auto-fit` with a
   floor does -- no media query, and correct at every width rather
   than at the four we happened to test. The 1.5fr/1fr ratio is kept
   above the floor by the explicit second track. */
.portal-split { grid-template-columns: minmax(0, 1.5fr) minmax(0, 1fr); }
@container portal-main (max-width: 57.4rem) {
  .portal-split { grid-template-columns: minmax(0, 1fr); }
}
.portal-rightcol { display: grid; gap: 16px; align-content: start; }

/* THE DRIVES LIST IS A <ul>, NOT A <table>, because each row is a
   grid of four unlike things — a route thumbnail, a trip, a score
   and a note — rather than cells to read down a column. */
/* THE TRIP COLUMN IS SIZED IN ch, NOT fr, and that is the fix for a
   real overlap. With 1.35fr it resolved to 158px inside the 60%
   column, while "Yesterday 17:46 · 3.1 mi · 12 min" needs about 215
   — and because .portal-when is nowrap, the text PAINTED OVER the
   note beside it instead of wrapping or clipping.

   Nothing caught that: the bounding boxes did not overlap, so a
   geometry check said the row was fine. Only the screenshot showed
   the collision. Sizing the column to its longest line makes the
   grid honour the content rather than the ratio. */
/* THESE ARE THE BOARD'S OWN TRACK SIZES, read from Home.dc.html
   (version 1791115430-3f17): `270px 48px minmax(0,1fr) 16px` with a
   14px gap. The trip column is FIXED, not content-sized. An earlier
   `auto` here let the nowrap trip text claim width and squeezed the
   AI note to 97px, which wrapped it to six lines and turned each row
   into a paragraph. The note column is the one that flexes. */
.portal-rowhead,
.portal-row {
  display: grid;
  grid-template-columns: 270px 48px minmax(0, 1fr) 16px;
  align-items: center;
  gap: 14px;
}
.portal-rowhead {
  margin: 0;
  padding: 0 16px 8px;
  font-size: 11px;
  letter-spacing: 0.06em;
  text-transform: uppercase;
  color: var(--ui-soft);
  font-weight: 700;
}
.portal-rows { list-style: none; margin: 0; padding: 0; }
.portal-row {
  padding: 11px 16px;
  border-top: 1px solid var(--ui-line);
  font-size: 13px;
}
.portal-row-trip { display: flex; align-items: center; gap: 10px; min-width: 0; }
/* THE DRIVES CARD CARRIES NO PADDING OF ITS OWN. On the board this
   card is padding: 0 -- its header (14px 16px) and its rows (11px
   16px) pad themselves, and the row separators run the full width of
   the card, which they cannot do if the card is inset. The shared
   14px cost the flexible note column 28px: 112.8px instead of the
   board's 140.8px, six wrapped lines instead of four. */
.portal-card--rows { padding: 0; overflow: hidden; }
/* The card no longer pads, so the header pads itself -- the board's
   own `padding: 14px 16px`, which lines its text up with the 16px
   row padding below it. */
.portal-card-head--rows { margin: 0; padding: 14px 16px; }

/* The route thumbnail: a drawn map, 56x38, carried from the board. */
.portal-route {
  flex: none;
  border-radius: 6px;
  background: var(--ui-bg);
  border: 1px solid var(--ui-line);
}

/* THE PORTAL ARTWORK IS DRAWN IN TOKENS, NOT IN HEXES.
   -------------------------------------------------------------------
   Both SVGs below -- the trip thumbnail and the premium chart -- used
   to carry their colours as literal hexes in index.html. That is how
   the chart came to be INVISIBLE in dark mode: every series was
   #142B49 navy on a #182B43 navy card, measured at 1.00:1, while the
   #EAF4FA gridlines -- which should sit quietly behind the data --
   measured 12.84:1 and became the brightest thing in the card. The
   chart was exactly inverted, and no DOM or text check can see that,
   because every element was present and correct in the markup.

   HomeV2.dc.html and HomeV2Dark.dc.html are BYTE-IDENTICAL in this
   region and contain no hexes at all: the board classes every part and
   lets the tokens do the theming, which is why one board serves both
   themes. These rules are that board's own stylesheet, carried over.
   Keep it that way -- a hex added here is a hex that cannot follow the
   theme, and it will be invisible in exactly one of the two. */
.portal-route .bg { fill: var(--map-bg); }
.portal-route .pk { fill: var(--map-park); }
.portal-route .wt { stroke: var(--map-water); }
.portal-route .rd { stroke: var(--map-road); }
.portal-route .mn { stroke: var(--map-main); }
.portal-route .rt { stroke: var(--ui-accent); }
.portal-route .st { fill: var(--ui-card); stroke: var(--ui-accent); }
.portal-route .en { fill: var(--ui-accent); stroke: var(--ui-card); }

/* The premium chart. `.ac` is the series stroke (both the solid paid
   line and the dashed projection), `.af` the band under the
   projection, `.gl` the gridlines, `.vl` the today marker. The
   gradient under the paid line takes `currentColor`, which the SVG's
   own `<g style="color:var(--ui-accent)">` supplies -- that is the
   board's mechanism for giving a <stop> a token, since stop-color
   cannot read a custom property through a fill="url(#...)". */
.portal-chart .gl   { stroke: var(--ui-line); }
.portal-chart .ac   { stroke: var(--ui-accent); }
.portal-chart .af   { fill: var(--ui-accent); }
.portal-chart .vl   { stroke: var(--ui-ink); }
.portal-chart .dot  { fill: var(--ui-card); stroke: var(--ui-accent); }
.portal-chart .tip  { fill: var(--ui-ink); }
.portal-chart .tipt { fill: var(--ui-card); }
.portal-chart .tips { fill: var(--ui-card); opacity: .8; }
.portal-chart text  { fill: var(--ui-soft); }
.portal-row-names { min-width: 0; }
.portal-trip { display: block; font-weight: 600; color: var(--ui-ink); white-space: nowrap; }
.portal-when { display: block; font-size: 12px; color: var(--ui-soft); white-space: nowrap; }
/* TOKENS, not the pastel literals this carried until 23 Oct 2026.
   HomeV2.dc.html tokenises every one of these chips (`.sc.g` is
   --ok-bg/--ok-ink, `.sc.n` is --ui-accent-soft/--ui-accent, the
   `.nochip .x` is --err-bg/--err-ink). The literals stayed internally
   legible in dark mode -- both ink AND surface were pinned, so they
   measured 6.43:1 and 5.74:1 -- which is exactly why nothing flagged
   them. They were simply six pale-mint badges glowing on a navy card,
   a contrast checker's blind spot and an eye's first complaint. */
.portal-score {
  display: inline-block;
  min-width: 34px;
  text-align: center;
  padding: 3px 8px;
  border-radius: 999px;
  background: var(--ok-bg);
  color: var(--ok-ink);
  font-weight: 700;
  font-size: 12px;
  font-variant-numeric: tabular-nums;
}
/* --ui-accent-soft / --ui-accent, as HomeV2.dc.html has `.sc.n`. */
.portal-score--mid { background: var(--ui-accent-soft); color: var(--ui-accent); }
/* The dot repeats what the score already says, in colour — so it is
   decoration, and the score carries the meaning for anyone who
   cannot tell green from amber. */
.portal-note {
  display: flex;
  gap: 8px;
  align-items: baseline;
  font-size: 13px;
  line-height: 1.4;
  /* --ui-ink, not a literal: #2A3E55 was chosen against the pinned
     white card and measures 1.30:1 on the dark --ui-card. This is
     the "What our AI noticed" column -- the most substantive text in
     the mock-up -- and it was invisible in dark mode. 14.27:1 light,
     12.03:1 dark. */
  color: var(--ui-ink);
  min-width: 0;
}
.portal-dot-note {
  flex: none;
  width: 7px; height: 7px;
  border-radius: 50%;
  background: #28A860;
  transform: translateY(-1px);
}
.portal-dot-note--warn { background: #D9822B; }
.portal-chev { color: #AAC0CD; flex: none; }

/* THE KEYS ARE LINES, NOT SQUARES, because the chart draws lines.

   NOT IN THE ITERATION-1 BRIEF — found in the side-by-side image.
   The page drew two 9x9 rounded squares, one solid purple and one
   pale; the board draws a 14x3 solid bar for Paid and a 14px dashed
   rule at 70% opacity for Projected. Read from Home.dc.html
   (version 1791115430-3f17).

   It matters more than it looks: the chart's whole claim is "solid is
   what you paid, dashed is what we project". A legend keyed by COLOUR
   rather than by line style asks the reader to map pale-purple onto
   dashed themselves, and the two swatches differ only in lightness —
   which is the one cue a colour-blind reader is least likely to get.
   Matching the board fixes an accessibility problem and a fidelity
   problem with the same rule. */
.portal-legend { display: flex; gap: 14px; font-size: 11px; color: var(--ui-soft); }
.portal-key { display: flex; align-items: center; gap: 6px; }
.portal-key::before {
  content: "";
  width: 14px;
  flex: none;
}
.portal-key--paid::before {
  height: 3px;
  border-radius: 2px;
  background: var(--brand);
}
.portal-key--proj::before {
  height: 0;
  border-top: 2px dashed var(--brand);
  opacity: 0.7;
}
/* Forced-colors drops `background` on the solid key; a border keeps
   both keys visible and still distinguishable by style. */
@media (forced-colors: active) {
  .portal-key--paid::before { border-top: 3px solid CanvasText; }
  .portal-key--proj::before { border-top: 2px dashed CanvasText; opacity: 1; }
}
.portal-chart-label { margin: 0 0 10px; font-size: 11px; color: var(--ui-soft); }

/* A CSS bar row rather than an image or a chart library: the CSP
   allows no CDN, and a picture of a chart cannot scale with text.
   --h is the bar's height as a percentage of the plot. NOTHING RISES
   — the board's shape only ever falls or holds, because a bar growing
   upward would read as a price going up. */
.portal-chart {
  list-style: none;
  margin: 0;
  padding: 0;
  display: grid;
  grid-auto-flow: column;
  grid-auto-columns: 1fr;
  align-items: end;
  gap: 5px;
  height: 108px;
}
/* THE CHART IS AN SVG LINE, NOT BARS. The board draws a trajectory:
   one line falling left to right, SOLID for months already paid and
   DASHED for projected, with a shaded band under each and a marker
   circle on the current month.

   It was twelve CSS bars until 4 Oct. Bars show twelve separate
   amounts; a line shows the thing actually being claimed, which is
   that the price comes down over time. The markup is the board's own
   SVG carried across rather than redrawn, so the shape is identical
   and cannot drift from it.

   role="img" with an aria-label describing the TREND in words,
   because a polyline is unreadable to a screen reader and the trend
   is the whole content. */
.portal-card svg[role="img"] { display: block; width: 100%; height: auto; }

/* TWO GREY TILES, NOT TWO RUNS OF TEXT ON A RULE.

   NOT IN THE ITERATION-1 BRIEF -- found by looking at the side-by-side
   image, which is the point of looking. Measured on Home.dc.html
   (version 1791115430-3f17): a `1fr 1fr` grid, 10px gap, 10px
   margin-top, each pair in its own #F7F6FA box with a 10px radius and
   10px 12px of padding, 162x61 at 1280.

   The page had a flex row on a 1px top rule with the pairs sized to
   their text -- 87px and 106px, transparent, no tile. Every string was
   present, so compare-board.js was green; the shape was wrong, which
   only the image shows. The figures are the panel's conclusion and the
   board gives them a surface to sit on.

   The 56rem block overrides this back to one inline line, because the
   PHONE board does it differently (P5) -- label and figure inline,
   both pairs on a single row, no tiles. */
.portal-chart-foot {
  margin: 10px 0 0;
  display: grid;
  grid-template-columns: minmax(0, 1fr) minmax(0, 1fr);
  gap: 10px;
}
.portal-chart-foot > span {
  background: var(--ui-bg);
  border-radius: 10px;
  padding: 10px 12px;
}
.portal-foot-label { display: block; font-size: 11px; color: var(--ui-soft); }
.portal-foot-figure {
  display: block;
  font-size: 18px;
  font-weight: 700;
  font-variant-numeric: tabular-nums;
  color: var(--ui-ink);
}
/* The board colours the saved figure green — it is money kept, not a
   neutral total — and leaves the projected one in ink. */
/* --ok-ink, not a literal: #17643A on the dark --ui-card is 1.60:1.
    The token is the per-theme pair -- #17643A light, #8FE0AD dark --
   and its dark value measures 9.17:1 on the card it now sits on. */
.portal-chart-foot > span:first-child .portal-foot-figure { color: var(--ok-ink); }

/* D2 — TWO COLUMNS, TWO ROWS, which is what Home.dc.html draws:
   `grid-template-columns: 1fr 1fr; gap: 10px 16px`, so Cover sits
   beside Car and Excess beside Renews.

   It was `repeat(auto-fit, minmax(min(9rem, 100%), 1fr))`, and that is
   why all four stacked. auto-fit asks each track for at least 9rem
   (144px) BEFORE it will place a second one, and the right-hand column
   at 1280 gives this panel about 338px of content box -- enough for
   two 144px tracks on paper, but "Ford Fiesta 1.0 · 2019" does not fit
   144px, so the implicit minimum content width of that item pushed the
   track over its share and auto-fit fell back to one column.

   TWO EXPLICIT 1fr TRACKS instead. 1fr floors at auto by default,
   which would reproduce the same problem, so each is minmax(0, 1fr):
   the tracks split the width evenly and the long value wraps inside
   its cell rather than widening it. The board's own values. */
.portal-policy {
  margin: 0;
  display: grid;
  gap: 10px 16px;
  grid-template-columns: minmax(0, 1fr) minmax(0, 1fr);
  font-size: 13px;
}
.portal-policy dt { color: var(--ui-soft); font-size: 11px; }
.portal-policy dd { margin: 2px 0 0; font-weight: 600; }

/* ---- PHONE-ONLY PARTS OF THE PORTAL ------------------------------
   The dark header card and the bottom tab bar exist on
   HomePhone.dc.html and NOT on Home.dc.html, so they are hidden here
   and switched on inside the 56rem block. Hidden by default rather
   than shown-then-hidden, so a width nobody tested shows the desktop
   composition the desktop board actually draws. */
.portal-hdr,
.portal-bottomnav { display: none; }

/* The short tile labels and the phone's "Renewal" are likewise the
   phone board's wording; desktop shows the long forms. */
.portal-lbl-short { display: none; }

/* BELOW 56rem THE SIDEBAR GOES. A 216px rail inside a 360px screen
   leaves nothing for the content, and the sidebar is the least
   informative part of the mockup — the tiles and the drives table are
   what make the point. */
@media (max-width: 55.9375rem) {
  .portal { padding-block: 3rem; }
  .portal-side { display: none; }
  .portal-body { grid-template-columns: minmax(0, 1fr); }
  .portal-main { padding: 14px; }

  /* P3 — NO BROWSER CHROME ON PHONE, AND A DARK HEADER CARD INSTEAD.
     HomePhone.dc.html draws no browser frame at all: no traffic-light
     dots, no URL pill. A phone does not show a desktop browser's
     window furniture, and the board is right about that. The chrome
     row goes and the dark navy card takes its place. */
  .portal-chrome { display: none; }
  .portal-hdr {
    display: block;
    background: #0B1626;
    /* A LITERAL, not --ui-card: this card is dark in BOTH themes
       (the board draws it that way on phone), so its ink must be
       light in both. --ui-card is the card SURFACE and goes dark in
       dark mode, which put dark text on dark navy here. */
    color: #FFF9F1;
    padding: 16px 16px 18px;
  }
  .portal-hdr-top {
    display: flex;
    align-items: center;
    justify-content: space-between;
    gap: 8px;
    margin-bottom: 14px;
  }
  .portal-hdr-brand {
    display: flex;
    align-items: center;
    gap: 8px;
    font-weight: 700;
    font-size: 14px;
  }
  .portal-hdr-right {
    display: flex;
    align-items: center;
    gap: 10px;
  }
  /* The avatar inside the header card is 30px, not the sidebar's 32px —
     the board's own figure. */
  .portal-hdr .portal-avatar {
    width: 30px;
    height: 30px;
    font-size: 12px;
  }
  .portal-hdr-greet { margin: 0; font-size: 13px; color: #BFD3E2; }
  .portal-hdr-price {
    margin: 2px 0 0;
    display: flex;
    align-items: baseline;
    flex-wrap: wrap;
    gap: 4px 10px;
  }
  .portal-hdr-figure {
    font-size: 30px;
    font-weight: 700;
    font-variant-numeric: tabular-nums;
    line-height: 1.1;
  }
  /* #9FE3B8 on #1F0D53 is the board's pairing and measures about 8.6:1,
     well past 4.5. The arrow carries the direction too, so the colour
     is not doing the work alone. */
  .portal-hdr-delta { font-size: 13px; font-weight: 600; color: #9FE3B8; }
  .portal-hdr-cap { margin: 2px 0 0; font-size: 12px; color: #BFD3E2; }
  /* THE SPLIT COLLAPSES. 60/40 inside 360px gives two columns of
     nothing; the right-hand panels go below the drives instead. */
  .portal-split { grid-template-columns: minmax(0, 1fr); }

  /* THE AI NOTE SHOWS ON EVERY ROW AT THIS WIDTH, and the row is laid
     out to make room for it. HomePhone.dc.html (version
     1791115430-3f17) draws three things per row -- a 48x33 route
     thumbnail, a flexible column holding the trip name, the
     time/distance line AND the contextual note wrapped over two to
     three lines, then the score badge. No chevron and no SCORE column
     header.

     THE BOARD IS THE REASON. An earlier rule here hid .portal-note
     below 56rem on the argument that it was "the least readable at
     this width". That is the opposite of what the board shows, and the
     note is the section's whole claim -- "our AI reviews every drive
     IN CONTEXT". A portal section that drops the context on phone
     keeps the furniture and discards the argument.

     What gives way instead is the chevron and the SCORE header, per
     the board: the chevron is a decorative affordance on a mockup
     nothing can click, and the header labelled a column that is no
     longer a column. */
  .portal-trip,
  .portal-when { white-space: normal; }

  /* THE ROW BECOMES A 2x2 GRID, which is how the board's nesting lands
     without moving any markup.

     The board nests the note INSIDE the text column, under the time
     line. This page's markup has it as a SIBLING after the score
     (because on desktop it is its own grid track), and the template is
     not restructured to suit one width -- quote-layout's "edit, do not
     regenerate" applies to the mockup too, and desktop depends on that
     order.

     So the grid does the placing instead:

       col 1 (48px)   col 2 (1fr)    col 3 (auto)
       +------------+--------------+-----------+
       | thumbnail  | trip + time  |   score   |   row 1
       +------------+--------------+-----------+
       |            | the AI note              |   row 2
       +------------+--------------------------+

     The note starts in column 2 and runs to the end, so it wraps
     against the text column's full width plus the score's -- the two to
     three lines the board shows -- and it sits under the trip name
     exactly as drawn. Nothing is reordered in the DOM, so the reading
     order is still trip, score, note. */
  .portal-row {
    display: grid;
    grid-template-columns: 48px minmax(0, 1fr) auto;
    align-items: center;
    gap: 5px 12px;
    padding: 11px 14px;
  }
  /* `contents` promotes the thumbnail and the name block to row items
     so the grid can place them in separate columns. */
  .portal-row-trip { display: contents; }
  .portal-route {
    width: 48px;
    height: 33px;
    grid-column: 1;
    grid-row: 1;
  }
  .portal-row-names {
    display: block;
    min-width: 0;
    grid-column: 2;
    grid-row: 1;
  }
  .portal-score {
    grid-column: 3;
    grid-row: 1;
    justify-self: end;
  }
  /* ROW 2, STARTING IN COLUMN 2: under the trip name, not under the
     thumbnail, and running to the right edge. */
  .portal-note {
    grid-column: 2 / -1;
    grid-row: 2;
    font-size: 12px;
    line-height: 1.4;
    align-items: baseline;
  }
  .portal-trip {
    font-size: 14px;
    white-space: nowrap;
    overflow: hidden;
    text-overflow: ellipsis;
  }
  .portal-when { display: block; font-size: 12px; }
  /* NO CLOCK TIME ON PHONE. The two boards differ by exactly this
     token, checked against both sources:

       Home.dc.html       "Today 08:12 · 9.4 mi · 24 min"
       HomePhone.dc.html  "Today · 9.4 mi · 24 min"

     Deliberate on desktop, deliberately absent on phone: the trip
     column is the narrowest on the page and the clock time is the
     least load-bearing thing in the row. The day word, the distance
     and the duration all survive.

     It is a span inside .portal-when rather than a second row, so
     there is one set of trip markup and the desktop keeps its time. */
  .portal-clock { display: none; }
  /* THE CHEVRON AND THE SCORE HEADER ARE WHAT GIVE WAY, per the board
     and per the task's instruction that if something must go it is
     these and not the note. The chevron is a decorative affordance on
     a mockup nothing can open; the header named grid columns the phone
     row no longer has. */
  .portal-chev,
  .portal-rowhead { display: none; }

  /* FOUR ROWS ON PHONE, not six — HomePhone.dc.html shows four. The
     list is an illustration, and two more rows of invented trips on
     a screen this narrow is scrolling without argument. The desktop
     board keeps all six. */
  .portal-rows li:nth-child(n + 5) { display: none; }

  /* P4 — THREE TILES ACROSS IN ONE ROW, SHORT LABELS, BARE DELTAS.
     HomePhone.dc.html: `repeat(3, minmax(0,1fr))` with an 8px gap,
     12px padding, a 22px figure, and labels of one word. The premium
     tile is not there at all — the dark header card above is where the
     price lives on phone, which is the whole reason the phone row is
     three tiles and the desktop row is four. */
  .portal-tiles { grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 8px; }
  .portal-tile { padding: 12px; }
  .portal-tile-figure { font-size: 22px; }
  /* The figure's "/100" goes with the long labels: the board's phone
     tile reads a bare "89". */
  .portal-tile-unit { display: none; }
  .portal-tile--premium { display: none; }
  .portal-lbl-short { display: inline; }
  .portal-lbl-long,
  .portal-delta-tail { display: none; }

  /* P5 — THE PREMIUM FOOTER IS ONE LINE. The board runs both pairs
     along a single row with the label and the figure INLINE —
     "Saved since April £40.60" then "Renewal ~£116/mo" — rather than
     the desktop board's two grey tiles with the figure beneath.
     justify-content: space-between puts one at each end, as drawn. */
  .portal-chart-foot {
    display: flex;
    flex-wrap: nowrap;
    justify-content: space-between;
    margin: 6px 0 0;
    gap: 10px;
    font-size: 12px;
    color: var(--ui-soft);
  }
  /* NO TILES ON PHONE. The desktop board boxes each pair; the phone
     board runs them as two plain inline runs along one line. */
  .portal-chart-foot > span {
    background: none;
    padding: 0;
    border-radius: 0;
  }
  .portal-foot-label { display: inline; font-size: 12px; }
  .portal-foot-figure { display: inline; font-size: 12px; font-weight: 700; color: var(--ui-ink); }
  .portal-chart-foot > span:first-child .portal-foot-figure { color: var(--ok-ink); }

  /* P6 — THE BOTTOM TAB BAR, INSIDE THE CARD. The board closes the
     phone portal with four items on a white strip: a 20px icon above a
     10px label, Home current in brand purple and the rest in grey. It
     is what makes the frame read as a phone app rather than a cropped
     dashboard.

     Decorative, like the rest of the frame (.portal-frame is
     aria-hidden), so these are spans — not links, not tabs, and not
     announced as either. Nothing here claims to be operable. */
  .portal-bottomnav {
    display: flex;
    justify-content: space-around;
    padding: 10px 8px 14px;
    background: var(--ui-card);
    border-top: 1px solid var(--ui-line);
  }
  .portal-tab {
    display: flex;
    flex-direction: column;
    align-items: center;
    gap: 3px;
    font-size: 10px;
    font-weight: 600;
    color: #8DA0B0;
  }
  .portal-tab.is-current { color: var(--brand); }

  /* WHAT THE PHONE BOARD DOES NOT HAVE, and the page was showing.
     None of these were in the iteration-1 brief; all four turned up in
     the board-beside-page image and were then confirmed against the
     board SOURCE, where each string appears zero times:

       "Your last drive synced"  0      "Search drives"  0
       "Last 30 days"            0      "Your policy"    0
       "Comprehensive"           0      "Excess"         0
       "Monthly price, April to next March"  0

     ("Good morning" appears once — in the dark header card, which is
     why the second greeting below it is the duplicate, not the card.)

     The greeting and the sync line are IN the dark header card on
     phone, so repeating them below it says the same thing twice. The
     search and range controls are desktop furniture. The policy panel
     is a desktop panel; the phone card ends at the premium chart and
     goes to the tab bar. The chart caption is dropped because the
     phone chart is half the width and the axis already reads Apr-Mar. */
  .portal-top,
  .portal-card--policy,
  .portal-chart-label { display: none; }
}

/* ---- Contextual AI ----------------------------------------------- */

/* DARK GROUND, per the board: #1F0D53 with cream text. It is the
   banner's purple, and it sets this section apart from the panel
   above and the page below — the argument gets its own surface.

   The tokens are written out rather than taken from --page/--ink
   because this block does NOT follow the theme: it is dark in light
   mode and dark in dark mode, the way the banner and footer are. */
/* --band / --band-text, NOT LITERALS, SINCE V2. The page pinned
   #0B1626 in both themes; sampled off the rendered boards, HomeV2
   draws this band #142B49 and HomeV2Dark draws it #0B1626. So the
   literal was the DARK board's value applied to both -- which is
   why it looked right in dark mode and a shade too deep in light.
   Found by profiling the two halves of the comparison image down
   the gutter, not by eye: the difference between #0B1626 and
   #142B49 is not something to spot at a glance, which is exactly
   why it survived.

   --band-text on --band is 13.64:1 light and 17.35:1 dark. */
.ctxai {
  background: var(--band);
  color: var(--band-text);
  padding-block: 5rem;
}
/* .ctxai-eyebrow IS GONE, 5 Oct 2026. Both boards removed the
   "CONTEXTUAL AI · PLANNED CAPABILITY" eyebrow and made "Contextual
   AI" the first line of the title instead. The rule is deleted
   rather than left orphaned.

   The honesty words it carried did NOT go with it — they are now
   .ctxai-hedge below, under the lede. index.html's comment on that
   element carries the full argument and the question for Guillaume.
   Do not restore this rule without reading it. */

/* THE TITLE IS TWO LINES AND THEY ARE ONE HEADING.

   Board: 52px desktop, 34px phone, both lines the same size, no
   full stops. Line one "Contextual AI" in #FFAA66, line two "It
   understands" in the section's cream.

   ONE <h2> WITH A BLOCK <span>, not two headings and not a <br>.
   Two headings would put a second entry in the document outline for
   what is one title; a <br> would work visually but leaves the
   colour change needing a span anyway, and a span that is already
   there can carry `display: block` for free.

   IT USES THE SHARED SECTION SIZE, --section-h: 48px at 1280 and
   32px at 390, identical to .portal-h. Canvas version
   1791232647-f669 normalised every section heading to one size and
   says so on the board: "ALL section headings one size
   (Guillaume): 48px, 800, -0.02em, line-height 1.1. Hero h1 stays
   larger."

   An earlier build today used 52/34, from version ...-4f97, which
   drew this title larger than the portal's. That was superseded
   about an hour later by the normalisation above. Recorded because
   a reader finding 52px in the git history should know it was
   right when written, not a mistake.

   line-height 1.1 is the board's, and it is tighter than the 1.18
   this rule used to carry because the title is now two short lines
   rather than one long wrapped one. */
/* "CoRoute AI" over a small "it understands", 16 Oct 2026.
   .ctxai-h keeps var(--section-h), so the brand line is already the
   same 48px/800 as "How it works" — measured both at 48px before the
   change, which is why nothing here sets a size for it. */
/* display:block so the brand line and "it understands" stack.
   The ring inside it is inline-block and shares the line box with
   the name, so no flex alignment is needed here. */
.ctxai-h-brand { display: block; }

/* THE C AND "AI" ARE ORANGE, the middle of the name is not.
   #FFAA66 on this section's dark ground is 8.67:1. --brand would be
   2.1:1 and unreadable — this section does not follow the theme. */
/* THIS SECTION DOES NOT FOLLOW THE THEME: it is the dark navy band
   in both. So --logo is overridden FOR THE SECTION rather than the
   colour being written onto the C -- the C, the "o" and the "AI"
   then all move together. #FFAA66 on this ground is 8.67:1; --logo's
   light value would be 2.1:1 and unreadable. */
.ctxai { --logo: #FFAA66; }
.ctxai-h-ai { color: var(--logo); }

/* The sub-line is small and quiet: it is a continuation of the brand
   line above it, not a second heading. Lowercase as Guillaume wrote
   it, and the cream the section already uses for body text. */
.ctxai-h-sub {
  display: block;
  margin-top: 0.3rem;
  font-size: 1.0625rem;
  font-weight: 600;
  letter-spacing: 0;
  color: #D1E1F0;
}

/* 200% TEXT: THE TITLE MUST BE ABLE TO BREAK INSIDE A WORD.

   Measured at 360px with 200% text, when this heading was 52px:
   "Contextual" alone rendered 367px in a 280px column and the page
   scrolled sideways by 100px. A heading this size doubles past any
   phone column's width for a ten-letter word — no amount of
   wrapping BETWEEN words helps, because the overflow is one word.
   The size has since come down to 48px, which does not make the
   hazard go away; the rule stays and is re-measured in the report.

   overflow-wrap: break-word only breaks when the word cannot fit,
   so at every normal text size this changes nothing and the title
   still reads as written — "CoRoute AI" over a small "it
   understands" since 16 Oct 2026, where it was "Contextual AI" over
   a large "It understands" before, which is what both boards still
   draw. It engages only where the alternative is a page
   that scrolls sideways, which WCAG 1.4.10 forbids.

   The hero h1 carries the same pair for the same reason; this is
   that rule applied to the one other oversized heading on the
   page. */
/* The spans changed on 16 Oct (.ctxai-h-lead became -brand/-c/-ai
   plus a -sub); the wrap guard follows them, because what it
   protects against is a long word in an oversized heading forcing
   sideways scroll at 200% text, and that hazard belongs to whatever
   span holds the words. */
.ctxai-h,
.ctxai-h-brand,
.ctxai-h-sub {
  overflow-wrap: break-word;
  hyphens: auto;
}
.ctxai-h {
  margin: 0 0 0.5rem;
  text-align: center;
  font-size: var(--section-h);
  line-height: 1.1;
  font-weight: 800;
  letter-spacing: -0.02em;
}
.ctxai-h { max-width: 22ch; margin-inline: auto; }
.ctxai-lede {
  /* The 2.75rem that used to sit here is now on .ctxai-hedge, which
     follows this paragraph. R7: the gap belongs to whatever is last
     before the cards, and stacking a bottom margin here on top of
     the hedge's own would double it. */
  margin: 0 auto 0.75rem;
  max-width: 62ch;
  text-align: center;
  font-size: 1.0625rem;
  line-height: 1.55;
  color: rgba(255, 249, 241, 0.82);
}

/* THE HONESTY LABEL FOR THIS SECTION'S PROSE — 5 Oct 2026.

   It replaces the words "planned capability" that used to ride in
   the eyebrow above the title. The boards deleted the eyebrow and
   placed nothing instead; the label is kept because the claim it
   hedges is still here, and because it is the only honesty marker
   on the page outside the video chips. index.html carries the full
   argument, the wording's provenance, and the question for
   Guillaume. Read it before changing this.

   DELIBERATELY QUIET. The eyebrow was uppercase, letter-spaced
   #FFAA66 above the title — more visual weight than some of the
   copy it hedged, and it competed with the title the board has just
   shortened. This is the lede's scale and a dimmer cream, sitting
   under the lede where it is read before the scene cards.

   CONTRAST, computed from the composite rather than estimated:
   rgba(255,249,241,0.72) over #1F0D53 flattens to #C0B7C5, which
   measures 8.72:1 against the ground. The <b> is #FFF9F1 at full
   opacity, 16.18:1, so the emphasised words cannot be the weaker of
   the two. Both clear 4.5:1 with room to spare; the label is quiet
   in WEIGHT, not in legibility. */
.ctxai-hedge {
  margin: 0 auto 2.75rem;
  max-width: 62ch;
  text-align: center;
  font-size: 0.9375rem;
  line-height: 1.5;
  color: rgba(255, 249, 241, 0.72);
}
.ctxai-hedge b {
  font-weight: 700;
  color: #FFF9F1;
}
.ctxai-scenarios {
  list-style: none;
  margin: 0 0 3rem;
  padding: 0;
  display: grid;
  gap: 1.25rem;
  grid-template-columns: repeat(auto-fit, minmax(min(20rem, 100%), 1fr));
}
/* Translucent white on the dark ground rather than --paper: a solid
   white card here would read as a hole punched in the section. */
.ctxai-scenario {
  background: rgba(255, 255, 255, 0.05);
  border: 1px solid rgba(255, 255, 255, 0.12);
  border-radius: 18px;
  padding: 1.125rem;
}
/* The scene clip. Both chips sit ON the frame, so they survive a
   screenshot of it — the same reasoning as the dashcam card, and on
   these it is load-bearing rather than tidy: the boxes and captions in
   this footage were drawn BY HAND, so a moving picture with no chip on
   it is a claim the page must not make.

   IT IS A <button> SINCE 4 OCT 2026 — clicking it plays the clip, and
   a control has to work without a mouse. Same undoing of the UA button
   styling as .driving-preview and .explainer-play: padding, border,
   background, the inherited-font exception, text-align. `display:
   block` matters as much as the rest — a button is inline-block by
   default, which adds a baseline gap under the video and makes the box
   taller than 16:9 asks for. */
.ctxai-shot {
  position: relative;
  display: block;
  width: 100%;
  aspect-ratio: 16 / 9;
  border-radius: 12px;
  overflow: hidden;
  background: #152C4A;
  padding: 0;
  border: 0;
  margin: 0;
  font: inherit;
  color: inherit;
  text-align: inherit;
  cursor: pointer;
  -webkit-appearance: none;
  appearance: none;
}
/* THE FOCUS RING IS NOT OPTIONAL. These are now controls, and the ring
   is the only way a keyboard user can tell they have reached one — the
   frame looks identical either way. :focus-visible so a mouse click
   leaves nothing behind; the offset puts it outside the rounded corner
   rather than on dark footage, where a dark ring would vanish.

   --accent is the ring the rest of the site uses. Measured on the
   #1F0D53 section ground it is the same ring the quote flow carries;
   these cards do not get their own. */
.ctxai-shot:focus-visible {
  outline: 3px solid var(--accent);
  outline-offset: 3px;
}
/* height: 100% AND object-fit: cover, both deliberate, both learned on
   4 Oct from the explainer's black bars. A <video> defaults to
   object-fit: contain, and the height="720" ATTRIBUTE beats
   aspect-ratio because the ratio only derives a height when the height
   is `auto` — together that gives a near-square box with the picture
   letterboxed inside it. The width/height attributes stay: they
   reserve the right box before the metadata loads, which is what stops
   the page jumping. TestVideosFillTheirBoxes guards this. */
.ctxai-video {
  position: absolute;
  inset: 0;
  width: 100%;
  height: 100%;
  object-fit: cover;
  /* The button is the control; the video must not swallow its clicks
     or become a second tab stop inside it. */
  pointer-events: none;
}
/* THE CHIPS PAINT ABOVE THE VIDEO, not only above the poster. This
   z-index is the rule that keeps the honesty note's first clause true
   while the picture is moving: without it the <video> is a later
   sibling in a positioned stacking context and covers them the moment
   it starts. */
.ctxai-chip {
  position: absolute;
  z-index: 2;
  left: 10px;
  bottom: 9px;
  background: rgba(11, 22, 38, 0.92);
  color: #FFF9F1;
  font-size: 0.625rem;
  padding: 3px 7px;
  border-radius: 4px;
  pointer-events: none;
}
/* .vid-chip--real and .ctxai-chip--real, the white "Real footage" chips
   at the top left of the hero card and the two scene cards, were DELETED
   11 Oct 2026 along with the markup. Guillaume: "it does not make much
   sense as is clearly real footage. We are adding nothing, it just looks
   a bit silly." The honesty claim that needed stating was always the
   hand-drawn ANALYSIS, which the remaining chip states. */
/* Duration AND the fact that the clip is SILENT, both before the
   click. The mirror of the explainer's "91 seconds · sound on": that
   one warns you it talks, this one tells you it will not, and either
   way the viewer knows what a click costs.

   BOTTOM RIGHT, NOT TOP RIGHT, and that was found by looking rather
   than by any script. At top right it sat ON TOP of the footage's
   OWN burned-in card — "Parked car blocking your lane" rendered as
   "Parked car b…" behind the badge on the second scene. Those cards
   are the hand-drawn analysis this section is about, so covering one
   is worse than untidy.

   The three other corners are each spoken for: top left is "Real
   footage", bottom left is "Planned service / Illustrative preview",
   top right is the footage's own card in BOTH clips. Bottom right is
   bonnet and road in both, and carries its own near-opaque ground
   anyway. */
.ctxai-badge {
  position: absolute;
  z-index: 2;
  right: 10px;
  bottom: 9px;
  background: rgba(255, 249, 241, 0.94);
  color: #142B49;
  font-size: 0.6875rem;
  font-weight: 700;
  line-height: 1.3;
  padding: 4px 7px;
  border-radius: 5px;
  pointer-events: none;
}
/* IT LIFTS ABOVE THE HONESTY CHIP WHEN THE CARD IS TOO NARROW FOR
   BOTH ON ONE LINE. Found 4 Oct 2026 by measuring the gap between
   the two at every width, then looking at the result:

     width   gap between chip right and badge left
       320   -70px     "Planned service / Illustrative p…"
       360   -30px     "Planned service / Illustrative pre…"
       393    +3px
       480   +90px
       768   -14px     "Planned service / Illustrative previe…"
       896   +45px
      1280  +192px

   Both are absolutely positioned on the same bottom edge, so where
   the card is narrower than chip (185px) + badge (107px) + gutters
   they simply overlap — and the badge, painted later, won. 360 is a
   very common phone width and 768 is the one-column-to-two-column
   handover, where each card is briefly at its narrowest.

   WHAT WAS COVERED IS THE WHOLE PROBLEM. The canvas note is explicit
   that "Planned service / Illustrative preview" is the ONLY thing on
   screen, once the clip loops, saying the boxes came out of Blender
   rather than a detector — so a duration badge truncating it is the
   one collision on this card that is not merely untidy. The chip
   that must never be obscured was being obscured by the chip that
   merely says how long the clip is.

   The badge moves rather than the chip because the chip's position
   is the board's and because the badge is the dispensable one: it
   disappears entirely on play, where the chip must not.

   UP, NOT ACROSS. Top right is the footage's OWN burned-in card in
   both clips — the comment on .ctxai-badge above records that being
   measured — so the badge sits immediately above the honesty chip
   instead, which is bonnet in both clips.

   A CONTAINER QUERY, NOT A WIDTH MEDIA QUERY, because what decides
   the collision is how wide THIS CARD is and the viewport does not
   predict it. Measured card width against the gap:

     viewport  card   gap
        320    242   -185 (lifted)
        360    282   -185 (lifted)
        393    315     +3
        480    402    +90
        700    591   +279
        768    298    -14   <- the grid goes two-column here and the
        800    313      0      card is SUDDENLY NARROWER than it was
        860    340    +28      at 700, which no viewport threshold
        896    357    +45      expresses
       1280    504   +192

   768 is the proof that a media query is the wrong tool: the card is
   591px at a 700px viewport and 298px at a 768px one, because the
   scenarios grid flips to two columns. A rule keyed to the viewport
   would have to encode that flip and would break the next time the
   grid's minmax changes. The container query just asks the card.

   THE THRESHOLD IS MEASURED, NOT GUESSED. The gap crosses zero at a
   card width of 313px (viewport 800 above). 20rem/320px is that
   measured crossing plus a few px of headroom, so the lift engages
   slightly before the two would touch rather than after. */
.ctxai-shot { container-type: inline-size; }
@container (max-width: 20rem) {
  .ctxai-badge {
    right: auto;
    left: 10px;
    bottom: 37px;
  }
}
/* THE PLAY CUE — the same device as the explainer's, scaled for a
   smaller card, so the two read as the same kind of thing. A triangle
   built from borders rather than an icon file: one shape, it must
   never fail to load, and the CSP allows no CDN.

   THE 3px --ink RING IS NOT DECORATION and the explainer learned it
   the hard way on 4 Oct: --cta orange over an orange frame measured
   1.03:1 and the circle effectively vanished, leaving only the dark
   triangle to read as a control. WCAG 1.4.11 asks 3:1 for the
   boundary of a non-text control. These frames are leafy green and
   grey rather than orange, but the ring costs nothing and a future
   re-cut changes the frame with nothing here to warn us. */
.ctxai-cue {
  position: absolute;
  z-index: 2;
  left: 50%;
  top: 50%;
  transform: translate(-50%, -50%);
  width: 56px;
  height: 56px;
  border-radius: 50%;
  background: var(--cta);
  box-shadow: 0 0 0 3px var(--ink);
  pointer-events: none;
}
.ctxai-cue::after {
  content: "";
  position: absolute;
  left: 50%;
  top: 50%;
  transform: translate(-40%, -50%);
  width: 0;
  height: 0;
  border-left: 18px solid var(--ink);
  border-top: 11px solid transparent;
  border-bottom: 11px solid transparent;
}
/* WHILE IT PLAYS, the cue and the duration badge go — both describe a
   click that has already happened — and THE TWO CHIPS STAY. That
   asymmetry is the whole point and it is the canvas's red note: "Real
   footage" is true of the picture and "Planned service / Illustrative
   preview" is true of the analysis drawn on it BY HAND, and the
   second must not vanish the moment the thing starts moving.

   THE CLIPS LOOP (Guillaume, 4 Oct: "looping is cool as well"),
   which is exactly why the chips must not be in this rule. A clip
   that repeats indefinitely means the "Planned service / Illustrative
   preview" chip is the ONLY thing on screen, for as long as anyone
   watches, saying the boxes came out of Blender rather than a
   detector. Adding .ctxai-chip to the selector below would be the
   single most damaging edit anyone could make to this section.

   flow.js adds .is-playing on the click and removes it on a second
   click, so this state lasts as long as the viewer wants it to. */
.ctxai-shot.is-playing .ctxai-cue,
.ctxai-shot.is-playing .ctxai-badge { display: none; }

/* ===================================================================
   THE EXPANDED STATE IS GONE, AND THAT IS WHAT MAKES THESE TWO CARDS
   SAFE RATHER THAN WHAT COSTS THEM SOMETHING.
   ===================================================================

   Removed 5 Oct 2026. Guillaume, trapped in fullscreen on his phone:
   "ok nothing full screen". Fourteen rules lived here — the
   fullscreened container, the video's object-fit, both chips enlarged
   and reseated, and the badge suppressed — each duplicated for
   :fullscreen and :-webkit-full-screen because an unrecognised
   selector invalidates a whole selector list.

   WHY THE DELETION IS THE FIX AND NOT A LOSS. The honesty constraint
   on this section is that the chip stays visible for as long as the
   clip runs: "Planned service / Illustrative preview" is true of the
   boxes, which were placed by hand in Blender. (A second chip,
   "Real footage", was removed as redundant on 11 Oct 2026; the
   measurement below was taken while both were present.) Native video fullscreen paints the <video> above
   everything and leaves positioned siblings unrendered — measured
   4 Oct at 1280x800, the video filled 0,0 1280x800 while the chips
   sat at y=2725 and y=2968 with elementFromPoint returning null at
   both centres, and every DOM assertion still passed.

   That is why the fullscreen target was the BUTTON: it put the chips
   inside the fullscreened element so these rules could enlarge them
   with it. And iOS Safari cannot fullscreen a container at all — it
   exposes webkitEnterFullscreen on the <video> alone — so on iOS the
   choice was inline-with-chips or fullscreen-without-them.

   WITH NO FULLSCREEN STATE THERE IS NOTHING TO CHOOSE. The chips
   cannot be orphaned, because the layer that orphaned them is never
   created. One state fewer is one state that cannot be got wrong.

   THESE TWO VIDEOS ALSO CARRY NO NATIVE CONTROL BAR, and they are the
   only videos on the page without one. Measured 5 Oct in WebKit with
   the iOS API surface: once this clip's metadata has loaded,
   video.webkitSupportsFullscreen is TRUE, so a bar would offer the
   viewer a full-screen button and put the <video> into the fullscreen
   layer — reaching the forbidden state by a route WE offered. The
   card stays its own control: one activation plays, a second stops.
   flow.js holds that measurement beside the code.

   The rules are deleted rather than commented out. A rule for a state
   that cannot occur is indistinguishable from a rule that matters,
   which is how a stylesheet stops describing the page. */

/* REDUCED MOTION. Nothing plays, so nothing should advertise that it
   will: the cue and the badge go and the poster is simply a picture
   again. The DOWNLOAD is stopped in the markup, not here — both
   <source> elements carry media="(prefers-reduced-motion:
   no-preference)", and a source whose media query does not match is
   never selected. The two halves are separate on purpose: this one
   guarantees nothing is OFFERED, that one guarantees nothing is
   FETCHED, and neither does the other's job.

   The chips stay in this state too. They are true of the still. */
@media (prefers-reduced-motion: reduce) {
  .ctxai-cue,
  .ctxai-badge { display: none; }
  .ctxai-shot { cursor: default; }
}
.ctxai-situation { margin: 1rem 0 0.25rem; font-size: 1.3125rem; font-weight: 700; }
.ctxai-detail { margin: 0 0 1rem; font-size: 0.9375rem; color: rgba(255, 249, 241, 0.72); }
/* BADGES ON RULES, NOT FILLED PANELS. The board's structure, restored
   4 Oct 2026 — Guillaume, looking at the running page: "the colour of
   the black box vs our ai does not match".

   What was here until today:

     .ctxai-sees--box { background: rgba(0, 0, 0, 0.22); }
     .ctxai-sees--ai  { background: rgba(255, 170, 102, 0.14); }

   That is a structural departure from the board dressed up as a
   colour choice. The board puts #FFAA66 SOLID in a 28px badge; the
   page spread the same hue at 14% opacity across a whole block, which
   over the #1F0D53 section ground computes to #3E2356, against the
   black-box row's #180A41. MEASURED: those two fills contrast with
   each other at 1.36:1 — two dark purple rectangles, which is
   exactly what Guillaume saw. The section exists to contrast "sees
   nothing" with "sees the situation", and a 14% tint cannot carry
   that. A small saturated badge can.

   NO BACKGROUND, NO BORDER-RADIUS, NO margin-bottom. The separation
   is border-top on each row, so two adjacent rows read as one list
   rather than two cards — and the spacing comes from the row's own
   symmetric padding rather than a margin on one side of it (R7).

   border-top on BOTH rows is deliberate and matches the board: the
   rule above the first row closes the gap under the scene's detail
   line, which is what makes the pair look like a table of two
   readings rather than two stray paragraphs. */
.ctxai-sees {
  margin: 0;
  display: flex;
  gap: 0.75rem;
  align-items: flex-start;
  padding: 0.75rem 0;
  border-top: 1px solid rgba(255, 255, 255, 0.10);
  font-size: 0.9375rem;
  line-height: 1.5;
}
/* THE ICON BADGE. `flex: none` so it keeps its 28px when the text
   beside it wraps — without it the badge is squeezed to an oval on a
   narrow card, which is what a flex item does by default.

   THE GLYPH IS CENTRED BY FLEX, not by line-height: a dash and a tick
   have very different ink heights and no single line-height centres
   both. Both glyphs are aria-hidden in the markup — see index.html.

   COLOURS ARE THE BOARD'S, SOLID, AND BOTH CARRY #1F0D53 INK.
   Measured (WCAG relative luminance, the card ground being
   rgba(255,255,255,0.05) over #1F0D53 = #2A195C):
     #9C93B5 badge — glyph 5.85:1, badge against card 5.21:1
     #FFAA66 badge — glyph 9.02:1, badge against card 8.04:1
   Both clear 4.5:1 for the glyph (1.4.3) and both clear 3:1 for the
   badge's own boundary (1.4.11) — which is the point of moving the
   colour out of a 14% wash and into something saturated. */
.ctxai-mark {
  flex: none;
  width: 28px;
  height: 28px;
  border-radius: 8px;
  display: flex;
  align-items: center;
  justify-content: center;
  font-size: 0.9375rem;
  font-weight: 800;
  color: #0B1626;
}
.ctxai-mark--box { background: #8FA2B4; }
.ctxai-mark--ai { background: #FFAA66; }
.ctxai-said { display: block; }
.ctxai-who {
  display: block;
  font-size: 0.6875rem;
  letter-spacing: 0.06em;
  text-transform: uppercase;
  font-weight: 700;
  color: #AFC2D4;
  margin-bottom: 0.25rem;
}
.ctxai-what { display: block; }

/* THE COMPARISON SITS BESIDE ITS ARGUMENT, not under it. The board's
   grid, chosen by Guillaume on 5 Oct 2026: minmax(0,1fr)
   minmax(0,1.6fr) with a 40px gap, vertically centred, so the
   heading and its line read next to the table rather than above it.

   minmax(0, …) on BOTH tracks is load-bearing. A bare 1fr has a
   min-content floor, and the table's longest cell ("Looks at the
   road, weather and traffic") would push the right track past its
   share and overflow the section. minmax(0, …) lets both tracks
   shrink to what the grid gives them.

   ONE COLUMN BELOW 56rem. Two columns of nothing on a phone is the
   fault the layout contract was written for, and the board's phone
   view stacks them. 56rem is an existing breakpoint; no new one. */
.ctxai-compare {
  display: grid;
  grid-template-columns: minmax(0, 1fr);
  gap: 1.5rem;
  max-width: 56rem;
  margin-inline: auto;
}
@media (min-width: 56rem) {
  .ctxai-compare {
    grid-template-columns: minmax(0, 1fr) minmax(0, 1.6fr);
    gap: 2.5rem;
    align-items: center;
  }
}

/* 26px, and deliberately SMALLER than the five section headings at
   48px. This introduces a comparison inside the Contextual AI
   section; it is not a section heading of its own, which is why the
   markup calls it .ctxai-sub and uses an <h3>. The board agrees. */
.ctxai-sub {
  margin: 0 0 0.625rem;
  font-size: 1.625rem;
  line-height: 1.2;
  font-weight: 800;
  letter-spacing: -0.01em;
}
.ctxai-subline {
  margin: 0;
  font-size: 1rem;
  line-height: 1.6;
  color: rgba(255, 249, 241, 0.82);
}
/* A REAL TABLE, because it is a real comparison: two columns of
   claims, read across. A pair of <ul>s would lose the pairing for
   anyone not seeing the layout. */
/* PADDING IN px, NOT rem, SO 200% TEXT DOES NOT OVERFLOW. At a 32px
   root the 1.125rem side padding became 36px per side -- 144px of
   padding across two columns on a 360px phone -- and the table
   scrolled the page sideways by 58px (WCAG 1.4.10 reflow). The text
   itself still scales; only the gutters are pinned. Measured 4 Oct:
   scrollWidth 418 vs clientWidth 360 before, equal after. */
.ctxai-table {
  width: 100%;
  max-width: 56rem;
  margin-inline: auto;
  border-collapse: collapse;
  background: rgba(255, 255, 255, 0.05);
  border: 1px solid rgba(255, 255, 255, 0.12);
  border-radius: 16px;
  overflow: hidden;
}
.ctxai-table th {
  text-align: left;
  font-size: 0.75rem;
  letter-spacing: 0.06em;
  text-transform: uppercase;
  color: #AFC2D4;
  padding: 14px 18px;
  border-bottom: 1px solid rgba(255, 255, 255, 0.12);
  background: rgba(0, 0, 0, 0.18);
}
.ctxai-table td {
  padding: 14px 18px;
  border-bottom: 1px solid rgba(255, 255, 255, 0.1);
  font-size: 1rem;
  vertical-align: top;
}
/* THE LAST PIXEL, and why it needs table-layout rather than a wrap
   rule. The px gutters above took this table from 58px of sideways
   scroll to 1px. The remainder is the table's MIN-CONTENT width,
   which `width: 100%` cannot go below: measured at 200% on a 360px
   phone it was 321px against 280px of available column, because
   `table-layout: auto` sizes every column to its widest word and the
   two worst here are "braking" (155px) and "junctions" (163px).

   `overflow-wrap` alone does NOT fix that — auto layout computes the
   column floor from the longest word before any wrapping rule gets a
   say, so the table stays 321px wide and simply overflows.

   `table-layout: fixed` makes the two columns split the width they
   are given instead of demanding what their content wants, and the
   overflow-wrap below then lets a too-long word break inside its
   cell. Two columns of equal width is what this table already looks
   like at every width, so nothing moves except in the failing case.

   The alternative was wrapping the table in an overflow-x: auto
   container, which the layout contract does allow for tables. It is
   not used here because it trades a 1px page overflow for a
   horizontally scrolling region inside the page, and this table only
   needs its columns to stop over-asking.

   Measured after: scrollWidth 360 = clientWidth 360, overflow 0. */
.ctxai-table { table-layout: fixed; }
.ctxai-table th,
.ctxai-table td { overflow-wrap: break-word; }
.ctxai-table tr:last-child td { border-bottom: 0; }
.ctxai-table td:first-child { color: rgba(255, 249, 241, 0.7); }

/* THE "OUR AI" COLUMN IS TINTED, HEAD TO FOOT. The board's structure,
   restored 4 Oct 2026 — Guillaume, looking at the running page: "this
   is not the same".

   He was right, and it is the SAME fault as the badges two rules up:
   the board's orange had been dropped and both columns given
   identical treatment, so the table read as a plain two-column list.
   A comparison table whose two sides look the same is not making a
   comparison.

   The board tints the whole right column rgba(255,170,102,0.08) and
   colours its header #FFAA66, against #B9AEDB on the left. Measured
   over the #1F0D53 section ground the tint computes to #2E1A54
   against the untinted #1F0D53 — a 1.27:1 difference, which is
   deliberately a WASH and not a boundary: it groups the column rather
   than boxing it, and the argument is carried by the orange HEADER
   (#FFAA66 on #2E1A54 = 7.9:1) and the brighter cell ink, both of
   which clear 4.5:1 on their own. The tint is the quietest part and
   nothing depends on seeing it.

   The header's 12px 12px 0 0 radius is the board's: the table clips
   to a 16px radius, so a square-cornered tinted cell would show a
   corner of tint outside the curve. */
.ctxai-table th:last-child {
  color: #FFAA66;
  background: rgba(255, 170, 102, 0.08);
  border-radius: 12px 12px 0 0;
}
.ctxai-table td:last-child {
  font-weight: 600;
  background: rgba(255, 170, 102, 0.08);
}

@media (max-width: 47.9375rem) {
  .ctxai { padding-block: 3rem; }
  /* px for the same reason as the base rule above: at a 32px root
     these rem gutters were what pushed the table past the viewport. */
  .ctxai-table th,
  .ctxai-table td { padding: 12px 14px; font-size: 0.9375rem; }
  /* A SMALLER CUE ON THE PHONE, because HomePhone.dc.html draws one:
     48px there against 56px on the desktop board. The card is the
     full column width on a phone rather than half of it, but the cue
     is a fixed pixel size, so at 56px it took visibly more of the
     frame than the board allows. Caught by comparing the device
     render against the phone board, not by any script.

     STILL WELL OVER THE 44px TAP TARGET — and the target is the
     BUTTON, which is the whole 16:9 card, not this cue. The cue is
     pointer-events: none decoration. */
  .ctxai-cue { width: 48px; height: 48px; }
  .ctxai-cue::after {
    border-left-width: 15px;
    border-top-width: 9px;
    border-bottom-width: 9px;
  }
}


/* ---- Ready when you are: two ways in ----------------------------- */

/* The board ends the page with the quote beside the waitlist rather
   than the waitlist alone. The two cards are equal width and equal
   height; only the button fill differs, and that difference is the
   whole point — the quote always works, the waitlist may be closed. */
/* ONE SECTION-HEADING SIZE, 5 Oct 2026 (Guillaume) — --section-h,
   plus the weight and tracking that go with it. This rule used to set
   30px/700/-0.015em, which made it the smallest and lightest of the
   five section headings; the boards draw all five identically at
   48/800/-0.02em/1.1. */
.ready-h {
  margin: 0 0 1.5rem;
  text-align: center;
  font-size: var(--section-h);
  font-weight: 800;
  letter-spacing: -0.02em;
  line-height: 1.1;
}
.ready-grid {
  max-width: 62.5rem;
  margin-inline: auto;
  display: grid;
  gap: 1.5rem;
  grid-template-columns: repeat(auto-fit, minmax(min(20rem, 100%), 1fr));
  align-items: stretch;
}
.ready-card {
  background: var(--paper);
  border: 1px solid var(--line);
  border-radius: 20px;
  /* THE SIDE PADDING STOPS GROWING WITH THE TEXT. `rem` scales with
     the user's font size, so at 200% this 1.625rem became 52px a side
     — 104px of a 280px card, before .consent took another 72px inside
     it. Three nested rem paddings compounded until the consent label
     had 23px to wrap 175px of words, and the page scrolled sideways.

     `min(1.625rem, 5vw)` keeps the designed 26px at every normal
     width (5vw only bites below 520px) and caps it at large text,
     where vw does not scale. The vertical padding is untouched: tall
     is free, wide is not. */
  padding: 1.625rem min(1.625rem, 5vw);
  display: flex;
  flex-direction: column;
}
/* .signup h2 (line 187) sets 1.375rem and a 2rem TOP margin, for the
   standalone waitlist page where that heading opens the section. In
   the Ready card it is one of two titles side by side, and that rule
   made them 24px vs 22px and 32px apart vertically — Guillaume,
   5 Oct 2026: "things don't seem aligned here".

   Two class selectors so this beats .signup h2 (0-1-1) without
   !important and without touching the standalone page. The <h2>
   stays an <h2>: the outline is right, only the look was wrong. */
.ready-card-h,
.ready-card .ready-card-h { margin: 0 0 0.625rem; font-size: 1.5rem; font-weight: 700; }
.ready-card-body {
  margin: 0 0 1.375rem;
  font-size: 1rem;
  line-height: 1.55;
  color: var(--ink-soft);
}
/* "Be first in line when we launch in 2027." — item 6, 4 Oct 2026.
   The waitlist card's equivalent of .ready-card-body, and styled from
   it rather than inventing a second scale: same size, same muted ink,
   16px below instead of 22px because the field label follows it
   directly where the quote card's button is further down. Both boards
   draw it at the same weight as the body copy beside it. */
.signup-promise {
  margin: 0 0 1rem;
  font-size: 1rem;
  line-height: 1.55;
  color: var(--ink-soft);
}
/* margin-top: auto pins the button to the bottom, so two cards with
   different amounts of copy still line their buttons up.

   align-self: start IS REQUIRED WITH IT. The card is a flex column,
   whose default align-items: stretch makes a child fill the cross
   axis — and .action is inline-flex, so it stretched into a 200px
   tall orange slab. Measured before the fix: 276x200 where the
   waitlist button beside it is 59px.

   flex: none stops the same thing happening on the main axis when
   the other card is taller. */
/* FULL WIDTH OF THE CARD, per the board and Guillaume on 5 Oct 2026.
   The board's markup is `display: block` on the <a>, which fills the
   card because a block-level element takes its container's width.
   The page had align-self: start, which shrank each button to its
   label: measured 267px and 224px inside a 436px column, so the two
   cards' buttons were different widths and neither reached the edge.

   align-self: stretch, not the removal of align-self. The card is a
   flex COLUMN, so the cross axis is horizontal and stretch is what
   fills it. The comment that used to sit here warned that stretch on
   the MAIN axis made .action a 200px-tall slab — that was
   align-items on the row, a different axis, and flex: none below
   still prevents it. Measured after: both buttons 436px, height
   59.44px, unchanged. */
.ready-cta {
  margin-top: auto;
  align-self: stretch;
  flex: none;
  justify-content: center;
}
/* The waitlist submit is a bare <button type="submit"> with no class
   of its own, so it needs its own selector to match. Same rule, same
   reason: the board draws both cards' buttons edge to edge, and one
   full-width button beside a shrink-wrapped one looks like a mistake
   rather than a hierarchy. Measured before: 434px and 224px. */
.signup.ready-card button[type="submit"] {
  width: 100%;
}

/* THE WAITLIST BUTTON IS SECONDARY, and the board is explicit about
   why: "two orange buttons don't compete". The quote keeps the
   orange because it is the one action that always works — the
   waitlist may be closed, as it is now.

   Paper fill with a brand edge, which is the same pairing the quote
   flow's decline button uses. Same size, same weight, same padding
   as the primary: only the fill differs, so the hierarchy is a
   colour decision rather than a size one. */
.signup.ready-card button[type="submit"] {
  background: var(--paper);
  color: var(--ink);
  border: 2px solid var(--brand);
}

/* ---- The differentiator, and the statement panel ------------------ */

.differentiator-band { background: var(--page); }

/* ===================================================================
   "YOUR DRIVING. NOT YOUR MARITAL STATUS." — item 5, 4 Oct 2026
   ===================================================================

   RESTYLED on Guillaume's "strangely formatted". What was here, and
   is deleted rather than overridden:

     - a lilac .statement-bg BOX with 32px padding and a 20px radius,
       sitting hard against the dark Contextual AI section above it;
     - LEFT-ALIGNED copy inside it;
     - a two-column grid at 52rem that put the launch caveat in a
       narrow side column beside the heading;
     - a 3px --accent left border on that caveat, which gave a hedge
       more visual weight than the claim it qualifies.

   THE 52rem BREAKPOINT WENT WITH IT. It was a fifth width in a
   contract that allows four, and it existed only to make that side
   column work. Nothing here needs it: the card row wraps on its own.
   So this change removes a breakpoint rather than adding one.

   NOW: no box at all, on the page ground, centred, with real room
   above it. Both boards draw 80px of top padding on desktop and 52px
   on phone — the "space above" the spec asks for is what separates
   this from the dark band and is the reason it no longer reads as
   jammed against it. */
.differentiator-band {
  padding-block: 3.25rem 2.5rem;
  text-align: center;
}
@media (min-width: 61rem) {
  .differentiator-band { padding-block: 5rem 4.5rem; }
}

/* ONE SECTION-HEADING SIZE, 5 Oct 2026 (Guillaume) — --section-h,
   48px at 1280 and 32px at 390, shared with the other four section
   headings. This rule used to set 40px/28px, which is what the boards
   drew before the normalisation note was added to both of them.
   line-height follows them to 1.1, from 1.12. */
.differentiator-h {
  margin: 0 0 0.75rem;
  font-size: var(--section-h);
  line-height: 1.1;
  font-weight: 800;
  letter-spacing: -0.02em;
  text-wrap: balance;
}
/* THE HIGHLIGHTER on "Not your marital status.", 5 Oct 2026
   (Guillaume) — the same felt-tip stroke .portal-h-mark carries on
   "Pay less", and both boards now draw it on this second sentence too.

   The stops are the boards', copied rather than approximated:
   transparent -> 58%, #FFCB9A 58% -> 92%, transparent 92%.

   TWO DIFFERENCES FROM .portal-h-mark, both deliberate and both the
   boards':

   1. box-decoration-break: clone, which "Pay less" does not need.
      "Not your marital status." is four words and DOES wrap — at 390
      it takes two lines. The default `slice` paints the gradient
      across the union of the line boxes, so the stroke runs through
      the gap and past the end of the short line; `clone` repaints it
      per line, so each line gets its own stroke that stops with its
      own text. Both boards set it explicitly for this reason.

   2. NO white-space: nowrap. "Pay less" is held on one line because
      it is short enough to never force overflow; forbidding a wrap on
      a 24-character phrase would push the page sideways on a narrow
      screen, which is the 188px overflow this section caused before.
      So it wraps, and `clone` is what makes a wrap look right.

   COLOUR IS PINNED TO --brand, NOT --accent, and that is a change
   from the plain-purple version this replaces. The mark is a
   self-contained pair — purple on peach — and does not sit on the
   page ground, so it must not follow a token that lightens for a dark
   page: on #FFCB9A a lightened purple loses contrast. Measured, the
   same pair .portal-h-mark records:
     light  --brand #6624C9 on #FFCB9A   5.36:1
     dark   --brand #5B21B6 on #FFCB9A   6.10:1
   Both clear 4.5:1 in both themes. */
/* COLOUR ONLY — see .portal-h-mark for the reasoning. This heading
   is the one that made the case: "Not your marital status." wraps to
   two lines on a phone, and box-decoration-break: clone gave each
   line its own stripe, which is worse than one badly-placed band
   rather than better.

   The clone property goes with the band it existed for. */
.differentiator-h-accent {
  /* --accent, NOT --brand. This band follows the theme (its fill is
     var(--page)), and --brand DARKENS to #5B21B6 in dark mode, so the
     accent converged on the near-black ground behind it: 2.06:1 for a
     32px heading that needs 3:1. --accent is #6624C9 in light —
     byte-identical to --brand, so the light appearance is unchanged —
     and lightens to #D8C4F5 in dark.

     The same convergence made the phone progress bar invisible
     earlier today (1.15:1) and was fixed the same way; .portal and
     .explainer have the mirror-image bug, where the FILL is pinned
     light and the ink had to be pinned with it. Found on Guillaume's
     phone, 6 Oct 2026. */
  color: var(--accent);
}

.differentiator-lede {
  margin: 0 auto 1.5rem;
  max-width: 46ch;
  font-size: 1.125rem;
  line-height: 1.5;
  color: var(--ink-soft, var(--ink));
}

/* THE THREE REFUSED QUESTIONS.

   A <ul>, because it is a list of three things and a screen reader
   should say so — the boards draw three boxes in a row, which is a
   list with its bullets replaced by cards. list-style is removed
   rather than the element being changed to divs.

   flex-wrap, NOT a grid with a fixed count. On desktop the three sit
   in a row; as the column narrows they wrap to two then one, and each
   card is only as wide as its question. A three-column grid would
   either squeeze "Do you own your home?" to two lines at 390px or
   need its own breakpoint. The phone board stacks them full width and
   left-aligns the text inside, which is what the 61rem block below
   does.

   justify-content: center is what keeps a wrapped row centred under
   the heading — the fault the design-canvas skill records for 3 Oct
   was a box jammed left in a centred section, and a wrapped flex row
   defaults to flex-start. */
.unasked {
  list-style: none;
  margin: 0 0 1.375rem;
  padding: 0;
  display: flex;
  flex-wrap: wrap;
  justify-content: center;
  gap: 0.875rem;
}
.unasked-item {
  display: flex;
  align-items: center;
  gap: 0.625rem;
  background: var(--paper);
  border: 1px solid var(--line);
  border-radius: 14px;
  padding: 0.875rem 1.25rem 0.875rem 0.875rem;
  font-size: 1.0625rem;
  font-weight: 600;
  color: var(--ink-soft, var(--ink));
  text-align: left;
}
/* The strikethrough. 2px so it reads at this weight, and in a muted
   purple rather than the text colour so the question stays readable
   THROUGH it — a full-contrast line through 17px text makes the words
   hard to parse, and the point is that you recognise the question. */
.unasked-item s {
  text-decoration-thickness: 2px;
  text-decoration-color: rgba(64, 86, 107, 0.55);
}
/* The cross. Decorative — <s> carries the meaning — but it still has
   to be visible, so it is measured: #8C1D1D on #FDECEC is 7.96:1. A
   red ONLY appears here, on a glyph that duplicates the
   strikethrough, so it is not the sole carrier of anything. */
.unasked-x {
  flex: none;
  display: flex;
  align-items: center;
  justify-content: center;
  width: 1.625rem;
  height: 1.625rem;
  border-radius: 50%;
  background: var(--err-bg);
  color: var(--err-ink);
  font-size: 0.9375rem;
  font-weight: 800;
  line-height: 1;
}
/* .differentiator-note WAS HERE AND IS DELETED, 5 Oct 2026
   (Guillaume) — the caveat line under this section came off both
   boards the same day, so the element it styled no longer exists in
   index.html. Deleted rather than left behind: a rule for an element
   nothing renders is the kind of claim CLAUDE.md warns about, and the
   next person would reasonably read it as evidence the line is still
   drawn somewhere.

   The honesty line this is NOT is .ctxai-hedge, in the Contextual AI
   section, which stays. See the note in index.html where this
   paragraph used to be. */

/* PHONE: the cards stack full width and their text stays left, which
   is what HomePhone.dc.html draws — a centred question inside a
   left-aligned cross looks accidental. 61rem, the width the rest of
   the phone layout already uses. */
@media (max-width: 60.9375rem) {
  .unasked {
    flex-direction: column;
    gap: 0.625rem;
    text-align: left;
  }
  .unasked-item {
    padding: 0.75rem 1rem 0.75rem 0.75rem;
    font-size: 1rem;
  }
}

/* THE HEADING'S CLOSING FULL STOP IS PHONE-ONLY, and it is the boards
   that differ rather than this rule that is clever. Read this session
   from canvas v180:

     Home.dc.html       "Your driving. Not your marital status"   (none)
     HomePhone.dc.html  "Your driving. Not your marital status."  (one)

   At 48px on one line it is punctuation on a display heading; at 32px
   wrapped over two lines it closes the sentence.

   THE MARKUP HOLDS THE DESKTOP FORM AND THIS ADDS THE STOP, which is
   the opposite of the obvious build and the reason is measurable. A
   <span class="...-stop">.</span> hidden above 61rem renders
   correctly at both widths, but the character is in the MARKUP at
   every width, and every text-level comparison this project runs
   reads markup rather than pixels. Measured on the running page:
   scripts/verify-against-board.sh reported

     FAIL  Home — MISSING: your driving. not your marital status

   against the desktop board, and would have done so on every future
   run, for a page that was drawing the right thing. A permanently
   red check is a check nobody reads.

   GENERATED CONTENT IS ACCEPTABLE HERE AND WOULD NOT BE FOR A WORD.
   ::after content is unreliable in the accessibility tree and must
   never carry meaning. A trailing full stop on a display heading
   carries none: the heading reads the same with or without it, and
   the sentence is already closed by the one after "Your driving".
   If the two boards ever differ by a WORD, this becomes two headings
   and a harder problem — do not reach for ::after then.

   61rem is the width the rest of this section's phone rules already
   use; no new breakpoint. */
@media (max-width: 60.9375rem) {
  .differentiator-h-accent::after { content: "."; }
}

/* ===================================================================
   THE MISSION BUTTON — canvas v214, 13 Oct 2026
   ===================================================================

   Guillaume's note, identical on Home.dc.html and HomePhone.dc.html:
   "MISSION BUTTON moved under the title, 13 Oct (Guillaume). The
   'That's why we're building CoRoute' bridge line was removed with
   it."

   .mission-bridge, .mission-bridge-text AND .mission-bridge-cta WERE
   HERE AND ARE DELETED. They styled a wrapper with a border-top
   divider, a 20px closing sentence, and a .cta-sec-skinned 400x52
   button at the FOOT of the section. The button moved under the h2,
   grew, and became purple; the divider and the sentence went with the
   move. Nothing selects the old three classes, so they are gone
   rather than left as dead rules — see the .signup-band note below,
   which is the same lesson.

   IT IS A NEW CLASS AND NOT AN EXTENSION OF .cta-sec, deliberately.
   The old button borrowed .cta-sec's entire raised treatment. This one
   cannot: .cta-sec is a PAPER-filled secondary control with an
   --accent edge, and this is a solid purple fill with white text. More
   to the point, .cta-sec is also worn by .action-decline in the quote
   flow, and quote-flow-spec §7.7 requires the decline control to stay
   exactly as prominent as the join — restyling the shared class to
   fill this button the hero would recolour a control that section is
   written about. One instance's own treatment belongs on its own
   class.

   IT IS STILL NOT ORANGE, AND THAT IS STILL LOAD-BEARING, but
   "secondary" no longer describes it and the old comment here said
   so. What is true now: the hero's "Get your advisory quote"
   (.action-join) is the only ORANGE button on the page and the only
   CTA in the hero, and both boards keep it that way — the quote is
   the action that always works even while the waitlist is closed, so
   it owns the one colour nothing else uses. This button is a
   SECTION-LEVEL PRIMARY: it is the loudest control inside the
   marital-status section and it leads to a different page, which is a
   different job from the hero's, not a competing instance of it.
   Purple is the brand fill and appears nowhere else as a button, so
   the two do not read as a pair.
   TestHomepageHasExactlyOneOrangeCTA pins the orange count.

   EVERY NUMBER HERE IS THE BOARD'S, copied rather than approximated.
   Desktop, Home.dc.html: 440x64, radius 14px, 21px/800, margin
   28px 0 40px, gap 12px, arrow 24px. Phone, HomePhone.dc.html:
   width 100%, height 58px, 19px/800, margin 22px 0 28px, gap 10px,
   arrow 22px.

   THE FILL IS A LITERAL #6624C9, NOT var(--brand), AND THAT IS A
   DEPARTURE FROM THE BOARD'S DECLARATION, measured before choosing.
   Both boards write `background: var(--brand)` and both define
   --brand: #6624C9 with no dark theme at all (grep
   prefers-color-scheme over either board returns 0), so the board's
   token and the literal are the same pixel in the only theme the
   board draws. They diverge in app.css's dark mode, where --brand
   DEEPENS to #5B21B6. Measured, white on each:

     #FFFFFF on #6624C9   7.89:1
     #FFFFFF on #5B21B6   8.98:1

   Both pass AA comfortably, so the choice is not a contrast one. It
   is a boundary one. Measured against the page each theme paints:

     #6624C9 vs light --page #FFF9F1   7.54:1
     #5B21B6 vs dark  --page #180F27   2.06:1
     #6624C9 vs dark  --page #180F27   2.34:1

   A solid-filled button needs its edge to be findable against the
   page, and 3:1 is the floor for a control boundary (the --accent
   dark-mode note above this file's dark block states exactly that
   rule for the same reason). NEITHER purple clears it on the near-
   black dark page, so the boundary is carried by the relief slab and
   the white label rather than by the fill, and the fill should
   therefore be the LIGHTER of the two: #6624C9 at 2.34:1 is 14%
   further from the page than #5B21B6 at 2.06:1, and it keeps the
   button recognisably the same object in both themes, which a reader
   switching themes mid-session notices. Pinning the literal also
   makes the three shadow colours, the hover and the fill one
   consistent set: #3E1590 and #7535D6 are literals on the board with
   no token to track, and a tokenised fill under literal shadows
   drifts apart the next time --brand moves.

   THE DEPARTURE IS RECORDED RATHER THAN SILENT. It is a dark-mode
   decision, and the boards draw no dark mode, so it is not a layout
   decision taken in CSS — board-leads is not breached by a rule for
   a state the board does not model. The skill says so in as many
   words: dark mode is "yours, and the board's silence is not
   permission".

   THE FILL AND THE SHADOW ARE TOKENS SINCE V2, AND THEY HAD TO BE.
   This was `background: #6624C9; color: #FFFFFF` with a hard slab in
   a darker purple, and one pair of literals served both themes
   because the purple read against the cream page AND against the
   near-black one. The v2 brand does not: --brand is NAVY in light
   mode and a LIGHT BLUE in dark, so a hardcoded navy fill would be
   a navy button on a navy page in dark mode -- which is precisely
   the "Read our mission button is hard to see" fault Guillaume
   reported on 13 Oct and which the --sec-fill note further up this
   file was written about. The same button, the same bug, reached by
   a different route.

   So: --brand / --on-brand for the fill and the label (14.27:1
   light, 9.72:1 dark), and --brand-edge for the slab. The slab is a
   DARKER shade than the fill in both themes, so it reads as depth
   against the BUTTON rather than against the page -- #102341 vs the
   light page is 14.99:1 and vs the dark page 1.08:1, and the second
   figure is correct rather than a failure: it is the button's own
   underside, not a border.

   THE FOCUS RING IS TWO STACKED CONTOURS, and v2 needs both of them
   in both themes where v1 needed the second one only on cream. An
   outline-offset ring sits on the PAGE as much as on the control,
   so it has to be seen against both:

     light   --logo #E8762A vs the navy button    4.80:1  (passes 3:1)
             --logo #E8762A vs --page #FFF9F1     2.84:1  (short)
     dark    --logo #FFB264 vs the blue button    1.21:1  (fails)
             --logo #FFB264 vs --page #101D30     9.52:1  (passes)

   NEITHER THEME CLEARS 3:1 ON BOTH SURFACES WITH ONE COLOUR, and
   the two themes fail on OPPOSITE surfaces -- light fails against
   the page, dark fails against the button. That is why the ring is
   drawn as two stacked box-shadow contours rather than one outline:
   3px of --logo immediately outside the button, then 2px of --ink
   outside that. --ink is 13.64:1 on the light page and 14.21:1 on
   the dark one, so whichever surface the orange loses on, the ink
   hairline is what makes the ring findable. Under forced-colors the
   UA outline below takes over entirely and neither matters.

   `outline: 3px solid #FFB264` ALONE IS WHAT THE BOARD WRITES and is
   NOT what ships, for the reason measured above. This is the same
   kind of departure as the fill and is recorded for the same reason.

   min-height, NOT height, AND THE BOARD WRITES height. Same reason
   as the button this replaced: at 200% text a fixed height clips the
   label, and "no fixed pixel heights on content" is in the layout
   contract's list. At 100% text min-height renders the board's 64px
   exactly, which is what a side-by-side compares.

   61rem IS THE BREAKPOINT, not a new one. Every other phone rule in
   this section already uses it (.unasked, .differentiator-h-accent,
   .differentiator-band), and a 440px button cannot overflow above it:
   at the 976px crossover the band's gutters leave 896px of content
   box, so the button occupies 49% of it and stays centred. Below the
   crossover the phone rule takes it to 100% of the gutters. There is
   no width at which a 440px box and a 40px gutter pair exceed the
   viewport. */
.cta-mission {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  gap: 0.75rem;
  box-sizing: border-box;
  width: 27.5rem;
  min-height: 4rem;
  margin: 1.75rem 0 2.5rem;
  border: 0;
  border-radius: 14px;
  background: var(--brand);
  color: var(--on-brand);
  font-family: inherit;
  font-size: 1.3125rem;
  font-weight: 800;
  line-height: 1.2;
  text-align: center;
  text-decoration: none;
  box-shadow: 0 5px 0 var(--brand-edge), 0 14px 28px rgba(20, 43, 73, 0.28);
  transition: transform 0.12s ease, box-shadow 0.12s ease,
              background-color 0.12s ease;
}
/* The arrow is 24px on the desktop board against a 21px label. */
.cta-mission span { font-size: 1.5rem; line-height: 1; }

/* A filter rather than a second hardcoded fill: one declaration that
   lightens a dark brand and the dark-theme override that deepens a
   light one, instead of two more per-theme literals. */
.cta-mission:hover {
  filter: brightness(1.18);
  color: var(--on-brand);
  text-decoration: none;
  transform: translateY(-2px);
}
:root[data-theme="dark"] .cta-mission:hover { filter: brightness(0.92); }
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .cta-mission:hover { filter: brightness(0.92); }
}
/* The press takes the slab down with it, so the button looks pushed
   into the page rather than sliding over it: 3px of travel against a
   slab cut from 5px to 2px. The board's own two declarations. */
.cta-mission:active {
  transform: translateY(3px);
  box-shadow: 0 2px 0 var(--brand-edge), 0 6px 14px rgba(20, 43, 73, 0.28);
}
/* Two stacked rings, measured above: the brand orange against the
   button, ink against the page. The slab is restated because
   box-shadow replaces rather than adds, and losing the relief on
   focus would make the button change shape when tabbed to. */
.cta-mission:focus-visible {
  outline: 3px solid transparent;  /* forced-colors paints this one */
  outline-offset: 3px;
  box-shadow: 0 5px 0 var(--brand-edge), 0 14px 28px rgba(20, 43, 73, 0.28),
              0 0 0 3px var(--logo), 0 0 0 5px var(--ink);
}
@media (prefers-reduced-motion: reduce) {
  /* The TRAVEL goes and the relief stays, which is the carve-out
     .cta-sec and .cta-btn already make: the slab is depth, not
     motion, and removing it would change what the control looks like
     rather than how it animates. */
  .cta-mission,
  .cta-mission:hover,
  .cta-mission:active { transition: none; transform: none; }
  .cta-mission:active {
    box-shadow: 0 2px 0 var(--brand-edge), 0 6px 14px rgba(20, 43, 73, 0.28);
  }
}
/* PHONE: full width to the section's own 20px gutters, which is what
   HomePhone.dc.html draws — at 390 the content box is 350px and the
   board's button measures exactly 350.

   display:flex NOT inline-flex, also the board's: an inline-flex box
   at width:100% still sits in a line box, and the board switches it
   to block-level flex.

   THE WIDTH IS min(440px, 100%) AND THE BOARD WRITES 100%. The cap
   does not change either width the boards draw, measured:

     390  content box 350px   min(440, 350)  = 350   board: 350
     1280 content box 1104px  min(440, 1104) = 440   board: 440

   It changes only the TABLET RANGE, which neither board draws. The
   media query runs to 60.9375rem (975px), and width:100% there made
   the button 897px wide — a 900px-long tap target for a two-word
   label, and the one control on the page that behaved that way: the
   hero's "Get your advisory quote" already uses
   `width: min(400px, 100%)` (.homehero .cta-hero) for the same
   reason and caps at its board width of 400px from 414px up. Leaving
   this one uncapped made the SECOND button on the page twice the
   width of the FIRST at 900px, which is a worse disagreement with
   the boards' intent than a cap the boards do not mention.

   The old bridge button capped at max-width 320px. That was ITS
   board value and is not carried over; 440px is this button's. The
   mechanism is the hero's, the number is this board's.

   THIS IS A JUDGEMENT THE BOARDS DO NOT COVER and is reported as
   such rather than presented as the board's. If Guillaume wants the
   button to run the full width of a tablet, the cap is the one line
   to delete — and it should then be deleted on the board too, where
   a tablet board does not exist to say either way. */
@media (max-width: 60.9375rem) {
  .cta-mission {
    display: flex;
    gap: 0.625rem;
    width: min(27.5rem, 100%);
    min-height: 3.625rem;
    /* margin-inline AUTO, NOT 0, AND THE BOARD WRITES 0. The board
       can write 0 because at 390 the button fills its content box
       exactly, so there is nothing to centre. The cap above means
       that stops being true from ~460px up, and `display: flex` makes
       this a BLOCK-LEVEL box, which the section's `text-align:
       center` cannot centre — a text-align centres inline content,
       not a block. Measured before the auto was added: at 975px the
       gutters came out 39px left and 496px right, the button jammed
       hard left under a centred heading, which is the 3 Oct explainer
       fault exactly (a base rule overriding `margin-inline: auto`,
       found by looking rather than by any script — nothing here
       asserts "centred"). With the auto: 267/267 at 975. */
    margin: 1.375rem auto 1.75rem;
    font-size: 1.1875rem;
  }
  .cta-mission span { font-size: 1.375rem; }
}

/* `.signup-band { background: var(--page); padding-bottom: 2rem }`
   WAS HERE AND IS DELETED, 10 Oct 2026.

   It was the last surviving rule of the "Ready when you are" section
   at the foot of the page, which was removed earlier today when the
   waitlist moved into the hero (56a36cd). The markup went; this one
   line did not, and nothing selected it — an orphaned rule for an
   element that no longer exists, which is the stale-claim failure
   CLAUDE.md is written about.

   IT WAS NOT HARMLESS, AND THAT IS WHY THIS NOTE IS LONGER THAN THE
   RULE IT RECORDS. Option A, built and then withdrawn the same
   evening, introduced a new band that reused the class name
   .signup-band — and this dead rule came back to life the moment
   that markup existed: measured on the built page, it added
   `padding-bottom: 2rem` on top of the band's own symmetric 44px,
   giving 44px top against 76px bottom. A band that looked
   bottom-heavy for no reason anybody could have found in the new
   rules. Found by measuring, not by reading the stylesheet.

   Option B has no band at all, so nothing selects this name today.
   The note stays because the hazard is general and this is the second
   time in one evening a dead rule has cost something: a class name
   outliving its markup is a trap for whoever next reaches for that
   name. See also `order: 7` in the phone block further up, which was
   the same accident by a different route. */


/* ---- The footer --------------------------------------------------- */

.sitefooter {
  background: var(--footer-bg);
  color: var(--footer-text);
  /* 32px vertical at desktop, 24px at 360px. The gutters come from
     .wrap, which is already 40px at desktop and 20px at 360px. */
  padding-block: clamp(24px, 3vw, 32px);
  /* The other half of the sticky footer declared on body: body is a
     column flex container of min-height 100dvh, and this margin eats
     the leftover space so the footer sits at the bottom on a short
     page and immediately after the content on a long one. Harmless
     where the page already overflows. */
  margin-top: auto;
}
.sitefooter-inner {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: 1rem 2rem;
}
.footer-brand {
  margin: 0;
  display: flex;
  align-items: baseline;
  gap: 0.5rem;
  font-size: 1.25rem;
  font-weight: 700;
  letter-spacing: -0.02em;
}
.footer-dot { opacity: 0.7; }
/* 14.49:1 on the footer purple. Set at 0.9375rem rather than dimmed
   with opacity, so the measured ratio is the rendered ratio. */
.footer-stage { font-size: 0.9375rem; font-weight: 500; }

/* ---- FOOTER LINKS — canvas v180, 11 Oct 2026 ---------------------

   Both boards draw "Our mission" and "Privacy notice" between the
   brand line and the tagline. The desktop footer is a space-between
   row of three, so the links land in the middle of it and read as
   centred without being centred by a rule of their own — that is
   .sitefooter-inner's existing justify-content doing the work, and
   adding text-align here would fight it.

   UNDERLINED AT REST. The boards draw them plain, and plain is wrong
   on this surface for a reason the board cannot show: these sit in a
   row of cream text on navy where the brand line and the tagline are
   also cream, so an un-underlined link is identified by nothing at
   all — not even colour. WCAG 1.4.1 again. The underline is the
   non-colour cue and it is the smallest honest departure from the
   board; flagged in the report rather than made silently.

   --footer-text on --footer-bg is 14.49:1, the same ratio .footer-legal
   is set at and documented with. */
.footer-links {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: 0.5rem 1.5rem;
  font-size: 0.9375rem;
  font-weight: 500;
}
.footer-links a {
  color: var(--footer-text);
  text-decoration: underline;
  text-decoration-thickness: 1px;
  text-underline-offset: 3px;
}
.footer-links a:hover { text-decoration-thickness: 2px; }

/* EXTERNAL PROFILE LINKS. Same visual treatment as .footer-links --
   they sit in the same band and a different underline weight would
   read as a different KIND of link, which these are not. They are
   separated semantically (their own element, outside the Footer nav),
   not visually.

   Why they exist at all: orgjsonld.go's sameAs names these two
   profiles, and a sameAs claim corroborated from both ends is worth
   more than one asserted from the site alone. _chrome.html carries
   the full note. */
.footer-social {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: 0.5rem 1.5rem;
  margin: 0;
  font-size: 0.9375rem;
  font-weight: 500;
}
.footer-social a {
  color: var(--footer-text);
  text-decoration: underline;
  text-decoration-thickness: 1px;
  text-underline-offset: 3px;
}
.footer-social a:hover { text-decoration-thickness: 2px; }

/* PHONE: the row stops distributing and the three blocks stack in DOM
   order — brand / links / tagline — which is exactly the order
   HomePhone.dc.html draws them in, so nothing needs reordering. The
   20px gap is the phone board's.

   THE STACK IS NOW DECLARED, NOT INFERRED — 12 Oct 2026, and the old
   comment was a claim this rule did not enforce. The three blocks
   stacked only because .footer-brand happened to be 176px wide, which
   left too little room for the 207px links block beside it. When the
   "· Pre-launch" badge came off the brand line that day the brand
   shrank to 81px, the links rode up beside it at 393px, and the
   footer stopped matching HomePhone.dc.html — which draws brand,
   links and tagline as three separate lines. It still stacked at
   360px, so a check at one phone width would have missed it.

   flex-direction: column states the intent instead of depending on a
   measurement that the content can change. align-items: flex-start
   stops the stretched children centring their own contents. */
@media (max-width: 34rem) {
  .footer-links { gap: 0.5rem 1.25rem; }
  .sitefooter-inner {
    flex-direction: column;
    align-items: flex-start;
    gap: 0.75rem;
  }
}

.footer-tagline {
  margin: 0;
  display: flex;
  flex-wrap: wrap;   /* the swoosh drops above the words when it must */
  align-items: center;
  gap: 0.875rem;
  min-width: 0;
}
/* max-width rather than flex:none. flex:none made the 120px swoosh
   incompressible and the footer was the one thing on this page that
   overflowed below 240px — found by measuring, not by looking. */
.footer-swoosh {
  display: block;
  height: auto;
  max-width: 100%;
  flex: 0 1 auto;
}
.footer-tagline-text {
  font-size: 0.8125rem;
  line-height: 1.5;
  letter-spacing: 0.08em;
  text-transform: uppercase;
  text-align: right;
  min-width: 0;
  /* The 0.08em tracking on an uppercase two-line phrase is what
     stopped this shrinking; allowing a break inside a word is ugly
     but it only ever fires far below 320px, and a broken word beats a
     horizontal scrollbar. */
  overflow-wrap: break-word;
}
@media (max-width: 34rem) {
  .footer-tagline-text { text-align: left; }
}

/* ON THE DARK CONFIRMATION PAGE THIS IS THE LAST CHILD OF <body>,
   standing in for the footer the template drops, so it takes the
   leftover height the way .sitefooter does elsewhere. Inside a
   normal footer it is not a grid item and margin-top:auto is inert,
   so this is safe in both places. */
.footer-legal {
  margin: auto 0 0;
  padding-top: 1rem;
  border-top: 1px solid color-mix(in srgb, var(--footer-text) 22%, transparent);
  font-size: 0.875rem;
  line-height: 1.5;
  /* Full --footer-text, 14.49:1. This sentence is the FCA-relevant
     one; it does not get set in a dimmed grey. */
  color: var(--footer-text);
}

/* The homepage supplies its own full-bleed bands, so <main> must not
   carry the page's own horizontal padding — the bands do it
   themselves, and .wrap inside each one restores the measure. */
body.home main { padding-inline: 0; max-width: none; }

/* ---- The advisory-form side panel -------------------------------
   Collapses at 58rem based on where the copy column stops holding a
   readable measure beside the portrait artwork — not on a device
   width. Below that the artwork follows the copy rather than pushing
   the form off the first screen. */

.hero-split { display: grid; gap: clamp(1.5rem, 4vw, 2.5rem); align-items: start; }

@media (min-width: 58rem) {
  .hero-split { grid-template-columns: minmax(0, 1.25fr) minmax(0, 1fr); }
}

.hero-aside {
  background: var(--panel);
  border-radius: var(--radius-panel);
  overflow: hidden;
}

/* The portrait variant. Capped so it cannot dominate the page at wide
   viewports, and never cropped or stretched — width/height + auto
   keeps the generated aspect ratio exactly. */
/* The portrait has calm cream sky at the top which, with nothing
   placed in it, reads as an empty gap. A negative top margin inside
   the clipping panel trims that headroom WITHOUT cropping the image
   box itself or changing its aspect ratio — the whole image is still
   laid out at its true proportions, the panel just shows less of the
   sky. No object-fit, no transform, no flip. */
.hero-art--portrait {
  width: 100%;
  height: auto;
  border-radius: 0;
  display: block;
  margin-top: -12%;
}

.hero-aside-copy { padding: 1.25rem; }
.hero-aside-copy h2 {
  margin: 0 0 0.5rem;
  font-size: clamp(1.25rem, 1.1rem + 0.8vw, 1.625rem);
  letter-spacing: -0.02em;
  text-wrap: balance;
}
/* Scoped so it cannot reach .tag below. `.hero-aside-copy p` is
   specificity 0-1-1 and `.tag` is 0-1-0, so the paragraph rule won
   regardless of source order and painted the tag's white text dark —
   #291449 on #6624C9, which measured 2.06:1. Caught by auditing
   rendered pixels rather than reading the stylesheet. */
.hero-aside-copy > p:not(.tag) { margin: 0 0 0.75rem; color: var(--ink); }

/* "Planned for launch" — a status, not a button. Deliberately not
   styled as one: nothing here is clickable and a pill that looks
   pressable invites a click that does nothing. */
.tag {
  display: inline-block;
  margin: 0 0 0.75rem;
  padding: 0.3rem 0.7rem;
  border-radius: 999px;
  background: var(--brand);
  color: var(--on-brand);
  font-size: 0.875rem;
  font-weight: 600;
}

/* ---- Status screens (confirmed, thanks) -------------------------
   Artwork is decorative and SECONDARY to the message: it follows the
   copy in the DOM, and on narrow screens it sits below rather than
   pushing the message off the first screen. */

.hero-status { display: grid; gap: clamp(1.5rem, 4vw, 2.5rem); align-items: center; }

@media (min-width: 54rem) {
  .hero-status { grid-template-columns: minmax(0, 1.2fr) minmax(0, 1fr); }
}

/* .status-art IS GONE. It sized the confirmed and thanks artwork with
   height:auto and, being later in the file than .flowsplit-art img,
   it silently overrode the bottom-anchored 16:9 crop those two
   screens are supposed to share with the waitlist and decline
   screens — so two of the four board-03 panels showed a half-empty
   sky and two did not. Found by comparing the rendered screens, not
   by reading the CSS.

   Both templates now use .flowsplit-art like the other two, and
   there is one rule for the four of them. */

/* "No policy has been created." The single most important sentence on
   the confirmation screen, so it is a panel and not fine print. */
.confirm-note {
  margin: 1.25rem 0;
  padding: 0.875rem 1rem;
  background: var(--panel);
  border-radius: var(--radius);
  max-width: var(--measure);
  font-size: 1rem;
}

/* The secondary action under the primary one, per the confirmation
   pack: "retain the text link below the primary action". It is an
   ordinary underlined link rather than a second button, so that the
   two actions do not compete -- the pack's mockup draws exactly
   that. Inherits the page's link colour, so it needs no colour of
   its own in either theme. */
.confirm-backlink {
  margin: -0.5rem 0 0;
}
.confirm-backlink a {
  font-size: 1rem;
}

/* ===================================================================
   THE FLOW SCREENS — design boards 02 and 03, built 22 Sep 2026
   ===================================================================

   The homepage was rebuilt to board 01 on 22 Sep and the other nine
   templates were not, so the site had two visual languages: a
   full-bleed purple/orange homepage and a narrow single cream column
   for everything after it. Everything below brings the advisory form,
   the result and the four waitlist/feedback screens onto the same
   board.

   THE TOKENS ARE THE ONES ALREADY HERE. Nothing new is invented
   above: this section introduces no colour that is not already
   defined at the top of this file or in the homepage block. Every
   pair was measured on the RENDERED, COMPOSITED page over CDP, both
   themes, not read off the token table — an earlier round put a
   headline at 1.00:1 by trusting a token value against artwork.

   Measured, light / dark, on the running local stack:

     --ink on --page (body, headings)     15.55:1 / 17.66:1
     --ink-soft on --page (lede, help)     8.33:1 / 10.12:1
     --ink on --panel (info note, pill)   14.12:1 / 12.31:1
     --ink on --secondary-panel-bg        13.32:1 / 12.94:1
     --on-cta on --cta (every CTA)         8.67:1 /  8.67:1
     nav link --ink on --page             15.55:1 / 17.66:1
     --accent focus ring on --page         7.54:1 /  9.85:1

   DESKTOP AND MOBILE ARE BOTH DESIGNED. The form splits into two
   columns at 64rem and the status screens at 60rem; both numbers are
   where the content stops fitting, measured, not a device width.

   NO LAYOUT SHIFT. Every <img> below declares width and height in the
   markup, so its box exists before the bytes do. */

/* ---- Shared flow chrome ------------------------------------------

   The nav here is on CREAM with dark-purple links, which is what
   boards 02 and 03 draw. The homepage's nav is on the hero because a
   hero starts immediately beneath it; these pages have no
   hero panel, so a purple bar would be a band of colour attached to
   nothing. Both are the board. */

.flownav {
  background: var(--page);
  color: var(--ink);
  border-bottom: 1px solid var(--line);
}
.flownav-inner {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  /* 72px min-height and a 32px brand-to-links gap at desktop, per
     layout.md; both collapse at 30rem where the wordmark and the
     three links stop sharing a line. */
  min-height: 72px;
  gap: 0.75rem 2rem;
  padding-block: 0.875rem;
}
/* ⚠️ THE "o" MEASURES 2.84:1 ON THE FLOW PAGES AND THAT IS THE
   BOARD'S OWN VALUE. #E8762A (--logo) on #FFF9F1 (--page) is what
   MainV2/VehicleV2/ResultV2 specify, and it is the same reading the
   homepage header has carried since v2 -- the flow pages only
   acquired it on 23 Oct because the square mark they used before had
   no text in it at all.

   1.4.11 EXEMPTS IT: "text that is part of a logo or brand name has
   no contrast requirement". The link's accessible name comes from
   aria-label="{{ .BrandName }}", not from this letter being
   discriminable, and the contrast walker flags it on every page that
   draws the wordmark. Dark is 9.52:1 and never flagged.

   DO NOT "FIX" IT BY DARKENING --logo HERE. That would put the flow
   pages' wordmark out of step with the homepage's and with every
   board, to satisfy a rule that does not bind. If the logotype
   colour should change it changes on the canvas, once, for all of
   them. */

/* ⚠️ gap:0 AND align-items:baseline, 23 Oct 2026, with Logo B.
   The 0.5rem gap and centre alignment were right for the square --
   a separate mark set beside a word. The ring is the LETTER C of the
   wordmark, so it needs neither: it carries its own -0.03em overlap
   and sits on the baseline with the letters. The same correction was
   made on .sitenav .brand and .mission .brand; leaving the gap here
   would have put a 8px hole between the C and the "o". */
.flownav-brand {
  display: inline-flex;
  align-items: baseline;
  gap: 0;
  color: var(--ink);
  text-decoration: none;
  font-size: 1.375rem;
  font-weight: 700;
  letter-spacing: -0.02em;
}
/* 28px between links at desktop, 20px when they wrap. No hamburger:
   three links and a wordmark fit on one line at 360px (measured), and
   layout.md forbids a full menu plus a redundant toggle. */
.flownav-links {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: 0.5rem 1.75rem;
  font-size: 1rem;
  line-height: 1.5;
}
@media (max-width: 30rem) {
  .flownav-inner { gap: 0.5rem 1.25rem; }
  .flownav-links { gap: 0.5rem 1.25rem; }
}
/* A PERMANENT UNDERLINE, not one that appears on hover. A link that
   is only a colour at rest is indistinguishable from bold text to
   anyone who cannot perceive the hue difference — WCAG 1.4.1. */
.flownav-links a {
  color: var(--ink);
  font-weight: 600;
  text-decoration: underline;
  text-underline-offset: 0.3em;
  text-decoration-thickness: 1px;
  /* ORANGE ON EVERY ITEM, 29 Oct 2026. Guillaume, on a screenshot of
     the /get-in-touch nav where only "Home" carried it: "they
     underlying colour should should be orange in both cases".

     It was a muted ink at 45% until then, and the orange arrived
     earlier the same day on "Home" alone -- which made the one link
     look selected rather than simply underlined. One colour for all
     of them removes that false signal, and the rule is still the
     WCAG 1.4.1 affordance separating a link from body text.

     --nav-home-rule carries the orange in BOTH themes, which is why
     it exists; three other tokens were measured and rejected. Its
     name is now narrower than its job -- it is the nav link rule,
     not Home's -- and renaming it is a bigger edit than this change
     earns. Left as a known misnomer rather than a silent one. */
  text-decoration-color: var(--nav-home-rule);
  padding-block: 0.25rem;
}
/* Hover goes to the ink, which is the inverse of the old behaviour
   and the only pairing that still reads as a state change now that
   the resting colour is the accent. */
.flownav-links a:hover { text-decoration-color: var(--ink); }
/* THE CURRENT PAGE IN THE SHARED NAV, added 13 Oct 2026.
   A page must not link to itself, so _chrome.html renders the matching
   item as a <span aria-current="page"> rather than an <a href>. This is
   what that span looks like.

   WEIGHT 700 AND NO UNDERLINE, which is Contact.dc.html: it draws
   "Contact" bold with an orange rule under it while its siblings are
   plain, and the boldness is the half that transfers. The orange rule
   does not, because on THIS nav the underline is the WCAG 1.4.1
   affordance that separates a link from body text (see the rule above);
   putting it on a non-link would say "link" about the one item that is
   not one.

   SAME COLOUR AS ITS SIBLINGS, deliberately. --ink on --page is
   15.55:1 light and 17.66:1 dark, so this is the highest-contrast text
   on the bar; dimming the current item to distinguish it would make the
   page you are on the hardest thing on the bar to read. aria-current
   carries the meaning; the weight carries the look.

   THE PADDING MATCHES THE LINKS' padding-block so swapping a link for
   this span moves nothing: the two states are the same height, so the
   nav does not shift between pages. */
.flownav-links .flownav-current {
  color: var(--ink);
  font-weight: 700;
  padding-block: 0.25rem;
}
/* The back link keeps its arrow on the same line as its words. */
.flownav-back {
  display: inline-flex;
  align-items: center;
  gap: 0.5rem;
}
.flownav-back svg { flex: none; }

/* --flowmain gives every flow page the same rhythm under the nav. */
.flowmain {
  padding-block: clamp(1.75rem, 4vw, 3rem) clamp(2.5rem, 6vw, 4rem);
}

/* ---- Flow typography ---------------------------------------------

   The board's headings are large and tight. 60px at desktop is the
   homepage hero's size and is too big for a page that also carries a
   form, so these run 40px–48px and drop to 32px at 360px. Line
   height 1.05 matches the board's stacked two-line headings. */
.flow-h1 {
  margin: 0 0 0.625rem;
  font-size: clamp(2rem, 1.35rem + 2.6vw, 3rem);
  line-height: 1.05;
  letter-spacing: -0.03em;
  font-weight: 700;
  text-wrap: balance;
}
.flow-lede {
  margin: 0 0 1.5rem;
  max-width: 34rem;
  font-size: 1.125rem;
  line-height: 1.5;
  color: var(--ink-soft);
}
.flow-back { margin-top: 2rem; font-size: 1rem; }
.flow-back a { color: var(--accent); }

/* ---- The info note, board 02 panel 01 ----------------------------

   A panel with an icon and two lines. --ink on --panel is
   14.12:1 light and 12.31:1 dark, both measured on the rendered
   page. The icon is currentColor so it follows the theme. */
.infonote {
  display: flex;
  align-items: flex-start;
  gap: 0.75rem;
  margin-block: 0 1.75rem;
  padding: 0.875rem 1.125rem;
  max-width: 34rem;
  background: var(--panel);
  border-radius: var(--radius);
  color: var(--ink);
  font-size: 1rem;
  line-height: 1.5;
}
.infonote-icon { flex: none; margin-top: 0.15rem; color: var(--brand); }

/* ---- The scope note, step 1 only ---------------------------------

   "Today's estimate uses your age, postcode and car (and licence
   years, for newer drivers). Driving analysis comes later."

   MOVED HERE FROM THE HOMEPAGE HERO, 10 Oct 2026 (item 4 of
   Guillaume's spec, identical in Options A and B). It was
   .hero-clarify, set in
   --hero-text on the hero; here it sits on --page, so the colour is a
   token swap rather than the same value re-used.

   IT IS DELIBERATELY NOT AN .infonote. An infonote is a lavender
   panel with an ⓘ, and there is one directly above this carrying
   D9's definition. A second panel immediately under the first would
   read as one block of boilerplate and the eye would skip both —
   which is the failure mode a qualifier cannot afford. So this is
   plain text at a smaller size, distinguished by weight of voice
   rather than by another box.

   --ink-soft on --page is 8.33:1 light and 10.12:1 dark. It is NOT
   dimmed below that: a qualifier set in a lower-contrast grey is a
   qualifier nobody reads. It is distinguished by size, which is the
   same argument .hero-clarify made on the purple.

   34rem is the measure every other element in this column uses
   (.flow-lede, .infonote), so the left and right edges line up down
   the page rather than this one sentence setting its own width. */
.flow-scope {
  margin: 0 0 1.75rem;
  max-width: 34rem;
  font-size: 0.9375rem;
  line-height: 1.5;
  color: var(--ink-soft);
}

/* ---- The conditional note ----------------------------------------

   The panel a board draws with an "If <answer>" tag on the left: step
   1's "If Not yet" brand-new-driver note (5 Oct 2026), and step 2's
   "If I've picked one". Both boards draw the same object, so it is one
   rule rather than two.

   IT IS NOT AN .infonote. An infonote is unconditional context about
   the whole screen and carries an ⓘ; this is a consequence of one
   answer and carries the answer's name, which is the information that
   makes it readable out of context. Same panel fill, smaller type, a
   tighter radius, and no icon.

   THE TAG IS TEXT, NOT A PSEUDO-ELEMENT. `content: "If \201CNot yet\201D"`
   would put the only words identifying which answer this belongs to
   somewhere a screen reader may or may not read and a translator
   cannot reach at all.

   WRAPS ON A PHONE. flex-wrap, so the tag sits above the sentence
   rather than squeezing it into a 12rem column at 390px: measured, the
   sentence needs the full width there. */
.condnote {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-start;
  gap: 0.375rem 0.625rem;
  margin-block: 0 1.25rem;
  padding: var(--ctrl-pad-y) var(--ctrl-pad-x);
  max-width: 34rem;
  background: var(--panel);
  border-radius: 0.75rem;
  color: var(--ink);
  font-size: 0.875rem;
  line-height: 1.5;
}
.condnote-tag {
  flex: none;
  padding-top: 0.125rem;
  font-size: 0.6875rem;
  font-weight: 700;
  letter-spacing: 0.06em;
  text-transform: uppercase;
  /* 7.56:1 on --panel — this is small, uppercase, tracked-out text and
     it still has to pass AA at its own size. */
  color: var(--ink-soft);
}
/* NO .condnote[hidden] RULE HERE, and that is checked rather than
   assumed: the global `[hidden] { display: none !important }` near the
   top of this file (added for the car combobox, for this exact reason)
   already beats the `display: flex` above. A second rule would be a
   duplicate claim that could later disagree with the first.

   WHY THE ATTRIBUTE AND NOT A CLASS: the server renders `hidden` from
   the licence answer and quote.js toggles the same property, so there
   is one mechanism and no state where a class says shown and the
   attribute says hidden. It also removes the subtree from the
   accessibility tree, which a visibility class does not. */

/* ---- The "Planned for launch" pill -------------------------------

   A statement of tense, not a badge. It sits on --paper inside the
   lavender sidebar so it separates from the panel behind it;
   --ink on --paper is 16.27:1 light and 15.55:1 dark. */
.pill {
  display: inline-block;
  margin: 0 0 1.25rem;
  padding: 0.3125rem 0.75rem;
  background: var(--paper);
  color: var(--ink);
  border-radius: 999px;
  font-size: 0.875rem;
  font-weight: 600;
  line-height: 1.4;
}

/* ---- Two-column form, board 02 panel 01 --------------------------

   Form left, sidebar right, side by side above 64rem. Below that the
   sidebar stacks BELOW the form: the form is the task and the
   sidebar is supporting material, so it never comes first.

   The sidebar is after the form in the SOURCE at every width, so tab
   order and reading order already match the visual order and nothing
   here reorders content for sighted users only. */
.formgrid { display: grid; gap: 2rem; }
/* THE GRID ENGAGES AT 56rem, NOT 64rem.

   It used to switch at 64rem/1024px, which left a DEAD ZONE from about
   900 to 1023: wide enough that a full-width form looks lost, not wide
   enough for the sidebar to be allowed beside it. Guillaume's own
   screenshot was at 993px, squarely inside it.

   Worse, it made the transition violent. Measured with the measure-cap
   removed: the form column ran 828 -> 914 -> 941px as the window grew
   to 1023, then COLLAPSED to 534 at 1024 when the sidebar finally took
   its 360px. A 400px drop from one pixel of window.

   56rem/896px is where 22.5rem of sidebar plus 3rem of gap still
   leaves the form over 30rem — enough for the three-up name row and a
   three-column card grid. Below that the sidebar stacks, which is
   correct: it is supporting material and the form is the task.

   Engaging earlier is the right half of the choice because the sidebar
   is the thing that should yield. Making it stack LATER would leave
   the form full-width at 1000px, where a 900px line of radio cards is
   worse than a 600px one beside a panel. */
@media (min-width: 56rem) {
  .formgrid {
    grid-template-columns: minmax(0, 1fr) 22.5rem;
    gap: 3rem;
    align-items: start;
  }

  /* STICKY, AND ONLY WHERE THERE ARE TWO COLUMNS.
     15 Oct 2026, from an external review: "on the form, on desktop, I
     would make the right banner fixed as you scroll down, otherwise
     the space on the right looks empty and the form itself is not
     centered on the page."

     Measured at 1280 before this: the panel is 689.7px tall in a
     2,561px form column on step 1, so roughly 1,870px of the scroll
     happened beside an empty 360px strip — and once that column reads
     as blank, the page looks left-weighted rather than centred. The
     second half of his remark is a consequence of the first, not a
     separate fault.

     Inside the 56rem query deliberately: below it the sidebar stacks
     BELOW the form, where sticky would pin supporting material over
     the task. Guillaume, 15 Oct 2026: "I think this is mostly for
     desktop."

     `align-items: start` above is load-bearing for this — a stretched
     grid item is full-height and sticky has nothing to scroll within.
     Do not change it to `stretch` without removing this.

     THE max-height AND overflow ARE NOT DECORATION. The panel is
     ~690px and a laptop viewport is often ~700px, so without a cap a
     short window pins something taller than itself and the bottom of
     the panel becomes permanently unreachable. Measured at 1280x600
     with no cap: 138px of the panel below the fold and no way to
     reach it.

     THE overflow-y HERE MUST BEAT .formaside's `overflow: hidden`,
     which is further down this file and exists so the artwork tile
     meets the panel's rounded corners. Without winning that, the cap
     below clips the panel instead of scrolling it: measured at
     1280x600, 138px of the panel sat below the fold with no way to
     reach it (38px at 700, 88px at 650).

     ON THE QUESTIONNAIRE PAGES .formaside AND .formgrid-side ARE THE
     SAME ELEMENT — the aside carries both classes. So a descendant or
     child selector matches NOTHING here; the two classes have to be
     combined on one compound selector, which also gives the 0,2,0
     that outranks the later `.formaside`. Do not "tidy" this into
     `.formgrid-side > .formaside`; it silently stops applying.

     overscroll-behavior stops a flick inside the panel from carrying
     on into the page once it bottoms out. */
  .formgrid-side {
    position: sticky;
    top: 1.5rem;
  }
  .formaside.formgrid-side {
    max-height: calc(100vh - 3rem);
    overflow-y: auto;
    overscroll-behavior: contain;
  }
}
.formgrid-main { min-width: 0; }
.formgrid-side { min-width: 0; }

.formaside {
  background: var(--secondary-panel-bg);
  color: var(--ink);
  border-radius: var(--radius-panel);
  padding: 1.75rem 1.75rem 0;
  overflow: hidden;   /* the artwork tile meets the panel's corners */
}
.formaside-h {
  margin: 0 0 0.875rem;
  font-size: clamp(1.5rem, 1.2rem + 1.2vw, 1.875rem);
  line-height: 1.12;
  letter-spacing: -0.02em;
  font-weight: 700;
}
.formaside-body {
  margin: 0 0 1.25rem;
  font-size: 1.0625rem;
  line-height: 1.5;
  color: var(--ink);
}
/* THE ARTWORK TILE IS CREAM AND SAYS SO.

   hero-sidebar.png is opaque cream-background artwork with generated
   texture — the illustration pack states this explicitly and says to
   "use an intentional cream image panel rather than pretending these
   images have transparent edges". The board draws the car directly
   on lavender; reproducing that honestly with this asset means a
   cream tile inside the panel, not a recolour of somebody
   else's illustration by eye.

   The tile bleeds to the panel's left, right and bottom edges, which
   is the board's composition. NOT FLIPPED: the car travels on the
   left of the centreline and the pack forbids a horizontal flip. */
.formaside-art {
  margin-inline: -1.75rem;
  margin-bottom: 0;
  background: #FFF9F1;   /* the artwork's own cream field */
  /* A FIXED 4:5 FRAME, and the image covers it anchored to the
     BOTTOM. The asset is a 2:3 portrait whose upper third is empty
     cream sky; shown whole it puts a tall blank band between the
     "Planned for launch" pill and the street, which the board does
     not have. Anchoring to the bottom crops sky and keeps the car,
     the houses and the road — every part the board shows.

     object-fit: cover is used HERE and must never be used on the
     dashcam poster or its annotation overlay: the illustration pack
     forbids it for that pair because the two layers have to stay in
     the same coordinate system. This image has no overlay to stay
     aligned with.

     NOT FLIPPED. object-position touches the vertical axis only, so
     the car still travels on the left of the centreline. */
  aspect-ratio: 4 / 5;
}
.formaside-art img {
  display: block;
  width: 100%;
  height: 100%;
  object-fit: cover;
  object-position: 50% 100%;
}
@media (max-width: 55.99rem) {
  /* Stacked on a phone the panel is full width, so a 4:5 frame is
     very tall. 16:10 keeps the same bottom-anchored crop in a band.
     Follows the grid's breakpoint, moved to 56rem. */
  .formaside-art { aspect-ratio: 16 / 10; }
}

/* ---- Status screens: copy left, artwork right --------------------

   Boards 03 panels 03–06. Two columns above 60rem, stacked below,
   with the artwork always AFTER the copy in the source. */
.flowsplit { display: grid; gap: 1.75rem; }
@media (min-width: 60rem) {
  .flowsplit {
    grid-template-columns: minmax(0, 1fr) minmax(0, 1fr);
    gap: 3rem;
    align-items: center;
  }
}
.flowsplit-main { min-width: 0; }
.flowsplit-art { min-width: 0; }
/* A 3:2 FRAME WITH THE IMAGE ANCHORED TO THE BOTTOM.

   road-trees.png is a 3:2 landscape whose upper HALF is empty cream
   sky. Shown whole it is a tall pale rectangle with the road in a
   band at the bottom, which is not what board 03 draws — the board
   fills the panel with road, hills and trees. Anchoring to the
   bottom of a shorter frame keeps every element the board shows and
   drops only sky.

   The artwork's own cream is declared as the tile background so that
   in dark mode it reads as a deliberate cream panel rather than as a
   rectangle that failed to invert — the illustration pack is
   explicit that these PNGs are opaque cream artwork and that a dark
   surface should use an intentional cream panel.

   NOT FLIPPED: object-position moves the vertical axis only. */
.flowsplit-art img {
  display: block;
  width: 100%;
  /* NOT height:auto. With width:100% and height:auto the intrinsic
     ratio wins and aspect-ratio is ignored, which is why an earlier
     attempt at this crop changed nothing on the rendered page. */
  height: 100%;
  /* 16:9, NOT the asset's own 3:2. Measured on the PNG: the road,
     hills and trees begin 38.3% of the way down and everything
     above that is empty cream sky. A 3:2 frame therefore renders as
     a pale rectangle with the illustration in a band at the bottom,
     which is not what board 03 draws. 16:9 bottom-anchored drops
     most of the sky and keeps the whole scene.

     A ratio wide enough to remove ALL the sky would be 2.43:1 — a
     letterbox strip. 16:9 is the compromise: a little sky stays,
     and it reads as headroom rather than as emptiness. */
  aspect-ratio: 16 / 9;
  object-fit: cover;
  object-position: 50% 100%;
  border-radius: var(--radius-panel);
  background: #FFF9F1;
}

/* ---- The confirmation page's artwork is a DIFFERENT SHAPE --------

   28 Oct 2026, with the confirmation pack. The rule above crops to
   16:9 and covers, which is right for road-trees: that asset is an
   opaque 3:2 whose top 38% is empty cream sky, so cropping loses
   nothing and the cream tile reads as a deliberate panel.

   NEITHER HALF OF THAT HOLDS HERE and copying it would break the
   picture twice over:

     - These assets are ALPHA-TRANSPARENT PNG/WebP, not opaque cream.
       A cream tile behind them is a cream rectangle in dark mode,
       which is exactly the "rectangle that failed to invert" the
       rule above is at pains to avoid. The panel shows through
       instead, which is what the transparency is for, and the pack
       says in terms not to fill it with a white matte.
     - object-fit: cover on a 4:3 composition inside a 16:9 frame
       crops the top and bottom. The car sits low and the foliage
       high, so cover clips one or both. The pack's instruction is
       "keep the complete car visible" and "contain, never stretch".

   So: contain, the asset's own ratio, no background. The intrinsic
   width/height attributes in the template stop it shifting layout
   while it loads. */
.flowsplit-art--confirm img {
  /* 5:6 PORTRAIT AND cover, NOT 4:3 AND contain.

     The pack says "contain, never stretch" and "keep the complete
     car visible", and the first build took that literally: a 4:3
     contain box. Put beside the mockup the panel came out as a flat
     band with the artwork letterboxed inside it, while the mockup
     draws a TALL card whose right half the scene fills completely.

     cover does not stretch — it crops, and the crop is chosen below
     so the car is never the thing cropped. The asset is 4:3 with
     generous sky and verge around the car, which is what makes a
     taller frame safe: it eats margin, not subject. Verified by
     eye at 1440, 1280 and 390. */
  aspect-ratio: 5 / 6;
  object-fit: cover;
  /* Bottom-left: the car sits low and left of centre in the asset,
     so this keeps it whole and drops sky from the top. */
  object-position: 30% 72%;
  background: none;
  border-radius: 0;
  /* THE MOCKUP'S ARTWORK FILLS THE PANEL'S RIGHT HALF AND BLEEDS TO
     ITS EDGES. The first build left it inside the content column at
     the page's ordinary gutter, which rendered it about half the
     size the mockup draws and made the panel a flat band rather
     than the tall card the pack shows.

     The negative margins pull it out to the panel's padding edge on
     the top, right and bottom; width:calc() gives back the space so
     nothing overflows. The left edge stays put so the gap to the
     copy column survives. */
  width: calc(100% + var(--confirm-pad));
  margin: calc(var(--confirm-pad) * -1) calc(var(--confirm-pad) * -1)
          calc(var(--confirm-pad) * -1) 0;
  border-bottom-right-radius: 28px;
  border-top-right-radius: 28px;
}
/* 54rem, matching .hero-status and the two-column rule above — see
   the note there. A mismatched pair of breakpoints is what put a
   patch of sky inside the cream panel between 54 and 60rem. */
@media (max-width: 53.999rem) {
  /* STACKED, AND BACK TO A LANDSCAPE RATIO.

     The 5:6 portrait above is for the desktop two-column panel. Left
     applying at phone width it made the artwork taller than the
     viewport, which pushed the copy out of the panel's
     overflow:hidden entirely — the first phone render showed a car
     and no words at all.

     The mockup's phone frame puts a LANDSCAPE crop under the copy,
     which is what this is. */
  .flowsplit-art--confirm img {
    aspect-ratio: 16 / 10;
    object-position: 40% 68%;
    width: calc(100% + var(--confirm-pad) * 2);
    margin: var(--confirm-pad) calc(var(--confirm-pad) * -1)
            calc(var(--confirm-pad) * -1);
    border-radius: 0 0 28px 28px;
  }
  /* THE ART GOES LAST ON PHONE, below the disclosure.

     The source order puts the copy first and the art second, which
     is right for desktop's two columns and right for a screen
     reader. Stacked, the grid placed the art above the copy because
     .flowsplit-main is the second child in the DOM for the desktop
     column order — so the phone showed a car and then the words.
     ConfirmV2Phone draws the opposite, and the pack says
     "confirmation and actions first, then artwork".

     order on the two children rather than a DOM change, so the
     reading order stays copy-first for assistive tech.

     IT STAYS A GRID. An earlier attempt set the panel to
     `display:flex; flex-direction:column`, which dropped the grid's
     column sizing — .flowsplit-main collapsed to its longest word
     and the copy rendered one word per line. The base .flowsplit is
     already a single-column grid below 60rem, so only the ROW each
     child sits in needs stating. */
  .confirmsky .confirmpanel { grid-template-rows: auto auto; }
  .confirmsky .flowsplit-main { grid-row: 1; max-width: none; }
  .confirmsky .flowsplit-art--confirm { grid-row: 2; }
}

/* ================================================================
   THE CONFIRMATION PAGE: SKY, PANEL, CLOUDS, TICK
   ================================================================

   28 Oct 2026, from the confirmation pack's approved mockups. The
   first build of this page reused the ordinary page shell and only
   swapped the artwork, which left it looking nothing like the
   mockup — Guillaume put the two side by side.

   Four things the mockup has and the page did not: a pale blue sky
   as the page ground, a cream panel floating on it, low-density
   clouds, and a navy tick on a pale mint circle. Tokens are the
   pack's own design-tokens.json.

   DARK MODE IS THE PACK'S DARK MOCKUP, not an inversion: a deep
   navy sky, a lighter navy panel, and the clouds dropped to a
   whisper so they read as night cloud rather than as grey smudges.

   SCOPED to .confirmsky / .confirmpanel so none of it reaches
   another page. */
.confirmsky {
  --sky-page: #DCEFFA;
  --sky-panel: #FFF9F1;
  --sky-cloud: #FFFFFF;
  /* The mockup's clouds are solid white and clearly read as clouds.
     An earlier 0.85 on a pale sky made them near-invisible. */
  --sky-cloud-a: 1;
  --sky-tick-bg: #C9EDDD;
  --sky-tick-fg: #142B49;
  background: var(--sky-page);
  position: relative;
  overflow: hidden;
  /* The sky needs room to be seen above and below the panel; the
     mockup shows roughly a panel's-worth of margin at desktop. */
  padding-block: clamp(1.5rem, 4vw, 3.5rem);
}
/* THE DARK VALUES ARE MEASURED OFF THE PACK'S DARK MOCKUP, not
   derived from the light ones. The first pass guessed #101D30 and
   #182B43 and put a dark-green disc behind the tick; sampling
   confirmation-dark-desktop-phone.png gave:

     sky    #081C39      panel  #0E2441      tick   #8DD3B3
     head   #EDF0F7      body   #A0B4CA

   Two things the guess got wrong and the mockup settles:

     - the panel is barely lighter than the sky (#0E2441 against
       #081C39, six points of luminance). It lifts by its shadow and
       its radius, not by contrast. A lighter slab reads as a
       different component.
     - THE TICK DISC STAYS MINT. It is the one warm, saturated note
       in a navy page and the mockup keeps it at #8DD3B3 with the
       navy check still on it. Darkening it to a green would have
       lost the only colour the design puts there. */
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .confirmsky {
    --sky-page: #081C39;
    --sky-panel: #0E2441;
    --sky-cloud: #7FA8D4;
    --sky-cloud-a: 0.16;
    --sky-tick-bg: #8DD3B3;
    --sky-tick-fg: #0E2441;
  }
}
:root[data-theme="dark"] .confirmsky {
  --sky-page: #081C39;
  --sky-panel: #0E2441;
  --sky-cloud: #7FA8D4;
  --sky-cloud-a: 0.16;
  --sky-tick-bg: #8DD3B3;
  --sky-tick-fg: #0E2441;
}

/* The panel. Corners 24-32px per the brief; 28px sits between. */
.confirmsky .confirmpanel {
  position: relative;
  z-index: 1;
  background: var(--sky-panel);
  border-radius: 28px;
  --confirm-pad: clamp(1.5rem, 3.5vw, 3rem);
  padding: var(--confirm-pad);
  /* overflow hidden so the bled artwork is clipped by the panel's
     own radius rather than squaring off its corner. */
  overflow: hidden;
  /* The mockup's copy column is NARROWER than half: the heading
     wraps once and the lede wraps at three lines. A 1fr/1fr split
     ran the copy the full width of its half and read as a different
     layout. 0.86fr against 1fr matches the mockup's proportion. */
  align-items: center;
  /* The mockup's panel is a raised card. Kept subtle: a hard shadow
     on a pale sky reads as a sticker. */
  box-shadow: var(--sky-panel-shadow);
}
.confirmsky {
  /* Declared on the scope so both themes override it in one place,
     beside the colours they belong with. */
  --sky-panel-shadow: 0 1px 2px rgba(20, 43, 73, 0.06),
                      0 12px 32px rgba(20, 43, 73, 0.10);
}
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .confirmsky {
    /* On navy, a navy shadow is invisible; the panel lifts by going
       DARKER underneath, the way the mockup's does. */
    --sky-panel-shadow: 0 1px 2px rgba(0, 0, 0, 0.30),
                        0 16px 40px rgba(0, 0, 0, 0.38);
  }
}
:root[data-theme="dark"] .confirmsky {
  --sky-panel-shadow: 0 1px 2px rgba(0, 0, 0, 0.30),
                      0 16px 40px rgba(0, 0, 0, 0.38);
}

/* CLOUDS. Each is one span; the puffs are ::before/::after, so a
   cloud costs no markup beyond its own element. Low density and low
   contrast — the pack says "keep cloud density low", and a cloud
   that competes with the panel is worse than no cloud.

   aria-hidden on the container, and pointer-events:none so they can
   never swallow a click meant for the panel. */
.confirmsky-clouds {
  position: absolute;
  inset: 0;
  pointer-events: none;
  z-index: 0;
  overflow: hidden;
}
.confirmcloud {
  position: absolute;
  display: block;
  background: var(--sky-cloud);
  opacity: var(--sky-cloud-a);
  border-radius: 999px;
}
.confirmcloud::before,
.confirmcloud::after {
  content: "";
  position: absolute;
  background: inherit;
  border-radius: 999px;
}
/* Two above the panel, two below, none behind the heading. */
/* ---- THE CLOUDS DRIFT ------------------------------------------

   Guillaume asked for it, and it suits the subject: the artwork is a
   car on a road, so slow horizontal travel reads as going somewhere
   rather than as decoration for its own sake.

   FOUR RULES IT FOLLOWS:

   1. TRANSFORM ONLY. translate3d is composited on the GPU, so the
      drift costs no layout and no repaint. Animating `left` would
      relayout the whole sky every frame.

   2. EACH CLOUD CROSSES THE WHOLE VIEWPORT AND IS NEVER SEEN TO
      POP. The keyframe runs from -30vw to 130vw: 30vw of lead-in
      and lead-out beyond each edge, which is wider than any cloud,
      so a cloud is always fully outside the sky when it wraps. Its
      `left`/`right` offset is its starting phase.

   3. FOUR DIFFERENT DURATIONS, with negative delays. The durations
      are deliberately not multiples of each other (83/97/113/127
      seconds -- primes), so the four never resynchronise into a
      visible formation. The negative delays start each one
      mid-journey, so the sky is already populated on first paint
      instead of filling up over two minutes.

   4. VERY SLOW. 83 to 127 seconds to cross. At that speed it reads
      as weather, not as a carousel; anything faster competes with
      the text for attention on a page whose job is to confirm
      something.

   REDUCED MOTION IS ALREADY HANDLED, globally: app.css:871 sets
   `animation: none !important` on everything under
   (prefers-reduced-motion: reduce). The clouds then sit at their
   static offsets, which is the layout this started as, so nothing
   is lost but the movement. No separate rule is needed here and
   adding one would duplicate a guard that already works. */
/* THE TRAVEL IS RELATIVE TO WHERE THE CLOUD ALREADY SITS, not an
   absolute sweep of the viewport.

   The first version ran -30vw to 130vw, which treats the keyframe as
   the cloud's POSITION. But each cloud already has a left/right
   offset, so that offset was being added to a 160vw journey: every
   cloud spent most of its cycle outside the sky and the screen was
   usually empty. Measured at 1440: one cloud visible where four were
   drawn.

   Translating 0 -> 100vw and back keeps the static layout as the
   arrangement and only moves it along. Each cloud still crosses the
   full width; the four stay spread because their offsets differ and
   their durations are primes. */
@keyframes confirmdrift {
  from { transform: translate3d(0, 0, 0); }
  /* 100vw + the 260px parking offset, so a cloud that starts fully
     off the left edge also exits fully off the right one. */
  to   { transform: translate3d(calc(100vw + 260px), 0, 0); }
}
.confirmcloud {
  animation: confirmdrift linear infinite;
  /* will-change is deliberately NOT set. Four permanently-promoted
     layers cost memory on a phone for an effect this subtle, and
     translate3d already gets its own layer while animating. */
}
/* THE DELAYS ARE A FRACTION OF EACH CLOUD'S OWN DURATION, not a
   round number of seconds.

   The first version used -12s/-48s/-71s/-95s against durations of
   83-127s, which put three of the four in the same part of their
   cycle: the sky showed ONE cloud at 1440px where four are drawn.
   A delay only means something relative to the duration it belongs
   to, and these durations differ by half.

   So each starts at roughly 10%, 35%, 60% and 85% of its own
   journey, which spreads them across the width at any moment while
   the prime durations keep them from ever lining up again. */
.confirmcloud--a { animation-duration: 83s;  animation-delay: -8s; }   /* ~10% */
.confirmcloud--b { animation-duration: 97s;  animation-delay: -34s; }  /* ~35% */
.confirmcloud--c { animation-duration: 113s; animation-delay: -68s; }  /* ~60% */
.confirmcloud--d { animation-duration: 127s; animation-delay: -108s; } /* ~85% */
.confirmcloud--e { animation-duration: 89s;  animation-delay: -42s; }  /* ~47% */
.confirmcloud--f { animation-duration: 103s; animation-delay: -76s; }  /* ~74% */
/* FOUR MORE, 29 Oct 2026. Guillaume, on the email page: "You and
   add some more clouds", then "a few more clouds". Durations stay
   PRIME and distinct (79/109/131/137) for the same reason the first
   six are: shared or harmonically related periods make the clouds
   visibly re-form the same arrangement on a cycle. */
.confirmcloud--g { animation-duration: 79s;  animation-delay: -20s; }  /* ~25% */
.confirmcloud--h { animation-duration: 109s; animation-delay: -60s; }  /* ~55% */
.confirmcloud--i { animation-duration: 131s; animation-delay: -118s; } /* ~90% */
.confirmcloud--j { animation-duration: 137s; animation-delay: -22s; }  /* ~16% */

/* EACH CLOUD IS PARKED OFF THE LEFT EDGE and its journey is the full
   width plus its own size, so it enters and leaves cleanly rather
   than appearing and vanishing mid-sky. The `left` offsets above are
   replaced by a uniform -260px start; SPREAD now comes from the
   negative animation-delays, which is what makes the four sit apart
   at any given moment instead of queueing. */
.confirmcloud { left: -260px; right: auto; }
.confirmcloud--a { width: 190px; height: 38px; top: 4%; }
.confirmcloud--a::before { width: 92px;  height: 62px; top: -28px; left: 30px; }
.confirmcloud--a::after  { width: 64px;  height: 48px; top: -20px; left: 104px; }
.confirmcloud--b { width: 150px; height: 32px; top: 9%; }
.confirmcloud--b::before { width: 76px;  height: 52px; top: -24px; left: 22px; }
.confirmcloud--b::after  { width: 52px;  height: 40px; top: -16px; left: 82px; }
.confirmcloud--c { width: 210px; height: 40px; bottom: 5%; }
.confirmcloud--c::before { width: 104px; height: 68px; top: -32px; left: 34px; }
.confirmcloud--c::after  { width: 70px;  height: 50px; top: -22px; left: 120px; }
.confirmcloud--d { width: 160px; height: 34px; bottom: 11%; }
.confirmcloud--d::before { width: 80px;  height: 54px; top: -26px; left: 24px; }
.confirmcloud--d::after  { width: 56px;  height: 42px; top: -18px; left: 88px; }
/* SIX, NOT FOUR. With the panel occupying the middle of the sky the
   clouds are only visible in the bands above and below it, and four
   spread across a 113-second cycle left those bands empty most of
   the time -- measured at 1440px, the band above the panel was
   uniform sky with a single cloud at the far edge. Two more, at
   different heights again, keep both bands occupied without
   crowding: the pack asks for low density, and this is still well
   under what the mockup draws. */
.confirmcloud--e { width: 175px; height: 36px; top: 6%; }
.confirmcloud--e::before { width: 86px;  height: 58px; top: -26px; left: 28px; }
.confirmcloud--e::after  { width: 60px;  height: 44px; top: -18px; left: 96px; }
.confirmcloud--f { width: 140px; height: 30px; bottom: 7%; }
.confirmcloud--f::before { width: 70px;  height: 48px; top: -22px; left: 20px; }
.confirmcloud--f::after  { width: 48px;  height: 36px; top: -14px; left: 76px; }
/* TEN, AND THE FOUR NEW ONES GO WHERE THE SIX DID NOT. 29 Oct 2026.

   The panel on THIS page is left-aligned and capped at 44rem, so
   unlike the confirmation page the sky is open down its whole right
   side as well as above and below -- a band the six were never
   placed for. These four sit at heights the others do not use
   (18/32/44/68%), which fills that column without raising density
   in the bands that were already working.

   Sizes vary deliberately: a row of same-sized clouds reads as a
   pattern rather than as sky. */
.confirmcloud--g { width: 120px; height: 26px; top: 18%; }
.confirmcloud--g::before { width: 60px;  height: 42px; top: -20px; left: 18px; }
.confirmcloud--g::after  { width: 42px;  height: 32px; top: -12px; left: 66px; }
.confirmcloud--h { width: 185px; height: 38px; top: 32%; }
.confirmcloud--h::before { width: 90px;  height: 60px; top: -28px; left: 30px; }
.confirmcloud--h::after  { width: 62px;  height: 46px; top: -18px; left: 100px; }
.confirmcloud--i { width: 135px; height: 28px; top: 44%; }
.confirmcloud--i::before { width: 68px;  height: 46px; top: -20px; left: 20px; }
.confirmcloud--i::after  { width: 46px;  height: 34px; top: -13px; left: 74px; }
.confirmcloud--j { width: 165px; height: 34px; top: 68%; }
.confirmcloud--j::before { width: 82px;  height: 56px; top: -26px; left: 26px; }
.confirmcloud--j::after  { width: 58px;  height: 42px; top: -17px; left: 92px; }
/* At phone width the panel fills the sky, so clouds would only peek
   out as fragments at the edges. The mockup's phone frame shows a
   clean sky band instead. */
@media (max-width: 48rem) {
  .confirmcloud--b, .confirmcloud--d { display: none; }
  /* AND ALL FOUR OF THE NEW ONES. g-j were placed in the OPEN
     COLUMN to the right of this page's left-aligned panel, which
     exists only above 48rem -- below it the panel spans the width
     and they would cross behind it, surfacing as fragments at both
     edges. That is the exact effect this block was written to
     prevent, so they follow the same rule rather than an exception
     to it. */
  .confirmcloud--g, .confirmcloud--h,
  .confirmcloud--i, .confirmcloud--j { display: none; }
}

/* THE TICK. Navy on pale mint, 48px — comfortably clear of the
   24px minimum for a decorative mark and matching the mockup's
   proportion against the heading. */
.confirmtick {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  width: 48px;
  height: 48px;
  border-radius: 50%;
  background: var(--sky-tick-bg);
  color: var(--sky-tick-fg);
  margin-bottom: 1.25rem;
}

/* ONE BREAKPOINT, 54rem, AND IT MUST MATCH .hero-status.

   This panel is also a .hero-status, which goes to two columns at
   54rem. Declaring the confirm layout at 60rem left a 6rem band in
   which .hero-status had made two columns and the rules below had
   not, so the artwork sat in column 2 while the single-column
   overrides still applied — which is what put a second patch of sky
   inside the cream panel at 563px.

   Everything that places a child therefore keys off 54rem too. */
@media (min-width: 54rem) {
  .confirmsky .confirmpanel {
    grid-template-columns: minmax(0, 0.86fr) minmax(0, 1fr);
    gap: clamp(1.5rem, 3vw, 2.5rem);
  }
  .confirmsky .flowsplit-main { max-width: 27rem; }
}

/* The mockup's heading is larger than this page's default and sits
   tight to the lede under it. */
.confirmsky .flow-h1 {
  font-size: clamp(2rem, 1.3rem + 2.4vw, 3rem);
  margin-bottom: 0.5rem;
}
.confirmsky .flow-lede { margin-top: 0; }

/* The italic thank-you line, between the lede and the actions. */
.confirm-thanks {
  font-style: italic;
  max-width: var(--measure);
  margin: 1rem 0 1.5rem;
}

/* The arrow on the primary action, as the mockup draws it. */
.confirm-arrow { margin-left: 0.4em; }

/* The disclosure is the LAST line now, under the actions, and is no
   longer a tinted panel: on a cream card inside a blue sky, a third
   surface was one too many. It keeps its weight by being the only
   small-print line there. */
.confirmsky .confirm-note {
  background: none;
  padding: 0;
  margin: 1.5rem 0 0;
  font-size: 0.9375rem;
  color: var(--ink-soft);
}

/* ---- The decorative script phrase --------------------------------

   The board sets a handwritten phrase near the foot of each panel.
   This is that phrase as TEXT, not as lettering: the script assets in
   the repo (drive-well-script.svg, and safer-drivers-script.svg
   before it) are stroked in cream for the hero and would be
   invisible on this cream page, and real text scales, reflows,
   translates and can be measured for contrast.

   IT IS NOT THE TAGLINE. This phrase comes from .ScriptMark, is set
   per page by the server, and belongs to the quote-flow boards. The
   two are deliberately separate — do not fold one into the other
   without checking those boards.

   THE THREE-PHRASE STATE THIS DESCRIBED IS OVER, 10 Oct 2026. From
   8 Oct this comment said the hero and the footer both read "Drive
   well, it pays"; from 9 Oct it said they disagreed, because
   pre-launch-gate item 6 step 4 moved the footers to "Drive well. It
   could pay." while the hero's SVG lettering kept the old wording for
   want of five glyphs. Canvas v153 removed the hero lettering
   entirely — "it stays in the footer" — so there are TWO
   phrases now, not three: this .scriptmark, which belongs to the
   quote-flow boards, and the footer tagline. They remain deliberately
   separate.

   aria-hidden because it carries no essential content — it is never
   a heading, a control or an explanation. --ink-soft on --page is
   8.33:1 light and 10.12:1 dark, so it passes as text even though
   nothing depends on it doing so. */
.scriptmark {
  display: flex;
  flex-direction: column;
  align-items: flex-start;
  gap: 0.25rem;
  margin: 2.5rem 0 0;
  max-width: 18rem;
}
.scriptmark-text {
  font-size: 1.0625rem;
  font-weight: 600;
  font-style: italic;
  line-height: 1.35;
  letter-spacing: -0.01em;
  color: var(--ink-soft);
}
/* flex: none and an explicit width. The parent is a flex COLUMN, so
   without this the swoosh is stretched to the column's width by the
   default align-items and renders four times its intended size —
   which is exactly what it did on the result page's narrow art
   column before this rule existed. */
.scriptmark-swoosh {
  display: block;
  flex: none;
  width: 7.5rem;
  height: auto;
  max-width: 100%;
}

/* ---- Holding pages ----------------------------------------------- */
.holding { max-width: 34rem; }
.holding-note {
  margin: 0;
  padding: 1rem 1.125rem;
  background: var(--panel);
  border-radius: var(--radius);
  color: var(--ink);
  font-size: 1.0625rem;
}

/* ===================================================================
   /get-in-touch — THE CONTACT FORM
   ===================================================================

   BOARDS: Contact.dc.html (desktop 1280) and ContactPhone.dc.html
   (phone 390), canvas 1791920675-8e53. Exported to
   docs/design/boards/project/ with their hashes.

   REBUILT 13 OCT 2026. The page shipped on 11 Oct as a heading, an
   address and two lists, styled with .holding--contact,
   .contact-address, .contact-note, .contact-note--legal, .contact-h2
   and .contact-list. ALL SIX ARE GONE — removed, not left behind — and
   that is the point rather than tidying: an orphan rule for a class no
   template uses is a rule the next reader has to prove is dead, and
   this file has 10,000 lines of history already. The old page's
   wording survives where it belongs, in the template's comment and in
   git.

   BOARD GEOMETRY, measured off both boards and reproduced here:

     column           680px centred      (desktop)     1fr with 20px gutters (phone)
     page padding     72px / 80px        (desktop)     36px / 32px           (phone)
     eyebrow          13px, .1em, upper, --brand
     h1               56px / 1.05 / -.03em              40px (phone)
     lede             21px / 1.5                        18px (phone)
     can't-do panel   --panel, radius 12, pad 12/16, 32px below (24 phone)
     card             --paper, radius 22, pad 36                 (24 phone)
     card heading     24px / 900                                 (21px phone)
     label            15px / 700, 6px below
     control          52px tall, 2px border, radius 12
     textarea         150px tall
     hint             13px, --ink-soft
     name+email       1fr 1fr, gap 16, 22px below     (stacked on phone)
     send             pad 17/24, 18px / 800, radius 12, full width
     fine print       13px, 12px above
     alt address      15px, 16px above, 14px above a 1px rule

   TWO THINGS THE BOARDS DO NOT MODEL, so they are decided here and
   the boards' silence is not permission (design-canvas, "What the
   boards cannot tell you"):

     * DARK MODE. Every colour below is a token, so the page follows
       the viewer's theme with no second block. The two literals the
       boards use that have no token — the dashed conditional outline
       and the orange dot — are derived from --accent and --cta.
     * ERROR AND SUCCESS STATES. The boards draw neither. They reuse
       .result.ok / .result.err, which are the flow's tested panels.

   NO LAYOUT SHIFT. #contact-slot is an empty div above the fields with
   no min-height and no margin of its own, so an error or success
   message pushes the fields down as one block and reflows none of
   them. Nothing on this page loads late — no image, no font swap
   affecting a box, no injected region. */

/* The column. 680px is the boards' figure, converted: 42.5rem.
   Deliberately the SAME measure the 11 Oct page used, which came from
   the /privacy fix on 10 Oct ("feels like its mobile" at 34rem) — so
   the board's 680 and this project's own reading measure already
   agreed, and nothing had to be reconciled. */
.contactpage {
  max-width: 42.5rem;
  /* CENTRED. .wrap is 75rem, so without this the column sits hard
     against the left gutter — which is exactly the fault found on
     3 Oct 2026, when a base rule after a media query overrode the
     margin-inline that centred the explainer. There is no media query
     over this one, for that reason. */
  margin-inline: auto;
}

/* The eyebrow. text-transform rather than typed capitals, so a screen
   reader says "Contact" and not "C-O-N-T-A-C-T". */
.contact-eyebrow {
  margin: 0;
  font-size: 0.8125rem;      /* 13px */
  font-weight: 700;
  letter-spacing: 0.1em;
  text-transform: uppercase;
  /* --accent is the purple in both themes and measures 7.54:1 on
     --page in light, 9.85:1 in dark. The boards' --brand is the same
     purple. */
  color: var(--accent);
}

/* .flow-h1 already carries the board's line-height, tracking and
   weight. This adds the board's larger size — 56px desktop, 40px
   phone — because .flow-h1 tops out at 48px for pages that also carry
   a form, and this page's h1 sits above a card rather than above
   fields. */
.contact-h1 {
  margin-block: 0.875rem 0.875rem;   /* 14px / 14px, both boards */
  font-size: clamp(2.5rem, 1.9rem + 2.4vw, 3.5rem);  /* 40px -> 56px */
}

/* THE ORANGE FULL STOP. Real punctuation, so NOT aria-hidden: the
   sentence reads correctly with it. --cta is the orange and carries no
   information, so its contrast is not a WCAG question — the word it
   follows is --ink at 15.55:1. */
.contact-dot { color: var(--cta); }

.contact-lede {
  /* The board sets the lede at full column width, not at the 34rem
     reading measure .flow-lede caps it to. Two lines at 680px is what
     both boards draw. */
  max-width: none;
  margin-bottom: 1.125rem;          /* 18px, both boards */
  font-size: 1.3125rem;             /* 21px */
  /* --ink, NOT --ink-soft. The boards set this paragraph in the full
     ink colour (`color: var(--ink)`), unlike the muted ledes elsewhere
     in the flow, and it carries the reply-time commitment — the one
     promise in the intro — so it is not set in the quieter colour. */
  color: var(--ink);
}
@media (max-width: 30rem) {
  .contact-lede { font-size: 1.125rem; }   /* 18px, phone board */
}

/* ---- The standfirst ---------------------------------------------

   Added 1 Nov 2026, boards first (canvas versions 280 and 281). It
   sits between the lede and the "can't handle claims" panel.

   --ink-soft, NOT --ink, and that is the difference from the lede
   directly above it. The lede carries the reply-time promise and both
   boards set it in full ink for that reason; this paragraph is
   context, so it takes the quieter colour the boards draw it in and
   does not compete with the commitment above it.

   Measured: --ink-soft (#56407A) on --page (#FFF9F1) is 7.9:1, well
   past AA for body text at this size.

   Full column width like the lede, for the same reason — the board
   sets this column at 680px and does not apply the 34rem reading
   measure inside it. */
.contact-standfirst {
  max-width: none;
  margin: 0 0 1.125rem;             /* 18px, both boards */
  font-size: 1.0625rem;             /* 17px, desktop board */
  line-height: 1.6;
  color: var(--ink-soft);
}
@media (max-width: 30rem) {
  .contact-standfirst { font-size: 1rem; }   /* 16px, phone board */
}

/* ---- "We can't handle claims" -----------------------------------

   ABOVE THE FORM, which is the design and not a preference: both
   boards' comments say it was "moved ABOVE the form so it is read
   before anyone types". Somebody with a claim should learn we cannot
   take it before they write it out. */
.contact-cantdo {
  display: flex;
  gap: 0.75rem;                     /* 12px */
  align-items: flex-start;
  margin-bottom: 2rem;              /* 32px desktop board */
  padding: 0.75rem 1rem;            /* 12px / 16px */
  background: var(--panel);
  border-radius: 0.75rem;           /* 12px */
  color: var(--ink);
}
@media (max-width: 30rem) {
  .contact-cantdo { margin-bottom: 1.5rem; }   /* 24px, phone board */
}
.contact-cantdo-dot {
  flex: none;
  margin-top: 0.4375rem;            /* 7px, aligns with the first line */
  width: 9px;
  height: 9px;
  border-radius: 50%;
  background: var(--cta);
}
.contact-cantdo-text {
  margin: 0;
  /* min-width:0 so a long word wraps inside the flex child instead of
     widening it, which is what pushes a horizontal scrollbar onto a
     narrow phone. */
  min-width: 0;
  font-size: 0.875rem;              /* 14px, both boards */
  line-height: 1.5;
}

/* ---- The card ---------------------------------------------------- */
.contactcard {
  padding: 2.25rem;                 /* 36px, desktop board */
  background: var(--paper);
  border-radius: 1.375rem;          /* 22px */
  /* The boards' two-layer shadow. The second layer is the lift; the
     first is the hairline that keeps the card's edge visible against
     --page, which in dark mode is the only thing separating 1.14:1 of
     surfaces. */
  box-shadow: 0 1px 0 rgba(20, 43, 73, 0.04),
              0 18px 40px rgba(20, 43, 73, 0.12);
}
/* The navy-tinted lift is invisible on a dark ground and the hairline
   disappears with it. A 1px --line edge is what reads there, and --line
   is #2F4A68 against a #101D30 page in dark mode.

   BOTH SELECTORS, for the usual reason: the media query is the no-JS
   fallback and the attribute is the explicit Dark choice. */
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .contactcard {
    box-shadow: 0 0 0 1px var(--line), 0 18px 40px rgba(0, 0, 0, 0.45);
  }
}
:root[data-theme="dark"] .contactcard {
  box-shadow: 0 0 0 1px var(--line), 0 18px 40px rgba(0, 0, 0, 0.45);
}
@media (max-width: 30rem) {
  .contactcard { padding: 1.5rem; }      /* 24px, phone board */
}

.contactcard-h {
  margin: 0 0 1.25rem;              /* 20px below, both boards */
  font-size: 1.5rem;                /* 24px desktop */
  font-weight: 800;
  letter-spacing: -0.01em;
  line-height: 1.2;
}
@media (max-width: 30rem) {
  .contactcard-h { font-size: 1.3125rem; }   /* 21px, phone board */
}

/* ---- Fields -------------------------------------------------------

   .field already gives a block label and the label gap. These set the
   boards' label weight and size, and the boards' 18px rhythm between
   fields — .field's own 1.5rem is the quote flow's rhythm and this
   card is tighter. */
.contact-field { margin-block: 0 1.125rem; }     /* 18px, both boards */
.contact-field label {
  font-size: 0.9375rem;             /* 15px */
  font-weight: 700;
  /* The boards' 6px, which is --label-gap's value, named rather than
     inherited so a change to the questionnaire's scale does not move
     this card. */
  margin-bottom: 0.375rem;
}

/* The boards draw every control at 52px with a 2px border and a 12px
   radius. The base rules give 44px, 1px and --radius (12px), so only
   the height and the border need stating.

   2px, NOT 1px, AND IT IS NOT ONLY THE BOARD. --field-border is
   14.27:1 on --paper, well past WCAG 1.4.11's 3:1, so 1px was already
   compliant — but this card's fields sit on --paper inside a --page
   surround, where a 1px edge at 390px was the thing Guillaume called
   out on the quote steps. The board's 2px is the
   better choice and the reason to prefer it is legibility, not
   obedience. */
.contact-field select,
.contact-field input[type="text"],
.contact-field input[type="email"] {
  min-height: 52px;
  border-width: 2px;
}
.contact-field textarea {
  display: block;
  width: 100%;
  /* ================================================================
     150px ON BOTH BOARDS, AND `height` RATHER THAN `min-height`.
     ================================================================

     THIS WAS MEASURED, NOT ASSUMED. With `min-height: 150px` and the
     template's rows="6", the box came out 205px against the board's
     154px — because `rows` sizes the control in LINES and this site's
     base text is 18-19px (body's clamp), not the 16px the boards are
     drawn at. Six lines at a 29.45px computed line-height is 177px of
     text plus 24px of padding and 4px of border. min-height could
     only ever raise that, never bring it back.

     So the height comes from the BOARD and rows is left as the
     no-stylesheet fallback. 150px is a little over four lines at this
     text size, which is still a comfortable box to write a paragraph
     in, and `resize: vertical` below means anybody writing more can
     drag it.

     THE LARGER BASE TEXT IS NOT A BUG TO CORRECT. 18-19px body text
     is this site's standing choice and is why the other controls come
     out 56px against the boards' 52px — a bigger touch target and
     bigger text in it. Matching the board by shrinking the text would
     be reading the board as a pixel spec rather than as a design. The
     textarea is different only because `rows` made its overshoot
     three times larger than anything else's. */
  height: 150px;
  padding: var(--ctrl-pad-y) var(--ctrl-pad-x);
  font: inherit;
  color: var(--ink);
  background: var(--paper);
  border: 2px solid var(--field-border);
  border-radius: var(--radius);
  /* VERTICAL ONLY. A textarea that resizes horizontally can be dragged
     wider than its column, which is a horizontal scrollbar a person
     creates for themselves. */
  resize: vertical;
}
.contact-field textarea:focus-visible {
  outline: 2px solid var(--accent);
  outline-offset: 2px;
}
.contact-field textarea[aria-invalid="true"],
.contact-field select[aria-invalid="true"] {
  border-color: var(--err-ink);
}

/* The hints under the message box. 13px --ink-soft, which is 8.33:1 on
   --page and 7.56:1 on --panel — both past 4.5:1, measured in the
   token block rather than eyeballed here. */
.contact-hint {
  margin: 0.375rem 0 0;             /* 6px, both boards */
  font-size: 0.8125rem;             /* 13px */
  line-height: 1.45;
  color: var(--ink-soft);
}

/* ---- The conditional "My data" note ------------------------------

   Both boards draw it with a DASHED OUTLINE to mark it conditional —
   "dashed outline = conditional". That is a board annotation, not a
   design: a dashed ring around a live paragraph reads as an error
   state or a drop target to anyone who has not read the board. So it
   renders as the boards' panel fill with a solid left rule, the same
   device the rest of this site uses for a qualifying note (.notyet,
   .contact-note--legal before it).

   REPORTED AS A DEPARTURE rather than slipped in. If Guillaume wants
   the dashed ring on the live page it is one line, but the board's
   dash is almost certainly annotation. */
.contact-datanote {
  margin-top: 0.625rem;             /* 10px, both boards */
  padding: 0.625rem 0.75rem;        /* 10px / 12px */
  background: var(--panel);
  border-left: 3px solid var(--accent);
  border-radius: 0 0.625rem 0.625rem 0;
}
.contact-datanote-p {
  margin: 0;
  font-size: 0.875rem;              /* 14px, both boards */
  line-height: 1.45;
  color: var(--ink);
}
.contact-datanote-p + .contact-datanote-p { margin-top: 0.5rem; }

/* ---- Name and email ---------------------------------------------

   TWO COLUMNS ON THE DESKTOP BOARD (`1fr 1fr`, gap 16), STACKED ON
   THE PHONE BOARD, which draws them as two separate blocks.

   THE BREAKPOINT IS ON THE CARD, NOT THE VIEWPORT, in effect: 34rem
   (544px) is where two 52px controls plus a 16px gap stop being wider
   than ~240px each inside a 24px-padded card. Below it they stack.
   min-width:0 on the children so a long address wraps inside its
   column rather than widening the grid. */
.contact-pair {
  display: grid;
  gap: 1rem;                        /* 16px, desktop board */
}
.contact-pair > * { min-width: 0; }
/* The fields inside already contribute 18px below; the boards put 22px
   under the pair, so this makes up the 4px rather than restating the
   whole figure — the alternative is zeroing the child's margin and
   setting 22px here, which breaks if the pair ever holds one field. */
.contact-pair { margin-bottom: 0.25rem; }
@media (min-width: 34rem) {
  .contact-pair { grid-template-columns: 1fr 1fr; }
}

/* ---- Send -------------------------------------------------------

   FULL WIDTH AT EVERY SIZE, which both boards draw (`display: block`
   with no max-width) and which the base `button` rule would undo at
   34rem (`width: auto; min-width: 14rem`). The phone CTA on the
   homepage was once 44px short of its column for exactly this reason
   — found by measuring at 390, not by looking at 1280 — so the
   override is explicit.

   THE RAISED ORANGE SLAB comes from .btn-primary, which is the
   project's tested implementation of the boards' `0 4px 0 #C8682A`
   edge plus lift, with the hover, active and focus states and the
   reduced-motion opt-out already in it. Not re-declared here: a second
   copy of a raised button is a second thing to drift. */
.contact-send {
  width: 100%;
  min-width: 0;
  margin-top: 1.375rem;             /* 22px, both boards */
  padding: 1.0625rem 1.5rem;        /* 17px / 24px */
  font-size: 1.125rem;              /* 18px */
  font-weight: 800;
  border-radius: 0.75rem;           /* 12px */
}
@media (min-width: 34rem) {
  /* Beats the base rule's `width: auto; min-width: 14rem`. */
  .contact-send { width: 100%; min-width: 0; }
}

/* ====================================================================
   THE FOCUS RING ON THIS BUTTON IS PURPLE, NOT CREAM. MEASURED.
   ====================================================================

   .btn-primary's ring is `3px solid #FFF9F1` — cream — and the comment
   where it is defined says exactly why: "these buttons sit against a
   hero and a 2px ring disappears; cream is the only thing in
   the palette that reads on the hero AND on the cream band where the
   other copies live."

   NEITHER OF THOSE GROUNDS HOLDS HERE. This button is inside a WHITE
   card on a CREAM page. Measured, with the same method as the token
   block:

     cream ring on the white card   1.05:1
     cream ring on the orange fill  1.79:1
     cream ring on the cream page   1.00:1

   WCAG 2.4.11 / 1.4.11 want 3:1 for a focus indicator. A cream ring
   here is, to within a rounding error, invisible — so a keyboard user
   tabbing to Send would see the button they are about to press
   indicated by nothing. That is the kind of fault this project has
   already been told it keeps shipping, and it was found by measuring
   the ring rather than by looking at a screenshot, where a 1.05:1
   edge reads as "no ring drawn yet".

   --accent is the site's standard focus colour and measures 7.89:1 on
   the white card and 7.54:1 on the cream page. 3px and 3px of offset
   are kept from .btn-primary so the ring is the same SHAPE as every
   other primary button's — only the colour changes, and only where
   the colour was chosen for a background this button does not have.

   THE HERO'S BUTTONS ARE NOT TOUCHED. .btn-primary keeps its cream
   ring everywhere else, which is correct everywhere else. */
.contact-send:focus-visible {
  outline: 3px solid var(--accent);
  outline-offset: 3px;
}

.contact-fineprint {
  margin: 0.75rem 0 0;              /* 12px, both boards */
  font-size: 0.8125rem;             /* 13px */
  line-height: 1.5;
  color: var(--ink-soft);
}
.contact-fineprint a { color: var(--accent); }

/* The alternative address, under a hairline. Both boards: 16px above,
   14px of padding over a 1px --line rule, 15px text. */
.contact-alt {
  margin: 1rem 0 0;
  padding-top: 0.875rem;
  border-top: 1px solid var(--line);
  font-size: 0.9375rem;
  line-height: 1.5;
  /* BREAKS RATHER THAN OVERFLOWS. At 320px "info@coroute.co" plus the
     lead-in is wider than the column; without this it pushes a
     horizontal scrollbar onto the narrowest phones, which is the exact
     fault .contact-address carried this guard for on the 11 Oct page. */
  overflow-wrap: anywhere;
}
.contact-alt a { color: var(--accent); font-weight: 700; }

/* ---- The two swapped states --------------------------------------

   .result.ok and .result.err carry the colours and the padding. These
   add the card's rhythm: the slot sits above the fields, so a message
   needs space below it and none above. */
.contact-ok,
.contact-err { margin: 0 0 1.25rem; }
.contact-ok .result-lead,
.contact-err .result-lead { margin: 0; }
.contact-ok .result-aside,
.contact-err .result-aside { margin: 0.5rem 0 0; }

/* ---- Buttons in the board's shape --------------------------------

   The base `button` rule above already gets the colours right: the
   orange fill with DARK ink (8.67:1) and a --accent border, because
   the orange fill alone is 1.79:1 against cream and fails WCAG
   1.4.11's 3:1 for the control's own boundary. These add the board's
   wider shape and the inline arrow. */
.btn {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  gap: 0.625rem;
}
.btn-wide { width: 100%; }
@media (min-width: 34rem) {
  .btn-wide { width: 100%; max-width: 24rem; }
}
.btn svg { flex: none; }

/* ---- The result page's head, board 02 panel 02 -------------------

   THE PRICE, THE SOURCE LINE AND THE BREAKDOWN NEVER SIT OVER
   ARTWORK. The board bleeds the illustration in from the top right
   and sets the script phrase over it. An earlier round on this
   project put a headline over a city skyline at 1.00:1, so the rule
   here is structural rather than a promise: the artwork lives in its
   own grid column, the copy lives in the other, and neither enters
   the other's box. The only thing that changes at any width is
   whether the artwork column exists at all.

   Below 64rem it does not: display:none, so the price and the
   breakdown have the full width and nothing is competing for it. The
   image is a CSS-sized <img> inside that column rather than a
   background, so the width/height attributes still reserve its box.

   NOTHING ABOUT THE ORDER OF THE PAGE CHANGES. §7.3 fixes the order
   of the eight elements and this only puts the first three of them
   in a narrower column at desktop. */
.resulthead { min-width: 0; }
.resulthead-copy { min-width: 0; }
/* Hidden below the two-column breakpoint AND not fetched: an <img>
   inside a display:none parent is still requested by every browser,
   so the source is inside the media query with the layout. */
.resulthead-art { display: none; }

@media (min-width: 64rem) {
  /* THE ARTWORK COLUMN MUST NOT DRIVE THE ROW HEIGHT.

     As a plain two-column grid this row is as tall as its tallest
     child, and the artwork is taller than the price and source line
     beside it — which pushed a ~250px band of empty cream between
     the source line and whatever came next. Measured, not guessed.
     (The panel it measured against was .improved, removed 9 Oct
     2026; the block below the head is now .breakdown, and the fix
     is unchanged because the cause was the artwork's height and not
     the panel's identity.)

     The fix is to take the artwork out of flow: the grid row is
     sized by the copy alone and the artwork is absolutely
     positioned into the column the grid reserves for it. The
     reserved column is still a real grid track, so the copy never
     runs under the artwork and no text is ever composited over it.

     The artwork may extend BELOW the copy's row, which is what the
     board draws — it bleeds past the price into the space beside
     the panel underneath. overflow is not hidden on the page, so
     nothing clips; it simply overlaps a region that is empty cream
     at every width where this rule applies (checked at 1024, 1280,
     1440, 1920 and 2560). */
  .resulthead {
    display: grid;
    grid-template-columns: minmax(0, 1fr) 20rem;
    gap: 2.5rem;
    align-items: start;
    position: relative;
  }
  .resulthead-art {
    display: block;
    min-width: 0;
    position: absolute;
    top: 0;
    right: 0;
    width: 20rem;
  }
  /* SCOPED TO THE PICTURE, not to every img in the column. As
     `.resulthead-art img` this rule also matched the script
     phrase's decorative swoosh, which is an <img> in the same box:
     the swoosh was given the 5:4 frame, object-fit:cover and the
     cream background, and rendered as a 320x256 cream card in dark
     mode. Found by looking at the dark-mode screenshot, not by
     reading the rule. */
  .resulthead-art > picture img {
    display: block;
    width: 100%;
    height: auto;
    /* 5:4 landscape band cropped to the BOTTOM of the portrait, the
       same crop the sidebar uses: the asset's top third is empty
       sky, and the board shows roofline and street, not sky.
       Not flipped — object-position moves the vertical axis only. */
    aspect-ratio: 5 / 4;
    object-fit: cover;
    object-position: 50% 100%;
    border-radius: var(--radius-panel);
    background: #FFF9F1;   /* the artwork's own cream field */
  }
  /* The script phrase sits BELOW the image on the flat page, not
     over it. The board draws it over sky; sky is the one part of
     this asset with no reliable contrast floor, and the pack says so
     explicitly ("no global minimum contrast can be certified across
     the generated illustration"). */
  .resulthead-art .scriptmark {
    margin-top: 1rem;
    margin-left: auto;
    align-items: flex-end;
    max-width: 100%;
  }
  .resulthead-art .scriptmark-text { text-align: right; }
}

/* The "check your inbox" fragment, board 03 panel 04's content in the
   live region on the waitlist page. Two paragraphs rather than one so
   the spam-folder line reads as a follow-up rather than as part of
   the same sentence. */
.result-lead { margin: 0; }
.result-aside {
  margin: 0.5rem 0 0;
  font-size: 0.9375rem;
  /* Deliberately NOT --ink-soft: the .result.ok panel sets its own
     --ok-ink, and mixing the two puts one of the two lines at a
     ratio nobody measured. Both lines inherit the panel's ink. */
  opacity: 1;
}

/* ===================================================================
   THE TWO-STEP QUESTIONNAIRE — boards 01 and 02
   ===================================================================

   Added 22 Sep 2026 with the About you / Your vehicle restructure.

   Everything here is measured against the surfaces it actually sits
   on, not against the design board's own pale values: the boards'
   "generated pale input borders and small text are not implementation
   measurements" (the specification says so in as many words), and this
   project has already shipped one round of 1.25:1 control edges.

   Control edges use --field-border, which is 7.89:1 on --paper light
   and 10.18:1 dark — comfortably past WCAG 1.4.11's 3:1 — and never
   the decorative --line. */

/* ---- Progress indicator -----------------------------------------

   THE ACTIVE STEP IS NOT MARKED BY COLOUR ALONE (WCAG 1.4.1). It
   carries aria-current="step", a FILLED marker where the others are
   outlined, and 700 weight where the others are 500. Any one of the
   three would do on its own; all three are cheap.

   Horizontal on desktop with a connecting rule; at narrow widths the
   items wrap onto their own lines rather than truncating. The final
   label — "Your advisory quote" — must never be clipped: it is the
   word "advisory" that keeps the page legal, and it is the one that
   an ellipsis would eat. */
.progress { margin: 0 0 2rem; }
.progress-list {
  display: flex;
  flex-wrap: wrap;
  gap: 0.5rem 1.25rem;
  margin: 0;
  padding: 0;
  list-style: none;
}
.progress-step {
  display: flex;
  align-items: center;
  min-height: 2.75rem;   /* 44px effective target for the linked ones */
}

/* ---- THE CONNECTING RULE BETWEEN STEPS ---------------------------

   From the quote-journey pack's mockups, 28 Oct 2026, which draw a
   line joining the three circles. The list was a plain flex row of
   three items with a gap and nothing between them, so the steps read
   as three labels rather than as one journey.

   CSS ONLY, NO MARKUP CHANGE. The list is shared by all three steps
   (_chrome.html) and already carries .is-done / .is-active, so the
   connector is a ::before on every step after the first and needs no
   new element. That also keeps it out of the accessibility tree,
   where a decorative line has nothing to say -- the <ol> and the
   aria-current on the active step already carry the structure.

   IT ONLY DRAWS WHERE THE ROW DID NOT WRAP. .progress-list wraps at
   narrow widths, and a connector on the first item of a second row
   would point leftwards at nothing. There is no CSS selector for
   "did this item wrap", so the line is drawn only at >=48rem, where
   the three fit on one line -- measured, not assumed: at 360px the
   list is three rows, at 768px it is one.

   The DONE colour differs from the pending one, so the line reports
   progress as well as joining: a completed span is the accent, an
   upcoming span is the faint rule. */
@media (min-width: 48rem) {
  .progress-step + .progress-step { position: relative; }
  .progress-step + .progress-step::before {
    content: "";
    flex: 0 0 auto;
    width: 1.75rem;
    height: 2px;
    margin-right: 0.75rem;
    border-radius: 1px;
    /* --line, the project's decorative-rule token. It is redefined
       for dark mode, so the connector follows the theme without a
       rule of its own. (--rule does not exist: a fallback would have
       silently hard-coded a light colour into dark mode.) */
    background: var(--line);
  }
  /* A step that is done is reached THROUGH its connector, so the
     connector before it is part of the completed path. */
  .progress-step.is-done::before,
  .progress-step.is-active::before { background: var(--accent); }
}
.progress-item,
.progress-link {
  display: inline-flex;
  align-items: center;
  gap: 0.5rem;
  font-size: 1rem;
  line-height: 1.3;
  color: var(--ink-soft);
  font-weight: 500;
}
.progress-link { color: var(--accent); }
.progress-link:focus-visible {
  /* var(--accent), NOT var(--focus). --focus has never been defined in
     this stylesheet — only --focus-on-purple — and an invalid var()
     makes the whole `outline` shorthand compute to its initial value,
     which is outline-style:none. These three rules therefore drew
     NOTHING, measured at 0 changed pixels on a real Tab.

     This one was the worst of the three: being more specific than the
     a:focus-visible rule above, it REMOVED a ring the link would
     otherwise have had. A rule intended to improve a focus indicator
     deleted it. */
  outline: 3px solid var(--accent);
  outline-offset: 3px;
  border-radius: 0.25rem;
}
/* The marker: outlined by default, FILLED when active or done. The
   shape difference is what survives a colour-blind reading. */
.progress-marker {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  width: 1.75rem;
  height: 1.75rem;
  flex: none;
  border-radius: 999px;
  border: 2px solid var(--field-border);
  background: transparent;
  color: var(--ink);
  font-size: 0.875rem;
  font-weight: 700;
}
.progress-step.is-active .progress-marker,
.progress-step.is-done .progress-marker {
  background: var(--brand);
  border-color: var(--brand);
  color: var(--on-brand);   /* white on #6624C9 = 7.89:1 */
}
.progress-step.is-active .progress-label {
  font-weight: 700;
  color: var(--ink);
}
.progress-label { min-width: 0; }

/* The phone counter and the phone bar: present in the markup on every
   width, drawn only below 30rem. Not display:none on a WRAPPER — each
   is its own element, so there is one copy of the progress in the DOM
   rather than a desktop copy and a phone copy. See _chrome.html. */
.progress-count,
.progress-bar { display: none; }

/* THE RESULT PAGE SHOWS THE INDICATOR ON PHONE ONLY, because its two
   boards disagree: ResultPhone draws "Your advisory quote / Step 3 of
   3" over a full bar, and Result (desktop, same screen) draws nothing
   between the nav and the h1.

   display:none rather than visually-hidden, deliberately. An
   indicator nobody can see but every screen reader announces would
   match NEITHER board and would be the worse of the two outcomes.
   Above 30rem this page simply has no progress indicator, which is
   what its desktop board shows.

   WHICH BOARD IS RIGHT IS AN OPEN QUESTION and is flagged rather than
   settled here — see the note in quote_result.html.

   NO FIFTH BREAKPOINT. A first draft wrote `min-width: 30.0625rem` to
   be the exact complement of the 30rem max-width block, which is
   precisely the kind of invented width R1 exists to stop. It is
   hidden by default and REVEALED inside the existing max-width: 30rem
   block below, so only the four allowed breakpoints appear. */
.flowmain-result .progress { display: none; }

/* ---- PHONE: "Step n of 3" and a bar, not three circles -----------

   THE DESKTOP INDICATOR DID NOT FIT. Three labelled circles at 390px
   wrapped onto two lines with "Your advisory quote" dropping below —
   Guillaume found it on his own phone, 6 Oct 2026. The wrap itself was
   correct (truncating "advisory" is forbidden and still is); the shape
   was the problem.

   MainPhone, VehiclePhone and ResultPhone on the canvas all draw the
   same replacement, and all three carry the same comment: "three
   labelled circles don't fit at 390px, so phone shows 'Step n of 3' +
   the current step + a progress bar." Board leads, page follows.

   THE BOARD DRAWS NO CIRCLES AT ALL HERE, not even for the current
   step: bold label on the left, soft "Step n of 3" on the right, a
   6px bar under both. A first build kept the markers and they formed
   a second row of two naked circles under the label, pushing the bar
   down — visible the moment the board and the page were put side by
   side, invisible to every text check. The markers are therefore
   hidden on phone, and the STEP NUMBER survives in words: "Step 1 of
   3" says everything the circle said, which is why dropping the shape
   does not reintroduce the WCAG 1.4.1 problem the desktop markers
   exist to solve. Nothing here is distinguished by colour alone.

   NO NEW BREAKPOINT. This is the existing 30rem (480px) block — one
   of the four the layout contract allows (R1). 30rem rather than
   48rem because a 700px tablet has ample room for three circles.

   THE INACTIVE LABELS ARE NOT DELETED. They go to the visually-hidden
   pattern, so the list still announces three items and the "About
   you" Back button is still reachable. display:none would be a
   content change dressed as a style and would silently remove the
   only route back from step 2. */
@media (max-width: 30rem) {
  /* The bar is a SIBLING of the <ol> inside .progress, so the gap
     between them comes from .progress's own gap (R7 — spacing from a
     parent, never a margin on the component). */
  .progress {
    display: flex;
    flex-direction: column;
    gap: 0.5rem;
  }
  /* One visible row. The non-current steps collapse to nothing
     visible, so no wrapping and no second row of markers. */
  .progress-list { gap: 0; }
  .progress-item, .progress-link { font-size: 0.9375rem; }

  /* The current step is the row: label left, count right. */
  .progress-step.is-active { flex: 1 1 100%; }
  .progress-step.is-active .progress-item,
  .progress-step.is-active .progress-link { width: 100%; }
  .progress-step.is-active .progress-label { font-weight: 700; }
  .progress-count {
    display: block;
    margin-left: auto;
    color: var(--ink-soft);
    font-weight: 400;
    white-space: nowrap;
  }

  /* No circles on phone — the board draws none. The number they
     carried is in the words "Step n of 3". */
  .progress-marker { display: none; }

  /* THE OTHER STEPS ARE REMOVED, NOT CLIPPED, AND THAT IS A FIX.

     The first build used the visually-hidden pattern here, to keep
     "About you" and "Your vehicle" in the accessibility tree, with a
     comment claiming display:none "would take the Back button with
     them". THE COMMENT WAS WRONG and the rule it justified was an
     accessibility fault:

       * Wrong, because step 2's real Back is the full-width button in
         .stepactions at the foot of the form (measured 353x56 at
         393px). The breadcrumb is a second route to the same submit,
         not the only one. Step 1 has no Back at all and needs none.

       * A fault, because .progress-back is a <button>. Clipped to 1px
         it stays FOCUSABLE: measured at 393px, el.focus() succeeded
         on a rect of 1.0 x 44.0. A keyboard user tabs into a control
         they cannot see — WCAG 2.4.7 — and a touch user has a 1px
         target where 44px is the floor.

     display:none removes it from the tab order and the tree together,
     which is the consistent outcome. What the phone loses is the
     breadcrumb ROUTE; what it keeps is the footer Back button and the
     step context in words ("Step 2 of 3"). The boards draw no
     breadcrumbs at phone width at all, so this is also what they
     show. Above 30rem nothing here applies and both routes exist. */
  .progress-step:not(.is-active) { display: none; }

  /* THE BAR IS ONE SEGMENT PER STEP, NOT A FILLED PERCENTAGE, AND
     THE CSP IS WHY. `style="width: 33%"` on a fill element is dropped
     by the browser: this site is served style-src 'self' with no
     'unsafe-inline', so the attribute reaches the HTML and never
     reaches the CSSOM. Measured 6 Oct: element.style.width was "" and
     the computed width 350px of 350px — a step-1 page drawing a bar
     filled to 100%.

     Segments need no percentage. Three steps make three equal
     thirds; a fourth step would make quarters with no edit here. */
  .progress-bar {
    display: flex;
    /* NO GAP. The segments are an implementation of a single
       continuous track — the board draws one rounded bar with one
       filled portion, and a 2px gap between segments drew two faint
       notches in it, visible the moment board and page were put side
       by side. The segments must not announce themselves. */
    gap: 0;
    height: 0.375rem;
    border-radius: 999px;
    overflow: hidden;
  }
  .progress-bar-seg {
    flex: 1;
    /* --line is the board's track colour. The bar is decorative in
       the strict sense — it repeats "Step n of 3" and carries nothing
       that is not already in words — so the 3:1 rule that bans --line
       on control edges does not apply. */
    background: var(--line);
  }
  /* --accent, NOT --brand, and dark mode is why. The board draws the
     fill in --brand, which is right in light mode: #6624C9 on the
     #E0D4F2 track is a strong, obvious bar.

     In dark mode the two tokens converge — --brand moves to #5B21B6
     and --line to #4A3570 — and the filled portion measured 1.15:1
     against the unfilled track. A progress bar nobody can read is
     not a progress bar.

     --accent is the token that already exists for this problem: it
     is #6624C9 in light mode, IDENTICAL to --brand, so the board's
     appearance is unchanged there, and lavender #D8C4F5 in dark,
     which the design system chose precisely because #6624C9 fails
     against the dark page. The raised secondary button uses the same
     token for the same reason (see the btnraise note on the canvas).

     The bar is aria-hidden decoration and the step is stated in
     words, so this is not a WCAG failure either way — it is a bar
     that should be legible if it is drawn at all. */
  .progress-bar-seg.is-filled { background: var(--accent); }

  /* The result page's indicator exists only here — see the note on
     .flowmain-result .progress above. */
  .flowmain-result .progress { display: flex; }
}

/* ---- Field groups ------------------------------------------------

   Each group is a real <fieldset> with a real <legend>, so a screen
   reader announces "Personal details, First name, edit" rather than
   an orphan input. The border is reset because the grouping is
   conveyed by the legend and the spacing, and a default fieldset
   border at every nesting level would draw boxes inside boxes. */
.fieldgroup {
  border: 0;
  margin: 0 0 var(--section-gap);
  padding: 0;
  min-width: 0;   /* fieldsets default to min-content; this lets it shrink */
}
.fieldgroup-legend {
  padding: 0;
  margin-bottom: var(--heading-gap);
  font-size: 1.125rem;
  font-weight: 700;
  color: var(--ink);
  line-height: 1.3;
}
/* Nested fieldsets — the radio groups inside a group — carry their
   question as the legend and need no second heading weight.

   THE DEFAULT FIELDSET BORDER IS RESET HERE TOO. A <fieldset> draws a
   groove border and notches its legend into it by default; nested one
   level down that renders boxes inside boxes and makes the question
   look like a control rather than a label. The grouping is conveyed by
   the legend and the spacing, which is what a screen reader uses
   anyway — the border was never the thing doing the work. */
fieldset.field {
  border: 0;
  padding: 0;
  margin-inline: 0;
  min-width: 0;
}
.field > legend {
  padding: 0;
  margin-bottom: 0.5rem;
  font-weight: 500;
  color: var(--ink);
  /* float:none keeps the legend in flow rather than in the border
     notch, so the margin-bottom above actually applies. */
  float: none;
  width: auto;
}
.fieldgroup .field { margin-block: var(--field-gap); }
/* THE FIRST THING AFTER A LEGEND OWES NOTHING. The legend's own
   margin-bottom is the "section heading -> first field" distance in
   the scale; a top margin on the first field adds to it, because a
   legend is not a block sibling that margins collapse against. The
   result measured 30px at 393 where the board draws 16, and 28px at
   1280 where the desktop had always drawn 28 — so the DESKTOP value
   is the sum and must keep summing, while the phone's must not.

   Zeroing it here would therefore change desktop. It is zeroed in
   the phone block only, where the board is explicit; desktop keeps
   the 28px it is signed off at. */
/* (The zeroing itself is in the 30rem block — see
   "THE FIRST FIELD AFTER A LEGEND" there.) */
.optional { font-weight: 400; color: var(--ink-soft); }

/* Two or three controls on one line above 40rem, stacked below.
   `min-width: 0` on the children is what stops a long value forcing
   the row wider than its container — the classic grid overflow. */
/* A paired row. `align-items: start` so a help line under ONE of the
   pair does not stretch its sibling's box to match — "When the car was
   built, not when you bought it" sits under Year of manufacture and
   Doors has no help at all, which made the row look uneven.

   The controls now top-align, so the two selects sit on the same line
   whatever hangs below them. */
/* CONTAINER-DRIVEN, for the same reason the radio cards are: a media
   query asks how wide the WINDOW is, and since the grid engages at
   56rem that stops matching how wide this ROW is.

   Measured at 993px after the grid change: the window said "three
   columns" while the form cell was 506px, and three 10rem minimums
   plus two 1rem gaps need 512 — so Last name wrapped to a line of its
   own and the name row looked broken.

   auto-fit with a floor sizes from the space actually there. An 8rem
   floor fits three name fields in 506px and still drops to two, then
   one, as the column narrows. */
.fieldrow {
  display: grid;
  gap: 0 1rem;
  align-items: start;
  grid-template-columns: repeat(auto-fit, minmax(min(8rem, 100%), 1fr));
}
.fieldrow > .field { min-width: 0; }
/* R4: a fieldrow holds 2-3 inputs that are ONE CONCEPT, and "such a
   row stacks below 30rem". It did not — `minmax(min(8rem, 100%), 1fr)`
   still fits two 8rem tracks in 350px, so Title sat beside First name
   and Town beside Postcode at 390px, where MainPhone stacks both.

   Three-part rows (day/month/year) are the exception the rule is
   written around and keep their row: a date in one column would be
   three full-width boxes reading as three unrelated questions. They
   opt out with .fieldrow-keep. */
@media (max-width: 30rem) {
  /* A STACKED FIELDROW MUST MEASURE THE SAME AS TWO PLAIN FIELDS.
     It did not: plain siblings collapse their 14px margins to 14px,
     but GRID ITEMS DO NOT COLLAPSE MARGINS, so the row's children
     summed to 28 — twice the board — while the fields on either
     side of the row were correct. Two different rhythms in one
     column, from one property nobody looks at.

     So the row's children give up their margins and the row's own
     gap sets the distance, which is R7 (spacing from a parent's
     gap, never a component's margin) doing exactly its job. */
  .fieldrow:not(.fieldrow-keep) {
    grid-template-columns: 1fr;
    gap: var(--field-gap) 0;
  }
  .fieldrow:not(.fieldrow-keep) > .field { margin-block: 0; }

  /* A KEPT ROW IS THE BOARD'S FLEX ROW, NOT TWO EQUAL HALVES.
     MainPhone draws Town/Postcode as `display: flex; gap: 16px` with
     Town `flex: 1 1 0; min-width: 0` and Postcode `width: 120px;
     flex: none` — a postcode is six or seven characters and does not
     want half the screen, while a town name does want the rest.

     The default .fieldrow grid gives two equal tracks, which put
     "Town or city" in 169px for no reason. */
  .fieldrow-keep {
    display: flex;
    gap: var(--heading-gap);
  }
  .fieldrow-keep > .field { margin-block: 0; flex: 1 1 0; min-width: 0; }
  .fieldrow-keep > .field-narrow { flex: none; width: 7.5rem; }  /* 120px */

  /* THE FIRST FIELD AFTER A LEGEND OWES NOTHING, on phone.
     --heading-gap is the legend's own margin-bottom and is meant to
     BE the "section heading -> first field" distance. A top margin
     on the first field adds to it (a legend is not a block sibling,
     so nothing collapses), and the pair measured 30px at 393 where
     the board draws 16.

     Phone only: at 1280 the same sum produces 28px, which is what
     the desktop is signed off at and must keep. */
  .fieldgroup > .field:first-of-type,
  .fieldgroup > .radiocards:first-of-type,
  .fieldgroup > .fieldrow:first-of-type { margin-top: 0; }
  .fieldgroup > .fieldrow:first-of-type > .field:first-child { margin-top: 0; }
}
/* .field-narrow caps a control that should not run the full column.
   16rem = 256px, which sat 10px off the 266px radio cards beside it —
   close enough to look like a mistake rather than a decision.

   It now matches the grid the cards use: one column of the two-column
   layout below 48rem, one of three above. A select and a card in the
   same section are then the same width, which is the whole point of
   having one grid.

   The £ value field keeps its own content-sized width (11ch) and is
   deliberately NOT in this rule: it holds 4-6 digits and looking
   deliberate means looking small. */
.field-narrow { max-width: 16rem; }

/* ...BUT THE CAP IS FOR THE CONTROL, NOT FOR THE PROSE UNDER IT.
   17 Oct 2026. Guillaume, on the live vehicle step: "looks like the
   text.. Enter your best... should be spreadout more along the width?"

   .field-narrow is on the WRAPPER, so the cap applied to the hint as
   well as the input. "Enter your best estimate of the car's current
   value. We can't look this up automatically yet." broke onto THREE
   lines, while the fuel hint right above it -- same .help class, no
   narrow wrapper -- ran on one.

   TWO WRONG FIXES, BOTH MEASURED BEFORE THEY SHIPPED:

   1. `max-width: 100%` on the help does NOTHING. The percentage
      resolves against the 16rem parent, so it stays 256px. Measured:
      256px, 3 lines, unchanged.
   2. Taking the cap off .field-narrow and moving it to the children
      breaks the three responsive rules below, which size the wrapper
      to a GRID COLUMN so long labels do not wrap. That cap is load
      bearing; it is not a plain 16rem.
   3. `max-width: 40rem` alone OVERFLOWS the column at phone width --
      558px of text in a 353px column, a horizontal-scroll bug.

   100cqw is the column's own width, so the help can exceed the
   wrapper but never the column. Measured at three widths:
     353px -> 353px, 2 lines, fits
     660px -> 558px, 1 line,  fits
    1120px -> 558px, 1 line,  fits
   The input keeps its own 11ch at every width. */
.fieldgroup, .flowsplit-main { container-type: inline-size; }
.field-narrow > .help { width: max-content; max-width: 100cqw; }

/* A select sized to its CONTENT, for the short ones.

   Title rendered at 49px — a bare dropdown arrow with no room for the
   word "Mr", which is the shortest option it offers. It was in a
   three-column .fieldrow and took a third of a capped 544px column.

   width:auto lets it size to its longest option plus the arrow;
   min-width keeps it from collapsing when that option is two
   characters. */
/* A select that must not be squeezed by a neighbouring column. Year of
   manufacture wrapped its label to two lines when .field-narrow put it
   in a third of the column; it holds a four-digit year and needs no
   more room than that, but its LABEL needs one line. */
.field-year { max-width: none; }
.field-year select { width: auto; min-width: 9rem; }

.field-auto { max-width: none; }
.field-auto select { width: auto; min-width: 7.5rem; }

/* PHONE: NOTHING IS NARROWER THAN ANYTHING ELSE. MainPhone draws the
   whole Personal details group as `grid-template-columns: 1fr` and
   gives no field its own width — Title and Postcode are the same
   353px as First name.

   Measured at 393px before this rule: Title 138px (from .field-auto's
   `width: auto`, which sizes the select to "Mr") and Postcode 256px
   (from .field-narrow's 16rem cap). Both are good rules at desktop
   width, where a full-width select for "Mr" looks absurd and R9 says
   a control should match its answer. At 390px there is one column and
   nothing to balance against, so a short control just reads as an
   unfinished row.

   R9 IS NOT BEING OVERRULED. "A control width matches the expected
   answer" is about not making a day field as wide as a surname when
   they sit side by side. In a single column with a 350px measure
   there is no side by side, and the board says so. Above 30rem both
   rules apply exactly as before. */
@media (max-width: 30rem) {
  .field-auto select,
  .field-year select { width: 100%; }
  .field-narrow { max-width: none; }
}
/* THE LABEL SETS THE FLOOR, NOT THE CONTROL.

   Capping at a grid column squeezed long labels onto two lines —
   "Years of no-claims discount", "How long have you held it?" and
   "Your car's current value" all wrapped while their controls had room
   to spare. A wrapped label above a short control reads as a broken
   row, which is most of what "looks like shit" was pointing at.

   17rem is measured, not guessed: it is what the longest label in this
   class needs on one line at the page's base size. min() then keeps it
   from exceeding a grid column on a narrow screen, and the control
   inside is still its own size — the £ field remains 11ch. */
@media (min-width: 30rem) {
  .field-narrow { max-width: max(17rem, calc((100% - 0.75rem) / 2)); }
}
@media (min-width: 48rem) {
  .field-narrow { max-width: max(17rem, calc((100% - 1.5rem) / 3)); }
}

/* ---- Date of birth ----------------------------------------------

   Three labelled boxes, not a date picker. Native date inputs differ
   between browsers, are awkward on several screen readers, and impose
   a locale-dependent order people get wrong. Three visible labels are
   unambiguous, and each one is a real <label for>. */
/* NO margin-block OF ITS OWN. It used to carry `margin-block: 1.25rem`
   from when it was a .field nested inside Personal details. It is now
   its own section — MainPhone marks it "SECTION 2 of 5", a sibling of
   Personal details and Where do you live? — and a 20px override left
   it the one section in the column spaced differently from the other
   four (measured 20px where its neighbours had 28 on phone and 32 on
   desktop). It takes .fieldgroup's --section-gap like every other
   section now. */
.dobgroup { border: 0; padding: 0; min-width: 0; }
/* The Day/Month/Year row is a group of options on one concept, so it
   is --option-gap: 12px desktop, 10px phone. The board draws 12 at
   390 — see the note at the phone block for why the row keeps the
   desktop value there. */
.dobrow { display: flex; flex-wrap: wrap; gap: 0.75rem; }
.dobpart { display: flex; flex-direction: column; }
.dobpart label {
  font-size: 0.9375rem;
  font-weight: 500;
  margin-bottom: 0.25rem;
}
.dobpart input { width: 4.5rem; }
.dobpart-year input { width: 6rem; }

/* ---- Radio cards -------------------------------------------------

   The board's card-shaped choices. They are ordinary radio inputs
   inside labels — no ARIA, no JavaScript, no div pretending to be a
   control — so keyboard behaviour, grouping and announcement are the
   browser's and cannot be got wrong.

   MINIMUM 44px EFFECTIVE TOUCH TARGET, which the spec requires. The
   whole card is the label, so the target is the card, not the 17px
   circle.

   THE SELECTED STATE IS NOT COLOUR ALONE: the border goes to 2px and
   the background changes, and the radio dot itself is still drawn by
   the browser. */
/* ONE GRID FOR EVERY RADIO GROUP ON BOTH STEPS.

   Measured before this change, at 660px: Transmission cards were
   266px, Fuel type 127px, Owner and Keeper 173px — three different
   widths on one screen, nothing lining up vertically, reading as four
   unrelated components rather than one form. Fuel was also 148px tall
   against everything else's 83px.

   THE SYSTEM IS A FIXED COLUMN COUNT PER BREAKPOINT, NOT auto-fit.
   auto-fit sizes columns from the group's own item count, which is
   precisely why four groups with 2, 4, 3 and 3 items produced four
   different card widths. A fixed count makes the column width a
   property of the PAGE, so a card in one group is the same width as a
   card in another.

   Two columns below 48rem, three above. The 4-item fuel group
   therefore runs 3 + 1 on a wide screen — which was the thing the old
   comment here was trying to avoid, and it was solving it the wrong
   way round: an even row of four narrow cards that match nothing else
   on the page is worse than a 3 + 1 that lines up with every other
   group. The stray card is the same width as its siblings, so it reads
   as a continuation rather than an afterthought.

   .radiocards-tight is kept as an alias and no longer differs: the
   distinction between "tight" and normal groups was what produced the
   mismatch. It stays so the templates do not all have to change, and
   so a future reader finds this note rather than wondering where it
   went. */
.radiocards {
  display: grid;
  gap: var(--option-gap);
  margin: 0;
  padding: 0;
  border: 0;
  /* Every card in a row is as tall as the tallest, so a description on
     one option (fuel's "Hybrid, or something else.") cannot leave the
     row ragged. This is why the fix is `stretch` rather than giving
     every option a description it does not need.

     stretch alone was NOT enough, and the measurement caught it: it
     equalises within a grid ROW, and fuel's fourth card wraps onto a
     row of its own, where it had nothing to match. Measured 84 and 128
     in the same group. grid-auto-rows: 1fr makes every row of the grid
     the same height as the tallest, so the whole group is uniform
     however it wraps. */
  align-items: stretch;
  grid-auto-rows: 1fr;
}
/* THE COLUMN COUNT FOLLOWS THE CONTAINER, NOT THE VIEWPORT.

   Media queries ask how wide the WINDOW is; these cards care how wide
   their COLUMN is, and since the grid engages at 56rem those two stop
   agreeing. Measured with viewport queries: at 900px the window says
   "three columns" while the form cell is 420px, giving 132px cards —
   narrower than the 173px this change set out to fix.

   auto-fit with a floor sizes from the actual space available. The
   floor stops it collapsing: it is wide enough for "Someone else" on
   one line, so a group gets fewer, wider cards rather than more,
   cramped ones. Every group on the page still shares one column width,
   because they all share the same container.

   THE FLOOR IS 10rem BECAUSE THE CONTRACT SAYS SO (quote-layout R8).
   It was 8.5rem, and the contract said 10rem, and nobody noticed until
   the board review put the two side by side. Measured 26 Sep 2026
   before changing it: 8.5rem, 10rem and min(10rem, 100%) give
   BYTE-IDENTICAL tracks at 320, 360, 480, 896, 993 and 1280 — the
   container binds before the floor does at every width we test, so
   this is a no-op in rendering and a real gain in the contract and the
   site saying the same thing.

   min(…, 100%) and not a bare 10rem: a bare floor cannot go below
   160px, so a container narrower than that overflows instead of
   wrapping. That is a sideways-scroll bug waiting for a narrower
   column. */
.radiocards,
.radiocards-tight {
  grid-template-columns: repeat(auto-fit, minmax(min(10rem, 100%), 1fr));
}
/* .radiocards-stacked — ONE COLUMN AT PHONE WIDTH ONLY.
   Added 6 Oct 2026 from MainPhone/VehiclePhone, which draw two
   specific groups as `grid-template-columns: 1fr` where every other
   group on the same boards keeps `repeat(auto-fit, minmax(10rem,
   1fr))`: step 1's licence question and step 2's "Do you have a car
   yet?".

   THE TWO HAVE SOMETHING IN COMMON — every option carries a
   description ("You're still learning to drive.", "Your own, or a
   family car you'll drive."). At 390px the shared grid gives two
   169px tracks, and a described option in 169px runs to four or five
   lines, so the pair comes out tall, ragged and much harder to scan
   than three full-width rows. Undescribed groups (excess, doors,
   fuel) are short enough that 2-up is better, and the boards keep
   them 2-up.

   THIS IS NOT A SECOND CARD SYSTEM, which R8 forbids and which is the
   bug that produced the layout contract. It is the SAME grid with its
   track count pinned for two groups at one width; the card itself —
   padding, radius, border, min-height, the equal-height rule — is
   untouched, and above 30rem the modifier does nothing at all, so
   every desktop board is unaffected. */
@media (max-width: 30rem) {
  .radiocards-stacked { grid-template-columns: 1fr; }
}
/* .field .radiocard, not .radiocard: these cards are <label> elements
   inside a .field fieldset, so `.field label { display: block }` at the
   top of this file (specificity 0,2,0) beat a bare `.radiocard` (0,1,0)
   and forced every card to stack its circle ABOVE its text.
   THAT is why the radio sat in the top-left corner detached from its
   label — measured: input at y=23, body at y=44 inside an 83px card,
   with align-items:center having no effect because the box was never
   flex to begin with. */
.field .radiocard,
.radiocard {
  display: flex;
  /* CENTRE, not flex-start. The circle used to sit in the card's
     top-left corner with the label below and to the right of it, so
     the control and its text did not read as one thing. Centring puts
     the circle on the text, which is what makes the whole card read as
     the target it already was. */
  align-items: center;
  gap: 0.625rem;
  min-height: 3.5rem;
  padding: var(--card-pad-y) var(--card-pad-x);
  background: var(--paper);
  border: 1px solid var(--field-border);
  border-radius: var(--radius);
  cursor: pointer;
  min-width: 0;
}
.radiocard input { flex: none; margin-top: 0; }
.radiocard-body { display: flex; flex-direction: column; gap: 0.125rem; min-width: 0; }
.radiocard-title { font-weight: 600; color: var(--ink); line-height: 1.3; }
.radiocard-note {
  font-size: 0.9375rem;
  color: var(--ink-soft);   /* 7.56:1 on --panel, 8.33:1 on --page */
  line-height: 1.4;
}
/* --accent, NOT --brand, and dark mode is why — the same swap, for
   the same reason, as .progress-bar-seg.is-filled above.

   SELECTING A CARD MADE ITS EDGE WORSE IN DARK MODE. Measured on the
   running page, border against the page behind it:

     unselected  --field-border #D8C4F5   10.18:1
     selected    --brand        #5B21B6    2.06:1   <- the fault

   So the state meant to stand out was the one that receded, and it
   failed the 3:1 WCAG 1.4.11 needs for a control boundary while the
   UNSELECTED state passed comfortably. Confirmed by eye on a dark
   render before changing anything: against two siblings with bright
   lavender edges, the selected card's edge read as DIMMER than
   theirs, and the only thing still marking selection was the small
   filled radio dot.

   --accent is #6624C9 in light mode, IDENTICAL to --brand, so the
   board's light-mode appearance is unchanged — 7.54:1 against the
   page there, exactly as before. In dark mode it is lavender
   #D8C4F5, which measures 11.56:1 against the page and now makes
   selection the STRONGEST edge on screen rather than the weakest.

   The fill still goes to --panel, which lifts 1.37:1 off the page in
   dark mode; the fill was never what carried this state, the border
   is, which is why only the border needed to move. */
.radiocard:has(input:checked) {
  border-width: 2px;
  border-color: var(--accent);
  background: var(--panel);
  /* The 1px→2px border would move the text by a pixel and shift the
     row. Absorbed in the padding so nothing reflows on selection. */
  padding: calc(var(--card-pad-y) - 1px) calc(var(--card-pad-x) - 1px);
}
/* The focus ring is on the CARD, not just the 17px circle, so
   keyboard focus is visible at the size of the thing being operated.
   :has() is not relied on for this — :focus-within works everywhere
   the rest of the page does.

   THIS COMMENT WAS FALSE UNTIL 22 SEP 2026 and is the CLAUDE.md
   failure mode exactly: the rule below used var(--focus), which is not
   defined, so the card drew no ring at all and the only thing a
   keyboard user saw was the browser's default ring on the 17px radio —
   the very thing the comment says this rule replaces. Measured: 152
   changed pixels confined to the radio, in a 41x41 crop. Fixed by
   using --accent, the token every working focus rule here uses. */
.radiocard:focus-within {
  outline: 3px solid var(--accent);
  outline-offset: 2px;
}

/* ---- £ prefix ----------------------------------------------------

   The symbol is aria-hidden decoration inside the box; the LABEL
   already says what the number means. A "£" announced before every
   keystroke is noise. */
/* SIZED TO WHAT IT HOLDS. A car value is 4-6 digits, and with
   thousands separators that is up to "1,500,000" — nine characters —
   so the box is sized for that plus the £ mark. It used to be a 192px
   box floating in a 607px column beside full-width controls above it,
   which read as broken.

   ch, not rem: the width is a statement about the content, and ch
   tracks the digit width of whatever font is actually resolved.

   THE OLD COMMENT SAID "about 9 characters wide" AND THAT WAS NOT
   WHAT RENDERED. 15 Oct 2026, from an external review: "your car
   current value field — make it wider and maybe add automatic
   formatting with thousands separator — would help avoiding typos".
   Measured at 1280: the box was 128px and the <input> inside it
   90.1px, because `min-width: 8rem` (128px) was beating `width: 11ch`
   and the £ mark plus padding took 38px of what was left. Ninety
   pixels is nearer SEVEN characters, so 4500 and 45000 looked alike —
   and this field feeds the advisory price directly, so a typo here is
   a wrong number rather than a validation message.

   Both values raised together, and min-width must stay BELOW the ch
   width or it silently wins again. quote.js groups the digits on blur;
   the server already strips separators (profile.go digits()). */
.inputprefix {
  display: flex;
  align-items: stretch;
  border: 1px solid var(--field-border);
  border-radius: var(--radius);
  background: var(--paper);
  overflow: hidden;
  width: 15ch;
  min-width: 11rem;
  max-width: 100%;
}
.inputprefix-mark {
  display: flex;
  align-items: center;
  padding-inline: 0.75rem;
  background: var(--panel);
  color: var(--ink);
  font-weight: 600;
  border-right: 1px solid var(--field-border);
}
.inputprefix input {
  border: 0;
  border-radius: 0;
  flex: 1;
  min-width: 0;
}
.inputprefix:focus-within {
  /* Same dead-token bug (see .progress-link). This one left the £
     vehicle-value field — a primary control on step 2 — with no ring
     on its own box; the only visible change was the input's inner
     ring, which the prefix box visually crops. */
  outline: 3px solid var(--accent);
  outline-offset: 2px;
}

/* ---- Address: search, then the manual fields -------------------

   The manual fields sit in the same <details> treatment as "My car
   isn't listed" (.fallback) so the two escape routes on the form look
   like the same kind of thing. The details is rendered OPEN, and
   quote.js closes it only when the search is running and the fields
   are empty — with JavaScript off it is simply the address field.

   The search's own label is VISIBLE, unlike the car combobox's
   off-screen .combo-label: there, a visible label already sits on the
   no-JS <select>; here, the search box is the only thing labelled
   "Find your address". */
.help-above { margin: 0 0 0.5rem; }
.address-manual { margin-top: 0.75rem; }
.address-manual-fields { padding: 0.25rem 1rem 1rem; }
.address-manual-fields > .field:first-child { margin-top: 0; }

/* ---- The omission callout ----------------------------------------

   "No overnight-parking question." Guillaume asked for this
   HIGHLIGHTED, and §7.7 explains why it earns the space: it is the
   only element that says what we refuse to do, and the refusal is the
   product.

   THE ARTWORK IS DECORATION BESIDE IT AND NEVER A CONTROL. The pack
   is explicit that the parking scene "must not become a selectable
   parking answer or imply that parking is being evaluated" — so there
   is no input in this box and nothing in it is clickable.

   The deeper --statement-bg used to be the surface behind the
   homepage's "Not your marital status." panel, which is the same
   argument made in the same voice. That panel lost its box on 4 Oct
   2026 and sits on the page ground now, so THIS is the token's only
   remaining user — it is no longer a shared surface, and a future
   change to it affects this box alone.

   --statement-text on --statement-bg measures 11.53:1 light /
   9.41:1 dark as of 23 Oct 2026, when the token moved to the v2
   flow boards' green-grey. The 10.18:1 recorded here before was the
   v1 lilac and no longer describes anything on the page. */
.noask {
  display: flex;
  align-items: center;
  gap: 1rem;
  margin: 2rem 0;
  padding: 1.25rem 1.5rem;
  background: var(--statement-bg);
  color: var(--statement-text);
  border-radius: var(--radius-panel);
}
/* ---- The email section at the foot of step 2 ---------------------

   Guillaume, 8 Nov 2026: "it should be a box like on the other page",
   then "it has to be clear that they need to enter the email to see
   the quote … check the design pannel.. we basically need that box
   section at the end", with a screenshot of QuoteEmailV2.

   MEASURED OFF THAT BOARD rather than approximated — it was read from
   the live canvas (version 1794135708-b835):

     container   --paper fill, 2px solid --brand, 20px radius,
                 26px/28px padding
     h1          44px / 1.04 / 700 / -0.02em      -> .emailbox-h
     lede        19px / 1.5 / --ink-soft           -> .emailbox-lede
     label       15px / 600, 6px below
     input       2px --brand border, 12px radius, 13/14px padding,
                 52px min-height
     fineprint   13px / 1.55 / --ink-soft

   THE HEADING SCALE IS THE ONE DIFFERENCE, AND IT IS DELIBERATE. The
   board draws this as a PAGE of its own, so its heading is an h1 at
   44px. Here the section sits at the foot of step 2, under "Tell us
   about the car." — which is already the page's h1. A second 44px
   heading would compete with it and would be the wrong element: this
   is an h2. It takes the flow's existing h2 scale instead.

   ⚠️ --paper AND A BORDER, NOT A TINTED PANEL. An earlier version of
   this rule used --panel with no border, reasoning from .basis. The
   board is explicit and it is right: the white box with a hard navy
   edge reads as a THING TO FILL IN, where a tinted panel reads as
   something to read — which is exactly what .noask and .basis are.
   This box is the only one on the page that wants an answer. */
.emailbox-head { margin-top: 2.5rem; }
.emailbox-h {
  margin: 0 0 0.5rem;
  font-size: 1.75rem;
  font-weight: 700;
  line-height: 1.1;
  letter-spacing: -0.02em;
}
.emailbox-lede {
  margin: 0 0 1.25rem;
  font-size: 1.0625rem;
  line-height: 1.5;
  color: var(--ink-soft);
  max-width: var(--measure);
}
/* ⚠️ fieldset.emailbox, NOT .emailbox, AND THE SELECTOR IS THE WHOLE
   FIX. `fieldset.field { border: 0; padding: 0 }` lives further down
   this file (the rule that makes ordinary question groups invisible
   boxes). It has the SAME specificity as a bare .emailbox and comes
   LATER, so it won: the box rendered white-on-cream with no border
   and no padding, which is how the first version shipped and what
   Guillaume saw. Matching the element as well outranks it. */
fieldset.emailbox {
  margin-block: 0 2rem;
  padding: 1.5rem;
  background: var(--paper);
  border: 2px solid var(--brand);
  border-radius: var(--radius-panel);
}
.emailbox .fineprint { margin-bottom: 0; }

/* ---- The two-line submit inside the email box --------------------

   Guillaume, 8 Nov 2026: "the orange button should be within the
   email box and it should say - Verify my email \n Show me my
   advisory quote".

   MEASURED OFF QuoteEmailV2.dc.html, which draws exactly this control:
     line 1   18px / 800
     line 2   14px / 600, 2px above
     button   full width, 12px radius, 12/20px padding, 56px min-height

   ⚠️ .btn IS display:inline-flex WITH align-items:center, SO THE TWO
   SPANS WOULD SIT SIDE BY SIDE. flex-direction: column is what stacks
   them, and the gap replaces the board's 2px margin so the two lines
   cannot drift apart if one is edited.

   ⚠️ THE 14rem min-width FROM THE BASE `button` RULE MUST NOT WIN
   HERE. That rule gives every button a sensible floor on a wide
   screen; this one is full-width inside a box, so the floor is
   irrelevant and width:100% carries it.

   LINE 2 IS NOT DIMMED. It is the more important of the two
   promises — it is the thing they came for — so it is smaller but
   full-strength ink. A faded second line would read as a caveat. */
.btn-stacked {
  flex-direction: column;
  gap: 0.125rem;
  width: 100%;
  min-height: 3.5rem;
  padding-block: 0.75rem;
  text-align: center;
}
.btn-stacked .btn-line1 { font-size: 1.125rem; font-weight: 800; line-height: 1.2; }
.btn-stacked .btn-line2 { font-size: 0.875rem; font-weight: 600; line-height: 1.25; }

/* The button is the last thing in the box, under the notice. */
.emailbox .fineprint + .btn-stacked { margin-top: 1.25rem; }

/* ---- "Open Gmail" on the check-inbox page ------------------------

   ⚠️ .btn-link EXISTS BECAUSE AN <a> DOES NOT MATCH THE BASE `button`
   ELEMENT RULE. That rule is where .btn gets its padding, font-size,
   weight and removed underline; .btn itself adds only inline-flex and
   the gap. Three .btn-secondary anchors shipped without this earlier
   today and rendered as stretched white slabs with underlined links
   inside them — Guillaume screenshotted one.

   THOSE THREE WERE TURNED INTO <button>s because they navigated
   within the site. THIS ONE STAYS AN ANCHOR: it goes to an external
   provider, in a new tab, which is what a link is for and what a form
   cannot do honestly. So the missing values are supplied here
   instead, matching the base rule rather than inventing a shape.

   THE NUMBERS ARE THE BASE RULE'S, not new ones: the same padding and
   weight every other button on the flow has. */
.btn-link {
  padding: 0.8125rem 1.25rem;
  border-radius: var(--radius);
  font-size: 1rem;
  font-weight: 700;
  line-height: 1.25;
  text-decoration: none;
  min-height: 3rem;
  box-sizing: border-box;
}
/* ⚠️ THE FILL, THE BORDER AND THE INK ALL HAVE TO BE RESTATED, AND
   EACH OMISSION SHIPPED ITS OWN FAULT — both caught by looking at the
   rendered page, neither by a test.

   FIRST PASS: no colour. "Open Gmail" rendered in rgb(0,0,238),
   browser link blue, because the generic `a { color: ... }` rule won.
   SECOND PASS: colour fixed, still no fill — a WHITE button with an
   orange shadow, reading weaker than the two plain buttons below it.

   WHY BOTH: `button { background: var(--cta); border: 2px solid
   var(--on-cta); color: var(--on-cta) }` is the BASE ELEMENT RULE
   (app.css:1021). .btn-primary adds only the shadow and the
   transition; .btn adds only inline-flex and the gap. An <a> matches
   NEITHER element rule, so it arrives with no fill, no border and no
   ink — and the three faults are the three missing declarations.

   THIS IS THE THIRD TIME TODAY this mechanism has bitten: the three
   .btn-secondary anchors (stretched white slabs), the .fineprint
   privacy link (browser blue), and this one twice over. THE RULE IS
   ALWAYS THE SAME — an anchor inherits nothing from the `button`
   element rule, so everything that rule supplies must be restated.

   --on-cta is the token named for ink ON the orange fill, 7.61:1
   against it; --cta is the fill itself. Both come from the same
   block, so the pair stays correct in dark mode. */
.btn-link.btn-primary,
.btn-link.btn-primary:hover,
.btn-link.btn-primary:visited {
  color: var(--on-cta);
  background: var(--cta);
  border: 2px solid var(--on-cta);
}
.btn-link.btn-primary:hover { background: var(--cta-hover); }
.btn-link.btn-secondary,
.btn-link.btn-secondary:hover,
.btn-link.btn-secondary:visited {
  color: var(--ink);
  background: var(--sec-fill);
  border: 2px solid var(--accent);
}
.btn-link:hover { text-decoration: none; }

.inbox-open {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: 0.75rem;
  margin: 1.5rem 0 0;
}
/* "Opens in a new tab." is said in TEXT, not only by an icon. The
   icon is aria-hidden, so without this the behaviour would be
   announced to nobody — WCAG 3.2.5's reason for warning about a new
   window at all. */
.inbox-open-note { font-size: 0.875rem; color: var(--ink-soft); }

/* ONE SECONDARY BUTTON IN THE ROW, NOT TWO. `.stepactions .btn`
   stretches its children with flex: 1 1 12rem, which is right for a
   Back/submit pair and made a lone Back span the whole column —
   measured on the rendered page, not assumed.

   ⚠️ .stepactions.stepactions-backonly, NOT .stepactions-backonly
   ALONE. Both classes are on the same element, so the single-class
   version ties with `.stepactions .btn` on specificity and LOSES on
   source order (it sits ~50 lines earlier in this file). Matching
   both classes outranks it wherever either rule ends up. This is the
   same trap as fieldset.emailbox a few rules above. */
.stepactions.stepactions-backonly .btn { flex: 0 1 auto; }
/* The notice is the last thing in the box, so it gets the gap that
   the field below it used to provide. */
.emailbox input[type="email"] + .fineprint,
.emailbox .result + .fineprint { margin-top: 1rem; }

.noask-art { flex: none; width: 120px; }
@media (min-width: 40rem) { .noask-art { width: 160px; } }
/* The pack's assets have genuine alpha and sit on this lavender
   without a tile behind them. No object-fit: cover — the whole scene
   is shown at its own aspect ratio, which is what the pack asks for
   ("Use width:100%, height:auto and preserve aspect ratio"). NOT
   FLIPPED: no transform is applied to any of these images. */
.noask-art img { display: block; width: 100%; height: auto; }
.noask-body { min-width: 0; }
.noask-h {
  margin: 0 0 0.25rem;
  font-size: 1.125rem;
  font-weight: 700;
  line-height: 1.25;
}
.noask-text { margin: 0; font-size: 1rem; line-height: 1.5; }
@media (max-width: 30rem) {
  /* On a phone the art goes ABOVE the copy rather than squeezing it
     into a 40%-width column. The pack asks for exactly this: "Put art
     below meaningful copy on mobile or omit purely decorative art
     where it creates excessive scrolling." Here it is small enough to
     keep, and stacking it reads better than a two-column squeeze. */
  .noask { flex-direction: column; align-items: flex-start; }
  .noask-art { width: 120px; }
}

/* ---- Step actions: Back and submit --------------------------------

   Both are real submit buttons. Back cannot be a link: this form
   holds a date of birth and an address, and a link would have to put
   them in a URL to carry them.

   Back is SECONDARY but not diminished — it is a full-size button
   with the same min-height, because going back to fix an answer is a
   thing people legitimately want and hiding it produces abandonment,
   not completion. */
.stepactions {
  display: flex;
  flex-wrap: wrap;
  gap: 0.75rem;
  margin-top: 2rem;
}
.stepactions .btn { flex: 1 1 12rem; min-height: 3rem; }
/* PHONE: PRIMARY FIRST, FULL WIDTH, BACK UNDERNEATH — VehiclePhone's
   own comment says exactly that, and implements it as
   `flex-direction: column-reverse` over a DOM order of Back-then-
   primary, which is the DOM order this page already has.

   column-reverse rather than reordering the markup, for two reasons.
   The DOM order is the TAB order and the screen-reader order: Back
   before Submit at the end of a form is the conventional reading, and
   `order`/`column-reverse` deliberately do not change it (a known
   WCAG 1.3.2 trap is reversing one and not the other — here nothing
   needs to move because the visual order is the one being changed).
   And the desktop row is untouched: above 30rem this rule does not
   apply, so the signed-off desktop boards keep Back on the left. */
@media (max-width: 30rem) {
  .stepactions { flex-direction: column-reverse; }
  .stepactions .btn { flex: 1 1 auto; width: 100%; }
}

/* .btn-secondary IS NOT DEFINED HERE ANY MORE, 5 Oct 2026.

   It used to carry its own fill, ink and a 1px --field-border here,
   which made it the SECOND definition of the secondary role: the
   homepage's waitlist button and step 3's decline were styled
   elsewhere and differently, and this one was the only one with a
   1px edge. Guillaume, on the live page: "it still does not look
   aligned or the same at the end of the page."

   The single definition now lives with its primary counterpart under
   "THE RAISED BUTTON TREATMENT" above, where .cta-sec, .action-decline
   and .btn-secondary are one rule. Do not re-add a fill, a border or
   a hover for .btn-secondary here — a second definition is the bug,
   not the fix. Layout (flex-basis, min-height) still belongs here; it
   is not part of the treatment. */

/* ---- Artwork panels with genuine alpha ---------------------------

   The questionnaire pack's assets are NOT the opaque cream artwork
   the hero sidebar uses. They have real transparency, so they sit
   directly on the panel with no cream tile and no crop —
   which is what .formaside-art's cover/cream treatment exists to work
   around, and is not needed here.

   .formaside-art-tall is the portrait about-you-street; the landscape
   vehicle asset uses the plain rule below it. */
.formaside-art-tall,
.formaside .formaside-art:has(img[src*="vehicle-hatchback"]) {
  margin-inline: 0;
  background: transparent;
  aspect-ratio: auto;
  padding-bottom: 1.75rem;
}
.formaside-art-tall img,
.formaside .formaside-art:has(img[src*="vehicle-hatchback"]) img {
  width: 100%;
  height: auto;
  object-fit: fill;
  object-position: initial;
}
/* The stacked-width rule above sets .formaside-art to a 16/10 frame
   with object-fit:cover, which is right for the opaque hero asset it
   was written for and wrong for these: a 2:3 portrait forced into a
   16:10 box with cover shows a sliver, and here it collapsed to
   nothing. These assets have real alpha and no cream field to crop, so
   they keep their own aspect ratio at every width.

   Declared after the media query it has to beat, because both have the
   same specificity and the later rule wins. */
@media (max-width: 55.99rem) {
  .formaside-art-tall,
  .formaside .formaside-art:has(img[src*="vehicle-hatchback"]) {
    aspect-ratio: auto;
  }
  /* STACKED, THE PANEL IS FULL WIDTH, and a 1.5:1 landscape image
     stretched across it renders 367px tall at 660px and 514px at
     900px — which is what read as "a large empty purple area" below
     the text. The panel was not over-sized; the IMAGE was, and the
     panel grew to hold it.

     Measured before: aside 569px at 660, 722px at 900. Capping the art
     the way the portrait asset is already capped brings both down to
     something proportionate to the two lines of copy above it.

     (The image itself loads fine at every width. An earlier reading of
     imgComplete:false was lazy-loading not yet triggered below the
     fold, not a failure.) */
  .formaside .formaside-art:has(img[src*="vehicle-hatchback"]) {
    max-width: 20rem;
    margin-inline: auto;
  }
  /* Portrait art full-width on a phone is a very tall block for pure
     decoration, so it is capped and centred rather than filling the
     column. The pack asks for exactly this restraint. */
  .formaside-art-tall {
    max-width: 16rem;
    margin-inline: auto;
  }
}

/* ---- The ambition panel's artwork — REMOVED 9 OCT 2026 ------------

   result-road was mapped to the .improved panel by the questionnaire
   artwork pack, and both are gone: ResultPhone.dc.html replaced that
   panel with the .savings card, which the board draws as a single
   column of figures with no illustration in it.

   The four image files (result-road-512/768.webp/png) are LEFT IN
   web/static/img/ and are now referenced by nothing. They were not
   deleted here because an unused asset costs nothing served and the
   board may yet want one; it is reported rather than decided. If the
   board stays as it is, they are the thing to delete.

   The two media queries this replaces also carried the only remaining
   use of a 40rem breakpoint on this page, which was off the four the
   quote-layout contract allows (R1). The savings card uses 30rem and
   48rem only. */

/* The "About you" breadcrumb is a <button>, not an <a> — going back
   has to post the form so the answers survive (see _chrome.html). It
   must still look and measure like the link it replaced: a button
   brings its own background, border, padding, font and cursor, none of
   which belong on a breadcrumb.

   "MEASURE LIKE THE LINK" WAS THE INTENT AND IT DID NOT HOLD until
   15 Oct 2026. An external reviewer: "when you move to the next page
   on the form, the stepper at the top gets distorted with strange
   spaces". Measured at 1280: advancing from step 1 to step 2 grew this
   item from 118.2px to 224px and shoved the two steps after it 105.8px
   and 110.6px to the right. The whole row moved, every time.

   TWO SEPARATE UA DEFAULTS CAUSED IT, and fixing either alone leaves
   the fault visible:

   1. `font: inherit` above is the SHORTHAND, so it expands to
      font-size/line-height/font-weight: inherit and lands AFTER the
      `font-size: 1rem` that .progress-item,.progress-link sets further
      up this file. Equal specificity, later wins — so the button drew
      at body's clamp(1.125rem…) = 19px while the spans beside it drew
      at 16px. Hence the three re-assertions below.

   2. The 224px was NOT the font. A flex <button> gets a UA
      `min-width` (measured: exactly 224px here) that a <span> does
      not, while the content needs ~115px. min-width:0 is the fix for
      the shift; the font only made it slightly worse.

   With both, step 2's items land within ~3.4px of step 1's — the
   residue is this rule's own underline plus tick-versus-numeral glyph
   width, which is invisible where 106px was not.

   DO NOT "SIMPLIFY" THIS BY DROPPING `font: inherit` AND KEEPING ONLY
   THE LONGHANDS. The shorthand also resets font-family, font-stretch
   and the font-variant-* group off the UA's button font, which is
   what stops this drawing in the browser's default UI face. */
.progress-back {
  background: none;
  border: 0;
  padding: 0;
  margin: 0;
  font: inherit;
  /* Re-asserted AFTER `font: inherit`, which would otherwise undo the
     .progress-item,.progress-link rule above. Keep these three in step
     with it. */
  font-size: 1rem;
  line-height: 1.3;
  font-weight: 500;
  /* The actual fix for the 106px jump: a flex <button>'s UA min-width,
     which the <span>s beside it do not get. */
  min-width: 0;
  color: var(--accent);
  cursor: pointer;
  text-align: left;
  min-height: 2.75rem;   /* the same 44px target the links have */
}
.progress-back .progress-label { text-decoration: underline; }

/* ---- Row confirmation --------------------------------------------

   A tick beside a SECTION HEADING once that whole row is satisfied.
   One per row, never one per field.

   IT USED TO BE PER FIELD, on the argument that "a tick that appears
   on everything stops carrying information" — which is true, and which
   still produced a form that looked broken: only the date of birth and
   the email were wired, so somebody who filled in their name saw
   nothing and assumed the feature was missing. A restriction nobody
   can perceive reads as an absence. Row scope keeps the restraint and
   makes it visible.

   THE COLOUR IS --ok-ink, THE THEME-AWARE SUCCESS TOKEN, re-measured
   for this placement because a tick beside a heading sits on a
   different surface from one inside a field:

     #17643A on --page (light)   6.87:1
     #8FE0AD on --page (dark)   11.82:1

   --tick-bg (#18724B), used for the filled tick elsewhere on the site,
   stays REJECTED here: 2.75:1 on dark --paper, failing WCAG 1.4.11's
   3:1 for a non-text indicator.

   NEVER COLOUR ALONE (1.4.1): the mark is a tick SHAPE and the green
   only reinforces it. The SVG is aria-hidden and the state is exposed
   through aria-invalid="false" on each control in the row — no live
   region, because announcing "valid" after every section is hostile.

   Added by quote.js only. With JavaScript off none of these classes
   ever appear. */
.validtick {
  flex: none;
  vertical-align: middle;
  /* No colour of its own: the disc is currentColor, inherited from the
     heading, so the mark and the words it confirms can never disagree
     about the theme. */
}

/* THE WHOLE HEADING GOES GREEN, not just the mark beside it.

   Guillaume asked for the section to read as confirmed at a glance
   from across the page, and a tick at the end of a line is a small
   target for that.

   NEVER COLOUR ALONE (WCAG 1.4.1). The green is added TO the tick, not
   instead of it — a green heading on its own would be exactly the way
   to get this wrong, and both are applied by the same rule so they
   cannot come apart. The class is only ever set when the tick is
   appended.

   IT IS A REAL STEP DOWN IN CONTRAST and that is worth stating: the
   heading ink is 15.55:1 on --page and --ok-ink is 6.87:1. Both pass
   4.5:1 comfortably, and the trade is deliberate — 6.87:1 at 1.125rem
   700-weight is a strong, legible green, not a pale one. Measured on
   composited pixels at the rendered size and weight, not computed from
   the token. */
.fieldgroup-legend.is-valid,
.dobgroup > legend.is-valid {
  color: var(--ok-ink);
}

/* The heading becomes an inline-flex row ONLY while it holds a tick,
   so a heading without one keeps its ordinary block layout and nothing
   reflows when the tick arrives.

   gap rather than a margin on .validtick: an inline-flex container
   collapses the whitespace between its inline children, which ate the
   space before "(optional)" the first time this was applied. */
.fieldgroup-legend:has(> .validtick),
.dobgroup > legend:has(> .validtick) {
  display: inline-flex;
  align-items: center;
  gap: 0.375rem;
}

/* No border or background change on the fields themselves. The tick is
   the signal; recolouring every input edge in a row would be five
   signals for one fact, and --field-border is load-bearing for WCAG
   1.4.11 at its measured 7.89:1. */


/* ---- Owner and registered keeper ---------------------------------

   Two questions with IDENTICAL options stacked directly on top of each
   other — "Who will own the car?" then "Who will it be registered
   to?", both Myself / Parents / Someone else. With no separation it
   read as a duplicated component or a rendering bug.

   The fix is separation plus subordination, not a different control:
   they are the same question shape and should stay so, because
   changing one would imply the answers mean different things.

   The keeper question is indented behind a rule and carries its
   explanation, so it reads as a follow-up to the owner question rather
   than a repeat of it. The rule is --line, which is decorative and
   correct here: it is not a control edge and carries no meaning that
   the indentation and the heading do not already carry. */
/* fieldset.keeperfield, not .keeperfield: `fieldset.field` resets
   border and padding to 0 and would otherwise win on specificity,
   which it did — the rule computed to `0px none` and nothing showed. */
fieldset.keeperfield {
  margin-top: 1.5rem;
  padding-left: 1rem;
  border-left: 3px solid var(--line);
}
/* THE PHONE KEEPS THE INDENT. THE BOARD DRAWS IT. 6 Oct 2026.

   This block used to REMOVE the indent and the rule below 30rem,
   with the comment "at phone width the indent costs more than it
   explains, so the separation is vertical space alone". That was a
   judgement made in CSS, and VehiclePhone makes the opposite one
   explicitly:

     margin-left: 6px; padding-left: 12px;
     border-left: 3px solid var(--statement-bg)

   Guillaume, 6 Oct: "board on design is always source of truth". So
   the indent stays at every width and the phone uses the board's
   tighter numbers. The rule colour is --statement-bg on phone, which
   is what the board names; desktop keeps --line, which it already
   had and which is signed off.

   3px STAYS PX while everything around it is rem. A hairline rule is
   a hairline: scaling it with the type would thicken it to 6px at
   200% text, where its whole job is to be a quiet edge.

   It also used a FIFTH BREAKPOINT, 29.99rem, where the four allowed
   are 30/48/56/64 — so it is folded into the 30rem block that the
   rest of the phone work uses. */
@media (max-width: 30rem) {
  fieldset.keeperfield {
    margin-top: var(--field-gap);
    margin-left: 0.375rem;   /*  6px, the board */
    padding-left: 0.75rem;   /* 12px, the board */
    border-left: 3px solid var(--statement-bg);
  }
}

/* ---- Companies Act registered-details disclosure ----------------
   ADDED 7 OCT 2026. Reg 25(2) of the Companies (Names and Trading
   Disclosures) Regulations 2015: the registered name, place of
   registration, company number and registered office must appear on
   the company's websites. CoRoute Ltd was incorporated that day.

   SMALLER THAN .footer-legal AND DELIBERATELY SO. The not-authorised
   sentence is the FCA-relevant one and keeps the full --footer-text at
   0.875rem; this is a statutory address disclosure, which has to be
   present and legible but is not the sentence a visitor needs to read.
   It still clears the contrast floor by a wide margin. COMPUTED, not
   estimated: --footer-text (#FFF9F1) at an 80% mix over --footer-bg
   (#291B56) composites to #D4CDD2, which is 9.72:1 against that
   ground — against a 4.5:1 requirement. Full --footer-text on the
   same ground is 14.49:1, which is what .footer-legal above uses. And
   0.8125rem is above the 0.75rem where small text starts failing
   people.

   The registered office is a home address that is already public on
   the Companies House register. Unobtrusive is the right register for
   it: present because the law requires it, not framed as an
   invitation. */
.footer-company {
  margin: 0.625rem 0 0;
  font-size: 0.8125rem;
  line-height: 1.5;
  color: color-mix(in srgb, var(--footer-text) 80%, transparent);
}
/* privacy.html, quote_under17.html and quote_unavailable.html carry
   footer.site rather than the shared .sitefooter, and footer.site p
   sets its own colour and size. Re-state both so the disclosure reads
   the same on every page that carries it. */
footer.site .footer-company {
  margin-top: 0.625rem;
  font-size: 0.8125rem;
  color: var(--ink-soft);
}

/* ===================================================================
   THE HERO WAITLIST BAR — OPTION B, canvas v153, 10 Oct 2026
   ===================================================================

   Board: Home.dc.html / HomePhone.dc.html, both read live from the
   canvas this session at version 1791664321-484f.

   THE HERO COLUMN IS, IN ORDER: headline, paragraph, this bar, the
   consent line, the orange quote button 18px below. Guillaume's words
   for Option B: "sign-up first, decluttered".

   OPTION A WAS BUILT AND IS REVERSED HERE. For part of today the
   canvas carried Option A (v152), which moved this bar out of the
   hero into a separate .signup-band below it, with a heading, a 460px
   white bar and an id="updates" anchor. Every one of those rules is
   DELETED rather than left in place — orphaned CSS for markup that no
   longer exists is how a stylesheet stops describing the page, and
   this file has been bitten by exactly that twice today (see the note
   where `.signup-band { padding-bottom: 2rem }` used to be, further
   up: a dead rule from a removed section came back to life when new
   markup reused the class name).

   EVERY NUMBER BELOW IS THE BOARD'S, not a judgement: cream #FFF9F1,
   radius 14px, padding 6px, shadow 0 10px 26px rgba(20,8,48,.3); the
   dark button #291449 at radius 10px with min-height 48px; the
   consent line 13px at 80% white; 400px column on desktop; 18px to
   the quote button.

   WHY A FIXED 400px AND NOT A PERCENTAGE. The board specifies that
   the bar and the quote button are both exactly 400px wide
   (box-sizing: border-box) so their edges line up. A percentage would
   track the column and the two would agree at exactly one viewport.
   The width is capped at 100% so it cannot overflow a narrow column,
   and below the two-column breakpoint it is full width. */

.hero-signup {
  margin: 0;
  width: min(400px, 100%);
  box-sizing: border-box;
}

/* position: relative is NOT decoration. The honeypot inside the bar
   is `position: absolute; left: -9999px` (.hp), which resolves
   against the nearest POSITIONED ancestor. Without this the hero bar
   is not one, so the honeypot anchored further out and the document
   grew 6px wider than the viewport on every phone size — measured by
   measure-layout.js, which reported "home @360: overflows by 6px"
   where it had been 0. Containing it here keeps the off-screen field
   off-screen without changing the page's width. */
.hero-bar {
  position: relative;
  display: flex;
  align-items: center;
  /* WRAPPING IS WHAT HAPPENS BELOW THE BOARD'S OWN WIDTH, and it is
     deliberately not a new breakpoint. The boards are fixed canvases:
     Home is 1280, HomePhone is 390, and NEITHER REFLOWS, so neither
     says what this bar should do at 360 or 320. Measured, a single
     row cannot hold both: at 320 the bar is 280 wide, which leaves
     99px of field content against a placeholder needing 152.22px —
     the field would still clip with a zero-width button.

     So the row wraps instead, driven by the field's own flex-basis
     below. Above that basis nothing wraps and the bar is exactly the
     row the board draws; below it the button drops to a second line
     inside the same cream bar. No media query, no fifth breakpoint
     (quote-layout R1), and the signed-off 390 view is untouched —
     measured at 393 and 390, the bar is still one row.

     THIS IS A TREATMENT THE BOARD DOES NOT DRAW. Reported rather than
     left silent: if the bar should do something else below 390 — a
     shorter placeholder, or a full-width button by design — that is a
     board decision, and the board should be changed first. */
  flex-wrap: wrap;
  gap: 8px;
  box-sizing: border-box;
  width: 100%;
  /* --field-bg, NOT A LITERAL, SINCE V2 -- and the comment below
     records exactly why it was a literal and why that stopped being
     true. #FFFFFF light, #101D30 dark, which is what the v2 boards
     declare as --field-bg. */
  background: var(--field-bg);
  border-radius: 14px;
  padding: 6px;
  box-shadow: 0 10px 26px rgba(11, 22, 38, 0.3);
}

/* The field carries no border of its own: the BAR is the input
   affordance, which is what the board draws. The focus ring therefore
   has to be visible against cream rather than against the purple
   hero, so it is the ink colour at 2px with an offset, not a browser
   default that would sit on the bar's own edge and read as a seam.

   ⚠️ THE PARAGRAPH THAT STOOD HERE WAS RIGHT AND ITS PREMISE IS GONE.
   It read: "THE CREAM BAR IS WHY THIS ONE KEEPS LITERAL HEXES rather
   than tokens. It is an opaque cream panel laid on the hero in both
   light and dark mode — THE HERO DOES NOT INVERT, so neither does
   this... Checked on the dark shot, not assumed."

   Every word of that was true of the v1 purple hero, which WAS the
   same dark panel in both themes. The v2 hero inverts: pale sky in
   light, navy in dark (see the token block). A cream bar on a pale
   sky hero is a cream rectangle on a pale field -- and the board
   does not draw it. HomeV2PhoneDark.dc.html draws a DARK field with
   a light-blue button, which is --field-bg #101D30; the light boards
   draw white.

   FOUND BY LOOKING, in the phone board comparison: the board's field
   was dark and the page's was white, side by side. No contrast check
   could have flagged it -- the bar passes in both themes either way,
   because the ink is pinned with the fill. It is a FIDELITY fault,
   not an accessibility one, which is the category only the image
   catches.

   --field-bg exists in app.css's palette for this, and the ink
   follows: 13.64:1 light, 14.21:1 dark. */
/* width:100% is INHERITED from the global input rule and is KEPT. It
   is the flex-basis clamp that stops the field overflowing the bar at
   narrow widths; removing it put a 2px overflow on the page at 360px,
   measured. min-width:0 is what lets it shrink below its content. */
.hero-bar input[type="email"] {
  /* 11rem basis = 176px, which is the measured requirement: the
     placeholder "Your email address" renders 152.22px in this field's
     own font (17px system-ui, read off getComputedStyle and measured
     with canvas measureText, not estimated), plus 12px padding each
     side = 176.22px. Rounded to the nearest rem up, so it holds at
     200% text as well as at 100%.

     ⚠️ THAT MEASUREMENT WAS TAKEN IN system-ui AND THE FIELD IS NOW
     'DM Sans' (7 Nov 2026, see the note on `body`). The number is
     left as recorded rather than quietly adjusted, because it is
     evidence of a measurement someone took and re-measuring is the
     only honest way to change it. Two reasons it is not urgent: the
     basis was ROUNDED UP from 176.22 to 176, so there is headroom
     only if the new face is narrower, and DM Sans at the same px size
     sets slightly narrower than most system-ui faces — so the risk is
     unused space, not a clipped placeholder. The failure mode if it
     IS wider is also benign: the flex-wrap below drops the button to
     its own line a few pixels earlier than before.

     RE-MEASURE IT before trusting the 152.22, and update both numbers
     together if it moved.

     THE BASIS IS WHAT DECIDES THE WRAP. While the bar can give the
     field 176px it stays beside the button, which is every width the
     boards draw — measured one row at 1280, 1104, 393 and 390. Below
     that the field's basis no longer fits next to the button and the
     flex-wrap above drops the button to its own line, rather than
     shrinking the field until the placeholder clips.

     min-width: 0 still lets it shrink below its CONTENT once wrapped,
     which is what keeps the page from scrolling sideways at 320. */
  flex: 1 1 11rem;
  min-width: 0;
  border: 0;
  background: transparent;
  /* THE BOARD'S 12px. It had been shaved to 10px to buy a clipped
     placeholder four pixels; that was treating a symptom. The
     placeholder was clipped because the BUTTON was 92px wider than
     the board's, inheriting the global `button { min-width: 14rem }`
     — see .hero-submit below. With the button at its own width the
     field has room to spare, so the board's padding is what is drawn.

     MEASURED over CDP at 1280, page against the board rendered in the
     same browser:
       board   input 247.53 wide, 224.00 content, placeholder 144.90 -> fits, 79.10 spare
       before  input 156.00 wide, 136.00 content, placeholder 152.22 -> CLIPPED by 16.22
     The placeholder's width differs between the two because the board
     renders in Arial and the page in system-ui; both are measured in
     their own font with canvas measureText, not assumed. */
  padding: 12px;
  font: inherit;
  font-size: 1.0625rem;   /* 17px */
  /* --ink follows --field-bg. 13.64:1 light, 14.21:1 dark. */
  color: var(--ink);
  border-radius: 8px;
}
/* --ink-soft on --field-bg is 5.67:1 light and 9.65:1 dark. The
   placeholder is the field's only VISIBLE label (the real <label> is
   visually-hidden), so it is held to body-text contrast rather than
   to the 3:1 a decorative grey would get away with. opacity:1
   defeats the UA's own dimming, which would undo that. */
.hero-bar input[type="email"]::placeholder { color: var(--ink-soft); opacity: 1; }
.hero-bar input[type="email"]:focus-visible {
  outline: 2px solid var(--ink);
  outline-offset: 2px;
}

.hero-submit {
  flex: none;
  /* THIS PAIR IS THE FIX FOR THE CLIPPED PLACEHOLDER, and it is the
     whole of it. 10 Oct 2026, measured, not reasoned about.
     ------------------------------------------------------------------
     app.css:382 sets `button { width: 100% }` for every button on the
     site, and app.css:405 replaces it with `min-width: 14rem` (224px)
     above 34rem. .hero-submit set NEITHER, so both applied to it:

       at 1280   the button took min-width 224px - 92px more than the
                 board's own 132.47px - and `flex: none` stopped it
                 giving any of it back, so the field was squeezed to
                 156px (136px of content) against a placeholder needing
                 152.22px. That is the clip, and it is 16.22px short.
       at 393    `width: 100%` resolved against the bar, so the button
                 asked for 341px INSIDE a 353px bar. The field collapsed
                 to 20px - no placeholder visible at all - and the
                 button's right edge landed at 395 against a bar ending
                 at 373: 22px outside its own bar and 2px outside the
                 viewport. Visible in board-vs-page.sh HomePhone.

     Three earlier attempts moved the input's and the button's padding
     instead, which is why the arithmetic kept saying the text should
     fit: the padding was never what was taking the space.

     `width: auto` undoes the phone rule and `min-width: 0` undoes the
     desktop one, so the button is sized by its own label and padding,
     which is what the board draws. The board's own button computes
     min-width: auto and 132.47px wide. DO NOT REMOVE EITHER. */
  width: auto;
  min-width: 0;
  /* The same global rule also carries `margin-top: 0.5rem`, meant for
     a button that follows a field down a page form. In a flex bar it
     pushed the button 4px below the bar's centre line: measured at
     1280, bar centre 463.56 against button centre 467.56, where the
     board has both on 485.63. */
  margin: 0;
  /* --brand/--on-brand, not literals, since v2: the light theme's
     brand is navy with white ink and the dark theme's is a light
     blue with navy ink, so a hardcoded pair would be a navy button
     on a near-navy field in dark mode. 14.27:1 light, 9.72:1 dark. */
  background: var(--brand);
  color: var(--on-brand);
  border: 0;
  border-radius: 10px;
  /* The board's 20px, restored with the input's 12px above. */
  padding: 13px 20px;
  font: inherit;
  font-size: 1rem;
  font-weight: 700;
  line-height: 1.2;
  min-height: 48px;      /* the board's number, and a 48px touch target */
  cursor: pointer;
}
/* The hover is a filter rather than a second pair of literals: one
   declaration that lightens a dark brand and deepens a light one is
   correct in both themes, where two hardcoded hexes would need the
   same per-theme duplication as everything else. */
.hero-submit:hover { filter: brightness(1.18); }
:root[data-theme="dark"] .hero-submit:hover { filter: brightness(0.92); }
@media (prefers-color-scheme: dark) {
  :root:not([data-theme="light"]) .hero-submit:hover { filter: brightness(0.92); }
}
/* The ring is drawn on the HERO GROUND, not on the button, so it
   takes the hero's own text colour: navy on pale sky, cream on navy.
   12.08:1 and 13.64:1 respectively. */
.hero-submit:focus-visible { outline: 3px solid var(--hero-text); outline-offset: 2px; }

/* THE CONSENT LINE. 13px in --hero-soft, which is the hero's own
   secondary ink and follows the theme.

   IT WAS rgba(255,255,255,0.8) UNTIL V2 AND THAT HAD TO GO. 80%
   white was right while the hero was a dark purple panel in both
   themes — it measured 9.5:1 there. On the v2 LIGHT hero the ground
   is pale sky, and white at 80% over it is about 1.2:1: the sentence
   a person is consenting to would have been very nearly invisible,
   which is the worst place on this site for that to happen. An
   opacity is a claim about the ground beneath it, and the ground
   changed.

   --hero-soft is #2F4660 light and #C9DAEA dark. Measured against
   BOTH ends of the hero's gradient, because the ground is a ramp:

     light  on --hero-bg #DCEFFA   8.20:1
            on --sky     #84CBFC   5.51:1   <- worst case
     dark   on --hero-bg #142B49   9.99:1
            on --sky     #021E4A  11.42:1

   The link is NOT dimmed further — it takes --hero-text at full
   strength and is underlined, because a dimmed qualifier is how
   qualifiers stop being read.

   THIS PARAGRAPH IS THE STORED CONSENT_TEXT, rendered from
   {{ .ConsentText }}. It is not copy. Its SHA-256 is written to
   signup.consent_events.text_hash against every signup, so changing
   the wording is a CONSENT_FORM_VERSION bump, never a style edit.
   Nothing here may change what it SAYS; this rule only sets how it is
   set. See the long note in index.html.

   THE max-width MATCHES THE BAR, so the line breaks inside the same
   400px column the bar and the button occupy rather than running out
   under the hero art. That matters more now than it did when this
   line was four words: the stored sentence is longer than the board's
   display copy, so it wraps to several lines and the cap is what
   keeps it a block under the bar instead of a paragraph across the
   hero. If the final wording lands short, the cap costs nothing.

   CHECKED AGAINST THE ART, 12 Oct 2026, AND 400px SURVIVES.
   ------------------------------------------------------------------
   Same report as .hero-lede. 400px was chosen against the BAR and
   had never been checked against the ART, so it was re-measured.
   Before the art was resized, this line's right edge landed 80px
   past the point where the picture becomes visible, and in
   Guillaume's 1913 screenshot the word "your" ending line 1 sat on
   the blossom tree with line 2 running along the wall.

   THE FIX WAS THE ART, NOT THIS NUMBER. Guillaume: "you could leave
   the text as is but make the graphics smaller". .hero-art-layer
   went from 980px to 880px, which moved its left edge 100px right
   and off this paragraph. 400px then needs no change — it already
   aligns with the email bar and the CTA, which is the alignment
   this rule was written for in the first place.

   SO THE CAP STAYS 400px AND STAYS A LITERAL. It is the bars' width
   and shares their right edge; it is not a reading measure and
   should not be converted into one. If the bars change width, this
   changes with them.

   RE-MEASURE IT IF THE ART CHANGES AGAIN. This paragraph renders
   CONSENT_TEXT, whose wording is an audit record and WILL change
   when [COMPANY] is replaced (O4). A longer sentence wraps to more
   lines at the same width rather than growing wider, so the art
   clearance holds — but the line COUNT grows, and that is the thing
   to look at then, not this cap.

   THE BOARD DOES NOT SETTLE THIS ONE, and that is worth saying
   plainly. Home.dc.html gives this paragraph no max-width at all: on
   the board it is the short display string "Launch news only.
   Unsubscribe any time. Privacy notice", one line, 709px of column
   available, and its right edge runs 342px into the art box — 205px
   past the opaque point. The board is therefore NOT a picture of the
   stored sentence, and following it literally here would put more
   text on the brickwork, not less. The divergence is deliberate and
   is recorded as a note on the canvas rather than left as a silent
   difference between board and page. */
.hero-consent {
  margin: 10px 0 0;
  max-width: 400px;
  font-size: 0.8125rem;
  line-height: 1.5;
  color: var(--hero-soft);
}
.hero-consent a {
  color: var(--hero-text);
  text-decoration: underline;
  white-space: nowrap;   /* the board: "Privacy notice" never wraps */
}

/* THE 976-1103 BAND NEEDS 300px, BECAUSE THE ART IS PINNED CLOSER.
   ===================================================================
   Added 12 Oct 2026, after the art was resized to 880px for the wide
   widths. Shrinking the art fixed 1280 and up — it moved the picture
   100px right and off the copy — and it did NOT fix this band,
   because here the art's left edge does not depend on its width at
   all. It is pinned at a fixed offset by the rule beside
   .hero-art-layer ("THE ART'S LEFT EDGE IS PINNED THROUGH THE
   976-1103 BAND"), so the picture stayed where it was while every
   wide width improved.

   Measured on the built page after the resize:

     vw            bars end   art box left   first VISIBLE art
     976..1103        440          319             ~356
     1280             473          455              492
     1913             789          772              809

   So at 1024 the 400px consent line ends at 440, which is 84px into
   a picture that starts reading at 356. Sampled from the 1024
   screenshot along the consent line's own rows, against the hero
   token rgb(75,31,174):

     x=348   delta  34   purple, the shadow under the bar
     x=356   delta  42   the tree starts to read
     x=372   delta  81   clearly picture
     x=404   delta 205   brickwork

   LOOKED AT, NOT ONLY MEASURED: in the 1024 render "We will not pass
   your" ends on the blossom tree and "Privacy notice" — an
   underlined link — runs along the roofline. That is the same fault
   Guillaume reported at 1913, surviving at laptop width for a
   different cause.

   300px, from 316px of measured clearance less a 16px margin so the
   text is not flush against the first visible pixel.

   ONLY .hero-consent, NOT .hero-lede. After the resize the lede
   already clears the art in this band: it wraps to three lines whose
   longest ends at ~409, inside the 440 the bars occupy and short of
   356 only on its final short line — checked in the render, where it
   sits on flat hero ground at both 1024 and 1100. The consent line is
   longer, at 13px, and reaches further. Narrowing both would cost
   the lede a line for nothing.

   THIS IS BELOW THE BASE RULES ON PURPOSE. The first version of this
   override was written beside the .hero-art-layer band rule at
   ~line 3030 and DID NOTHING: the base .hero-lede and .hero-consent
   rules are at ~4115 and ~9540, same specificity but later in the
   file, so they won. The 1024 screenshot was unchanged and that is
   how it was caught. A media query carries no extra weight; only its
   position does. The design-canvas skill records the same bug from
   3 Oct 2026, when a base .explainer-figure rule after its media
   query cancelled the margin-inline that centred it.

   IT DOES NOT REOPEN THE BAND'S SIGNED-OFF CONTRAST FIGURE. The note
   beside .hero-art-layer accepts 13.6% of background pixels below
   4.5:1 inside the clarify text's box at 1104, and pins the art so
   the band holds that rather than degrading to 23.5%. That figure
   was for the clarify paragraph, which moved to quote step 1 on
   10 Oct and no longer exists, and it was accepted for a 418px box
   with no choice but to sit under the art. This paragraph has a
   choice. Taking it is strictly better, and the pin is untouched —
   the art does not move. */
@media (min-width: 61rem) and (max-width: 68.9375rem) {
  .hero-consent { max-width: 300px; }
}

/* FORCED COLORS, the same pattern as .portal-saving-pill and the
   chart keys above. Found by LOOKING at the forced-colors shot that
   measure-layout.js --shots already takes, rather than by reasoning:
   the whole bar is built out of `background` and `border: 0`, and
   forced-colors drops every background. The result was the cream bar
   gone and "Get updates" rendered as bare text with no boundary at
   all, beside a quote button that kept its border and still read as a
   button. A submit control that looks like a word is not operable by
   anyone relying on this mode.

   Borders rather than colours: the system repaints the text and the
   ground, and all these rules add back is the EDGE that the dropped
   backgrounds were carrying. Nothing here changes normal rendering. */
@media (forced-colors: active) {
  .hero-bar { border: 1px solid CanvasText; }
  .hero-submit { border: 2px solid CanvasText; }
  /* The field's own edge, which the bar was standing in for. */
  .hero-bar input[type="email"] { border: 1px solid CanvasText; }
}

/* THE CLOSED STATE. SignupEnabled is false in production and locally
   by default, so this is the state the deployed page is actually in.
   It keeps the bar's box exactly — same cream, radius, padding and
   shadow — so the hero does not change shape when the flag flips. */
.hero-bar-closed { display: block; padding: 14px 16px; }
.hero-closed {
  margin: 0;
  font-size: 0.9375rem;
  line-height: 1.5;
  color: #142B49;
}

/* THE SUCCESS STATE, swapped in place by htmx.
   The board asks for the bar to be "replaced IN PLACE with a
   same-size bar showing an envelope-tick icon, 'You're on the list!'
   on the left and the entered email on the right. No pop-up, no page
   jump." .hero-slot is empty until the swap, so it must not take
   space or affect the bar's layout while it is. */
.hero-slot:empty { display: none; }
.hero-slot {
  flex: 1 1 100%;
  min-width: 0;
}

/* "REPLACED IN PLACE" HAS TO MEAN REPLACED, AND IT DID NOT.
   ------------------------------------------------------------------
   FOUND BY DRIVING THE REAL SUBMIT AND PHOTOGRAPHING THE RESULT,
   10 Oct 2026, not by reading the markup. Both boards say the bar is
   "replaced IN PLACE by 'You're on the list!' + the email entered".
   It was not being replaced — it was being PREPENDED:

     before submit   bar 400 x 62
     after  submit   bar 400 x 154, and `#hero-email` still in the DOM,
                     still holding the address the person typed, with
                     a live "Get updates" button under the success
                     message

   So a confirmed signup showed a success message sitting on top of a
   form still inviting the same submission, in a box two and a half
   times its own height. The hero visibly changed shape. Measured
   identical at clean HEAD, so this shipped in 56a36cd and is fixed
   here rather than reported, because the success state is item 5 of
   the spec this commit is building.

   WHY IT HAPPENED. hx-target is #hero-form-slot, a SIBLING of the
   field and the button inside .hero-bar, so the swap fills the slot
   and leaves its siblings exactly where they were. Targeting the bar
   itself instead would have worked visually and been wrong: the live
   region has to be PERMANENT and pre-existing or it announces nothing
   (code review A2), and swapping the bar would replace the region
   with an injected copy of itself.

   THE FIX IS CSS, AND IT IS WHY THE SLOT IS A SEPARATE ELEMENT AT
   ALL. Once the slot has content, the bar's other children are taken
   out of the layout. :has() is supported everywhere this site
   targets; where it is not, the page degrades to exactly the
   behaviour above — a visible success message with a stale form under
   it, which is ugly but not broken, and no worse than what shipped.

   display: none RATHER THAN visibility OR opacity: the field and the
   button must leave the ACCESSIBILITY TREE as well as the picture. A
   hidden-but-focusable submit after a success message is a tab stop
   that resubmits an address already on the list, and a screen-reader
   user would hear "Email address, edit text" after being told they
   are on the list. */
.hero-bar:has(.hero-slot:not(:empty)) > :not(.hero-slot) { display: none; }

/* 18px from the consent line to the quote button, per the board.
   .cta-row is the existing wrapper; only its top margin and the
   button's width change in the hero.

   THE BUTTON IS 400px AND BLOCK, matching the bar above it, which is
   what makes their edges line up — the board's stated reason for both
   being "exactly 400px wide (box-sizing: border-box)". This overrides
   the site-wide `.cta-row .action { flex: 0 1 auto; display:
   inline-flex }`, which sizes a CTA to its text; a 267px button under
   a 400px bar reads as misaligned rather than as deliberate. Below
   61rem the phone rule on .cta-row .action already makes it full
   width and min(400px,100%) lets it, so the two agree without a
   second media query. */
.homehero .cta-row { margin: 18px 0 0; }
.homehero .cta-hero {
  display: block;
  width: min(400px, 100%);
  box-sizing: border-box;
  text-align: center;
}

/* THE TWO FACES OF form_success.
   One fragment serves the hero bar and the page form (see the note in
   form_success.html). Each context shows one and hides the other, so
   there is no second template for the same success path.
   DISPLAY:NONE RATHER THAN visually-hidden on the hidden one: both are
   inside a live region, and a visually-hidden copy would be announced
   as well, reading the success message twice. */
.hero-ok { display: none; }
.hero-slot .result { display: none; }
.hero-slot .hero-ok {
  display: flex;
  align-items: center;
  gap: 10px;
  width: 100%;
  min-width: 0;
  padding: 8px 10px;
  color: #142B49;
}
.hero-ok-icon { flex: none; color: #0F6B3F; }
/* flex: none IS THE FIX FOR A WRAPPED MESSAGE, 10 Oct 2026, and the
   comment under it used to assert the opposite of what was happening.
   ------------------------------------------------------------------
   It said the address "shrinks and truncates rather than pushing the
   message out". MEASURED with a 38-character address — the real
   submit, photographed — the message was the thing that gave way:
   "You're on the list!" wrapped onto two lines, "Check your inbox to
   confirm." onto two more, and the bar went from 62px to 96px. The
   hero changed shape on success, which is what "replaced in place"
   exists to prevent.

   WHY. .hero-ok-text had no `flex` of its own, so it took the
   default `0 1 auto` — shrinkable — exactly like the address beside
   it. Two shrinkable siblings both shrink; the browser was not
   choosing wrongly, nothing had told it which one matters.

   `flex: none` says it. The message is fixed at its content width and
   the address is now the only thing that can give, which is what the
   comment below always claimed and now describes. Re-measured after
   the change: bar 62px before and after submit, at 1280 and at 390,
   with the same 38-character address. */
.hero-ok-text {
  display: flex;
  flex: none;
  flex-direction: column;
  line-height: 1.25;
  min-width: 0;
}
.hero-ok-text strong { font-size: 0.9375rem; }
.hero-ok-sub { font-size: 0.75rem; color: #40566B; }
/* The address is the least important thing in the bar and the most
   likely to be long, so it is the part that shrinks and truncates
   rather than pushing the message out. That is true because of the
   `flex: none` above, not on its own. title= is not set: it would be
   the person's own address repeated on hover for no benefit, and an
   address that has been ellipsised is still fully visible in the
   field they typed it into a moment earlier. */
.hero-ok-email {
  flex: 0 1 auto;
  min-width: 0;
  margin-left: auto;
  padding-left: 0.75rem;
  font-size: 0.8125rem;
  color: #40566B;
  overflow: hidden;
  text-overflow: ellipsis;
  white-space: nowrap;
}

/* ---- The no-htmx form response page -----------------------------

   form_page.html, added 14 Oct 2026. It renders a form-result
   FRAGMENT as the body of a real page, for the browser that posted
   the form without htmx — the bug that put unstyled Times New Roman
   at /quote/email/<uuid> on production.

   ONE RULE, AND IT IS A READING MEASURE ON PROSE. .result has no
   max-width of its own, because everywhere else it is swapped into a
   form column that already has one. On a 75rem page it would be
   1128px wide, which is an unreadable line length for the two
   paragraphs inside it. 34rem is .flow-lede's measure, so the
   message box lines up with the lede on every other flow screen
   rather than introducing a width of its own.

   LAYOUT CONTRACT R2 PERMITS THIS. The rule it forbids is a
   max-width on a column HOLDING CONTROLS, which is the bug that
   contract was written for. This page has no controls at all: two
   links and some text. */
.formpage-msg { max-width: 34rem; }

/* THE WRAPPER IS .formpage-copy, NOT .hero-copy, AND THAT IS NOT
   TIDINESS. The homepage block at max-width 60.9375rem styles a BARE
   `.hero-copy` — purple ground, cream text, its own padding —
   restoring the hero background that display:contents takes away. It
   is not scoped to .homehero, so this page picked it up and rendered
   its message in a purple card at 393px while looking perfectly
   correct at 1280. Found by screenshotting the page at phone width,
   which is the only thing that could have found it.

   This class is deliberately styled by NOTHING. It exists so the
   wrapper cannot collide with a hero rule, and `min-width: 0` is the
   one thing .hero-copy gave it that is worth keeping: it stops a long
   unbroken string forcing the column wider than its container. */
.formpage-copy { min-width: 0; }

/* THE PANEL ON THE SKY PAGES NARROWS TO ITS CONTENT. 29 Oct 2026.

   form_page.html took .confirmsky/.confirmpanel so the no-JS form
   responses get the confirmation pack's sky and clouds. The panel
   is inside .wrap, which is 75rem, and this page has NO artwork
   column -- so at 1440 the card spanned 1200px with its copy in the
   left 540 and half the card empty. Measured, not guessed: the
   message box is capped at 34rem by .formpage-msg above, which is
   the prose measure the layout contract wants and is not the thing
   to widen.

   44rem is that 34rem measure plus the panel's own padding at the
   desktop end of its clamp, so the card hugs the column instead of
   floating around it. Centred, because a left-aligned narrow card
   under a centred nav reads as a mistake.

   SCOPED TO THE PANEL THAT HAS NO ART. confirmed.html's panel is a
   two-column .flowsplit with artwork in the second column and must
   stay full width; it never matches this selector because it does
   not carry .formpage-copy. */
.confirmsky .confirmpanel:has(> .formpage-copy) {
  max-width: 44rem;
  /* LEFT, NOT CENTRED. Guillaume, 29 Oct 2026: "we want messge +
     car to the left". Centred was the first pass and it read as a
     floating card; left-aligned it sits under the wordmark and the
     sky opens out to its right, which is the pack's composition. */
  margin-inline: 0 auto;
}
/* THE ARTWORK UNDER THE COPY, which is the pack's PHONE composition
   used at every width -- "not two column", "that is the mobile
   view". confirmed.html's .flowsplit-art is the desktop half and is
   deliberately not reused here.

   Full-bleed to the panel's edges via negative margins against its
   own --confirm-pad, so the picture meets the card's rounded
   corners the way the mockup draws it; the panel's overflow:hidden
   does the clipping. 16/10 is the pack's phone crop. */
.formpage-art {
  margin: var(--confirm-pad) calc(var(--confirm-pad) * -1)
          calc(var(--confirm-pad) * -1);
}
.formpage-art img {
  display: block;
  width: 100%;
  height: auto;
  aspect-ratio: 16 / 10;
  object-fit: cover;
  object-position: 50% 62%;
}

/* ===================================================================
   THE CONTEXTUAL FILM AND ITS END PANEL — 15 Oct 2026, canvas v239/v242

   Replaces the two .ctxai-scenario cards and the .ctxai-table that sat
   under them with one film plus an HTML end panel. The old rules are
   left above deliberately: .ctxai, .ctxai-h and .ctxai-lede still
   dress this section's heading and prose, which did not change.

   The film sits in the same white frame as "How it works" — 12px white
   border, 16px radius, max 760px — because the board draws them the
   same and they are two films on one page.
   =================================================================== */
.ctxfilm {
  max-width: 47.5rem;        /* 760px, the board's frame */
  margin-inline: auto;
  background: var(--paper);
  border: 12px solid var(--paper);
  border-radius: 16px;
  box-shadow: 0 8px 30px rgba(20, 43, 73, 0.16);
  overflow: hidden;
}
.ctxfilm-stage {
  position: relative;
  aspect-ratio: 16 / 9;
  background: #000;
  border-radius: 6px;
  overflow: hidden;
}
.ctxfilm-video { width: 100%; height: 100%; object-fit: cover; display: block; }

/* THE CHIPS STAY. "Planned service / Illustrative preview" is the
   load-bearing one — the boxes and labels on these frames were drawn
   by hand — and "Real footage" earns its place here in a way it did
   not on the old still cards, because this is a 37-second film and a
   viewer may reasonably wonder whether it is rendered. */

.ctxfilm-chip {
  background: rgba(11, 22, 38, 0.82);
  color: #fff;
  font-size: 0.72rem;
  font-weight: 600;
  padding: 5px 9px;
  border-radius: 999px;
}

/* TOP-LEFT, NOT TOP-RIGHT. The footage carries its own "Oncoming
   vehicle" / "Parked car blocking your lane" card burned into the
   top-right corner, and a button there sat on top of it. */


.ctxfilm-play {
  position: absolute; inset: 0; margin: auto; z-index: 4;
  width: max-content; height: max-content;
  display: flex; align-items: center; gap: 9px;
  background: rgba(11, 22, 38, 0.92);
  color: #fff;
  border: 1px solid rgba(255, 255, 255, 0.3);
  font: 700 1rem/1 inherit;
  padding: 14px 22px;
  border-radius: 999px;
  cursor: pointer;
}
.ctxfilm-play[hidden] { display: none; }
.ctxfilm-play:focus-visible { outline: 3px solid var(--cta); outline-offset: 3px; }

/* NO CUSTOM CAPTION PILL, NO CUSTOM PROGRESS BAR, NO CC BUTTON.
   16 Oct 2026, Guillaume: "make the video play for the CoRoute AI the
   same as how it works".

   All three were built here while this film had no control bar of its
   own — it autoplayed muted, so it needed a pill to carry the words,
   a bar to show position, and later a toggle to get the pill out of
   the way of the road. The film now hands over to the NATIVE controls
   on the first click, exactly as the explainer does, and the browser
   draws the scrubber and offers its own CC control. Keeping ours
   underneath would mean two progress bars and two ways to toggle one
   track.

   The <track> in index.html is unchanged and still `default`. What
   went away is our copy of the furniture, not the captions. */


/* ---- the end panel ---------------------------------------------- */
.ctxend {
  position: absolute; inset: 0; z-index: 5;
  display: flex; flex-direction: column; justify-content: center;
  padding: 20px 22px;
  background: linear-gradient(160deg, #102341 0%, #0B1626 100%);
  color: #fff;
  opacity: 0;
  visibility: hidden;
  transition: opacity 0.4s ease, visibility 0.4s;
}
.ctxend.is-shown { opacity: 1; visibility: visible; }
@media (prefers-reduced-motion: reduce) {
  .ctxend { transition: none; }
}
.ctxend-top {
  display: flex; align-items: flex-start; justify-content: space-between; gap: 12px;
}
.ctxend-h { margin: 0; font-size: 1.5rem; font-weight: 900; line-height: 1.15; }
.ctxend-h em { font-style: normal; color: #FFAA66; }
.ctxend-chip { background: rgba(255, 255, 255, 0.14); font-size: 0.66rem; white-space: nowrap; }

.ctxend-grid {
  display: grid;
  grid-template-columns: 112px 1fr 1fr 1fr;
  gap: 0 14px;
  align-items: center;
  align-content: start;
  margin-top: 8px;
}
.ctxend-head {
  padding-bottom: 7px;
  font-size: 0.66rem;
  font-weight: 700;
  letter-spacing: 0.09em;
  text-transform: uppercase;
  color: #B7C5D6;
}
.ctxend-head--box { color: #FF8A8E; }
/* The wordmark is a logo, not a column label: the uppercase transform
   above would render it "COROUTE" and lose the ring-C entirely. */
.ctxend-head .ctxend-wm { text-transform: none; letter-spacing: normal; }

/* display:contents so each cell lands in the grid's own columns and
   the three rows share one set of tracks. */
.ctxend-row { display: contents; }
/* box-shadow, not border-top: the thumbnail is narrower than its
   track, so a border on the cell stopped short of the column edge. */
.ctxend-row > * { box-shadow: 0 -1px 0 rgba(255, 255, 255, 0.14); padding: 10px 0; }
.ctxend-grid img {
  width: 112px; height: 63px;
  object-fit: cover;
  border-radius: 8px;
  display: block;
}
.ctxend-what { font-size: 0.84rem; font-weight: 700; line-height: 1.3; }
.ctxend-verdict {
  display: flex; align-items: center; gap: 8px;
  margin: 0;
  font-size: 0.84rem;
  font-weight: 700;
  line-height: 1.3;
  color: #fff;
}
/* FULL OPACITY AND NO STRIKETHROUGH on the black-box verdict. The
   brief is explicit: it has to be readable. Dimming it would make the
   comparison look like a straw man. */
.ctxend-badge { flex: none; display: grid; place-items: center; }
.ctxend-badge--x { width: 22px; height: 22px; border-radius: 6px; background: #E5484D; }
/* LITERAL #FFB264, for the reason already recorded at .radiocard's
   focus mark: the boards draw this orange specifically, var(--cta) is
   #FFAA66, and both tokens have other consumers. The end-panel brief
   names #FFB264 too. */
.ctxend-badge--t { width: 22px; height: 22px; border-radius: 50%; background: #FFB264; }

.ctxend-wm {
  /* gap WAS 1px, to part a typed C from the rest of the name. The
     ring sets its own -0.03em overlap and a positive gap here fights
     it, so this is 0 now -- removed with the ring conversion,
     23 Oct 2026. */
  display: inline-flex; align-items: baseline; gap: 0;
  font-size: 0.875rem; font-weight: 800; color: #fff;
}
/* The end panel is dark in both themes, so it pins --logo too.
   font-weight stays on the span; only the colour moves to the token.
   The > span selector no longer reaches the C -- that is a .brand-c
   ring taking --logo from the same override. */
.ctxend { --logo: #FFB264; }
.ctxend-wm > span { color: var(--logo); font-weight: 900; }

.ctxend-score {
  /* ONE LINE AT DESKTOP, which took two attempts. flex-wrap:wrap
     dropped Replay onto its own row; nowrap alone then let the three
     labels wrap to two lines each, because they are allowed to shrink.
     white-space:nowrap on the labels plus a slightly smaller size is
     what actually fits the board's single row at 760px. */
  display: flex; align-items: center; gap: 9px; flex-wrap: nowrap;
  margin-top: 14px;
  padding: 9px 12px;
  background: rgba(255, 178, 100, 0.12);
  border: 1px solid rgba(255, 178, 100, 0.35);
  border-radius: 10px;
  font-size: 0.78rem;
  font-weight: 700;
}
.ctxend-score-l { color: #fff; white-space: nowrap; }
.ctxend-score-ctx { color: #FFAA66; white-space: nowrap; }
.ctxend-score .ctxend-wm { white-space: nowrap; }
.ctxend-replay {
  margin-left: auto;
  flex: none;
  /* min-width:0 — THE SAME UA DEFAULT THAT MOVED THE QUOTE STEPPER
     106px this morning. A flex <button> gets an intrinsic min-width
     (measured here: 224px, against ~100px of actual content) that a
     <span> does not, so the bar's children needed 707.6px in a 668px
     box and Replay hung over the rounded right edge. */
  min-width: 0;
  display: flex; align-items: center; gap: 7px;
  background: #fff;
  color: #0B1626;
  border: 0;
  font: 700 0.8125rem/1 inherit;
  padding: 9px 15px;
  border-radius: 999px;
  cursor: pointer;
}
.ctxend-replay:focus-visible { outline: 3px solid var(--cta); outline-offset: 2px; }

/* Desktop shows the two verdicts as their own grid columns; the phone
   stacks them under the "what happened" line inside .ctxend-cell. */
.ctxend-verdict--m { display: none; }

@media (max-width: 37.5rem) {
  /* THE PANEL FLOWS BELOW THE FILM ON A PHONE, at its own height. The
     board asks for auto height here: forcing this much text into a
     16:9 box at 390px would either clip it or shrink it to nothing. */
  .ctxend {
    position: static;
    display: none;
    opacity: 1;
    visibility: visible;
    padding: 16px;
    border-radius: 6px;
  }
  .ctxend.is-shown { display: flex; }
  /* Once the panel is up the film is finished, so the video and its
     furniture come out of the flow and the card is the panel's height
     rather than 16:9 of dead video plus the panel. */
  .ctxfilm-stage.is-ended { aspect-ratio: auto; }
  .ctxfilm-stage.is-ended .ctxfilm-video { display: none; }

  .ctxend-h { font-size: 1.1875rem; }
  /* Chip to the bottom on a phone, per the board — which also gives
     the title the full row so it sets on two lines, not three. */
  .ctxend-top { display: block; }
  .ctxend-top .ctxend-chip { display: none; }
  .ctxend-foot { display: flex; align-items: center; margin-top: 10px; }
  .ctxend-foot::before {
    content: "Planned service / Illustrative preview";
    background: rgba(255, 255, 255, 0.14);
    color: #fff;
    font-size: 0.66rem;
    font-weight: 600;
    padding: 5px 9px;
    border-radius: 999px;
  }

  /* The desktop grid relies on .ctxend-row{display:contents}. On a
     phone the ROW is the layout unit, so the container becomes a plain
     block and each row its own two-column grid. Leaving .ctxend-grid
     as a grid made every row a grid ITEM and they overlapped. */
  .ctxend-grid { display: block; margin-top: 6px; }
  .ctxend-head { display: none; }
  .ctxend-row {
    display: grid;
    grid-template-columns: 96px minmax(0, 1fr);
    gap: 10px;
    align-items: start;
    padding: 10px 0;
    border-top: 1px solid rgba(255, 255, 255, 0.14);
  }
  .ctxend-row:first-of-type { border-top: 0; }
  .ctxend-row > * { box-shadow: none; padding: 0; }
  .ctxend-grid img { width: 96px; height: 54px; }
  .ctxend-cell { display: flex; flex-direction: column; gap: 5px; }
  .ctxend-what { font-size: 0.8125rem; }
  .ctxend-verdict { font-size: 0.78rem; }
  .ctxend-verdict--m { display: flex; }
  .ctxend-verdict--d { display: none; }
  .ctxend-badge--x, .ctxend-badge--t { width: 18px; height: 18px; }
  /* The phone DOES wrap: the board puts Replay under the bar. */
  /* The phone DOES wrap and its labels may break: the board puts
     Replay under the bar there. */
  .ctxend-score { margin-top: 12px; flex-wrap: wrap; font-size: 0.8125rem; }
  .ctxend-score-l, .ctxend-score-ctx { white-space: normal; }
  /* Bottom-RIGHT and sized to its text, as the board draws it.
     margin-left:auto pushes it right on its own wrapped line;
     without align-self it stretched the full column width. */
  .ctxend-replay { margin-left: auto; align-self: flex-end; }
  }

/* TESTING THE FILM BEHIND THE PREVIEW WALL: do not set the preview
   Authorization header globally.

   16 Oct 2026, cost about an hour, twice. A CDP probe that calls
   Network.setExtraHTTPHeaders with the preview Basic credential
   attaches it to EVERY request, including the cross-origin fetch of
   the video from storage.googleapis.com. GCS tries to parse that
   header as a Google credential and answers 401 with
   `www-authenticate: Bearer realm="https://accounts.google.com/"`.

   The <video> then reports networkState 3 (NO_SOURCE), readyState 0,
   duration null AND error null — which reads exactly like a broken
   deployment, a CSP block or a missing file, and is none of those.
   A real browser never sends that header to the bucket.

   Put the credential in the URL instead
   (https://preview:pass@host/), which scopes it to the app's own
   origin. Verified: with the header the bucket returns 401, without
   it 200, same URL and same Origin. */

/* THE FILM'S LAST 32 SECONDS ARE THE END PANEL, BURNED INTO THE VIDEO.
   16 Oct 2026: the narrated panel was joined onto the dashcam footage
   (contextual-v1 -> v2, 37.5s -> 69.7s), so from 37.3s the <video> is
   showing a title, a chip and three rows of its own. The furniture
   that belongs over FOOTAGE then collides with it.

   IT ONCE DID THREE JOBS AND NOW DOES ONE. It also moved the "Sound
   on" button off the panel's title and hid the duplicate chips;
   both of those went on 16 Oct when the chips were removed and the
   sound button was replaced by a poster play control. What is left
   is the captions, below.

   .is-ended would cover this too, but that class only lands when the
   video FINISHES — thirty-two seconds too late. .is-panel is set by
   flow.js on timeupdate once currentTime passes the join, and
   cleared on replay and on seeking back into the footage. */


/* The native caption rendering sits inside the video box and the
   browser positions it; nothing of ours needs to move out of the
   panel's way any more. */

/* The phone block that used to live here went with the sound button
   (16 Oct). It reflowed the stage — auto height, paper background, a
   flex column, progress bar hidden — purely so a "Sound on" button
   could sit under the panel instead of across row 2. With no button
   on the video there is nothing to make room for, and the stage keeps
   its 16:9 on a phone like everywhere else. */

/* The caption/chip collision this used to fix is GONE WITH THE CHIPS
   (16 Oct). It was real — at 390 the pill occupied y 432-483 and the
   chip stack 441-503, "Real footage" printed through the caption —
   but there is no longer anything in that corner to collide with, so
   the caption keeps its default position. The .ctxfilm-chip sizing
   that was here went with it; the END PANEL still uses that class and
   is styled by .ctxend-chip, which is untouched. */

/* THE FILM IS CLICK-TO-PLAY NOW, SO IT WEARS THE EXPLAINER'S CONTROL.
   16 Oct 2026. The poster button reuses .explainer-play and
   .explainer-badge verbatim — Guillaume asked for "the same style as
   the previous video on How it works", and the way to be sure it is
   the same style is to use the same rules rather than copies of them
   that drift. .ctxfilm-poster-btn only carries what differs.

   UPDATED 8 Nov 2026: .explainer-watch was the third shared rule and
   is deleted with the player look. What both covers now share instead
   is the .vp-* set AND THE MARKUP ITSELF — the overlay is one
   {{ define "video_cover_overlay" }} in _chrome.html, so the furniture
   can no longer drift between the two films even by hand. */
.ctxfilm-poster-btn {
  /* .explainer-play sets border-radius: inherit, and the stage's 6px
     is what this needs, so there is nothing to override there. The
     stage paints #000 behind the video; until the film is started the
     poster is what should show through. */
  background: transparent;
  border: 0;
  padding: 0;
  cursor: pointer;
}

/* IT INHERITS `display: none` FROM .explainer-play, AND THAT IS THE
   CONTRACT, NOT A BUG TO OVERRIDE.
   ---------------------------------------------------------------
   .explainer-play is hidden by default so that a visitor without JS
   never meets a dead control over a video — flow.js adds .is-ready
   once the handler is actually wired. The reasoning is written out
   at that rule and it applies here identically: this button's only
   job is to call play().
   
   But .is-ready is added to .explainer-stage, and this button lives
   in .ctxfilm-stage, so it would stay hidden for ever. Rather than
   force it visible — which would hand a no-JS visitor exactly the
   dead button that rule exists to prevent — the film's stage earns
   the same class the same way, from its own controller.
   
   Found by a click that did nothing: the button measured 0x0 with
   display:none, and elementFromPoint at its centre returned the
   SECTION behind it. */
.ctxfilm-stage.is-ready .ctxfilm-poster-btn { display: block; }
/* Once it is playing, the poster furniture must not come back on
   top of the picture. The button is hidden in the DOM, but the
   progress bar and captions share the stage and are positioned
   against it. */
.ctxfilm-stage.is-started .ctxfilm-poster-btn { display: none; }

/* THE CHIPS ARE GONE FROM THE FILM. .ctxfilm-chips no longer exists
   in the markup; the rule is left dead rather than deleted only if
   something else used it — nothing does, so it goes. The honesty
   claim it carried moved into .ctxai-hedge above the film, which is
   where the "placed by hand" sentence now lives. */

/* THE DURATION-DROPS-ON-A-PHONE RULE IS GONE — 8 Nov 2026, with the
   player look. It hid .badge-secs below 37.5rem because "70 seconds ·
   sound on · planned capability" measured 248px inside a 326px video
   — 76% of the width, against 42% for the explainer's — and crowded
   the footage's own "Oncoming vehicle" card in the opposite corner.

   The duration has LEFT this chip for the player strip, where the
   total is read off the film's metadata rather than typed. So the chip
   is "sound on · planned capability" at every width, short enough to
   need no override, and the element the rule selected no longer
   exists. Deleted rather than left dead — a selector matching nothing
   is a claim that something is still there. */

/* ===================================================================
   THE THEME SWITCH — System / Light / Dark
   ===================================================================

   WEBSITE V2, 22 Oct 2026. The markup is page_theme_switch in
   _chrome.html: a <fieldset> with a visually-hidden <legend> and
   three radios, one per value.

   IT IS A RADIO GROUP AND THAT IS THE WHOLE ACCESSIBILITY STORY.
   Arrow-key navigation between the three, a single tab stop for the
   group, the "Colour theme" grouping announcement and the
   checked/unchecked state all come from native radio inputs and
   NOTHING HERE RE-IMPLEMENTS ANY OF IT. The temptation with a
   segmented control is to build it from buttons with aria-pressed,
   which then needs a keydown handler, a roving tabindex and a
   focus-management story, all of which can be got wrong. Three
   radios cannot be.

   THE INPUT IS NOT display:none AND MUST NOT BECOME IT. It is
   stretched over its label at opacity 0. A display:none or
   visibility:hidden radio is removed from the accessibility tree and
   from the tab order — the group stops being operable by keyboard
   entirely, which is exactly the failure this file has been through
   before. Opacity 0 with real geometry keeps it focusable, hit-
   testable and announced, and the LABEL is what paints.

   THE FOCUS RING IS DRAWN ON THE LABEL, via :has(input:focus-visible),
   because the input itself is invisible. :has() is in every browser
   this site supports (Chrome 105+, Safari 15.4+, Firefox 121+); the
   fallback for anything older is that the ring is drawn on the
   invisible input's own box, which occupies the same rectangle, so
   the ring still appears in the right place. Nothing is lost.

   MEASURED:
     label rest   --ink-soft on --paper        7.60:1 / 8.17:1
     label on     --on-brand on --brand       14.27:1 / 9.72:1

   ON A PHONE THE TEXT IS HIDDEN, NOT REMOVED. .theme-switch-t takes
   the visually-hidden geometry below 30rem, so each control is an
   icon with a real accessible name. Removing the text would leave
   three unnamed radios; a title attribute would name them in a way
   most screen readers announce badly and no voice-control user can
   say out loud.

   MINIMUM 32px TALL AT DESKTOP AND 44px ON A PHONE. The phone value
   is the touch-target floor; the desktop one is what fits a 72px
   header beside a 22px wordmark without the header growing. */
.theme-switch {
  display: inline-flex;
  align-items: center;
  gap: 2px;
  margin: 0;
  padding: 3px;
  border: 1px solid var(--line);
  border-radius: 999px;
  background: var(--paper);
  /* A fieldset carries a default min-width:min-content that refuses
     to shrink and pushes a narrow header sideways. */
  min-width: 0;
}
.theme-switch label {
  position: relative;
  display: inline-flex;
  align-items: center;
  gap: 0.375rem;
  min-height: 32px;
  /* ⚠️ margin: 0 IS THE FIX FOR "too high with respect to the white
     bar", 8 Nov 2026, AND IT TOOK FIVE WRONG GUESSES TO FIND.

     MEASURED: the labels sat 4px from the pill's top and 10px from
     its bottom. The cause was ONE INHERITED DECLARATION —

         label { display: block; font-weight: 500;
                 margin-bottom: 0.375rem; }        app.css:912

     — the global FORM-label rule, giving every label 6px underneath
     for the control below it. These labels have no control below
     them: they ARE the control. The 6px became a flex item's bottom
     margin, so align-items:center centred a 38px box containing a
     32px label plus 6px of margin, and the label landed high.

     WHAT I CHECKED FIRST AND RULED OUT, recorded so nobody repeats
     it: the <legend> (removing it from the DOM changed nothing), the
     inherited line-height (setting it to 0 changed nothing), the
     fieldset's legend-reserved area (a plain <div> behaved
     identically), the UA fieldset padding (zeroing all padding left
     the 6px exactly where it was), and align-items (already center).

     THE LESSON IS CHEAPER THAN THE SEARCH WAS: when a flex child
     sits off-centre and align-items is already correct, read the
     child's MARGIN before reading anything else. getComputedStyle
     said "0px 0px 6px" the whole time. */
  margin: 0;
  padding: 0.25rem 0.6875rem;
  border-radius: 999px;
  font-size: 0.8125rem;
  font-weight: 600;
  line-height: 1.2;
  color: var(--ink-soft);
  cursor: pointer;
  /* The label is the hit target, so it must not shrink below its own
     icon when the header is tight. */
  flex: none;
}
.theme-switch input {
  position: absolute;
  inset: 0;
  width: 100%;
  height: 100%;
  margin: 0;
  opacity: 0;
  cursor: pointer;
}
.theme-switch svg {
  width: 16px;
  height: 16px;
  flex: none;
}
.theme-switch label:has(input:checked) {
  background: var(--brand);
  color: var(--on-brand);
}
.theme-switch label:has(input:focus-visible) {
  outline: 3px solid var(--accent);
  outline-offset: 2px;
}
.theme-switch label:hover { color: var(--ink); }
.theme-switch label:has(input:checked):hover { color: var(--on-brand); }

/* THE SWITCH ON THE HOMEPAGE'S OWN HEADER. That header sits on
   --hero-bg rather than on --page, so the resting label colour has to
   come off the hero's soft ink or it reads as disabled. The fieldset
   keeps --paper, which is the board's treatment: a pale pill on the
   sky, not a transparent outline. */
.sitenav .theme-switch label { color: var(--ink-soft); }
.sitenav .theme-switch { flex: none; }

@media (max-width: 48rem) {
  /* ICONS ONLY. The accessible name survives — see the note above. */
  .theme-switch-t {
    position: absolute;
    width: 1px;
    height: 1px;
    margin: -1px;
    padding: 0;
    overflow: hidden;
    clip: rect(0 0 0 0);
    clip-path: inset(50%);
    white-space: nowrap;
    border: 0;
  }
  .theme-switch label {
    min-height: 44px;     /* the touch-target floor */
    min-width: 44px;
    justify-content: center;
    padding: 0.25rem 0.5rem;
    gap: 0;
  }
}

/* The three .site headers (privacy, quote-unavailable, under-17) had
   a bare wrap holding a mark and a wordmark. The switch needs a row
   with the brand at one end and the control at the other; this is
   that row, and nothing else about those headers changed. */
.sitehead-row {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: 0.75rem 1.5rem;
}
.sitehead-brand {
  display: inline-flex;
  align-items: center;
  gap: 0.5rem;
  min-width: 0;
}

/* The mission page's nav is scoped under .mission in mission.css and
   draws its own row; the switch there takes the same pill but has to
   sit on that page's dark hero. See mission.css for the override. */

/* ===================================================================
   DAY / NIGHT ARTWORK PAIRS
   ===================================================================

   WEBSITE V2, 22 Oct 2026. Where the artwork is an <img> rather than
   a CSS background, the theme swap needs an ELEMENT per theme. These
   two classes hide the inactive one.

   ONLY THE ACTIVE THEME'S FILE IS EVER DOWNLOADED, and that is the
   whole reason for doing it this way rather than swapping a src in
   JavaScript. A lazily-loaded <img> inside a display:none subtree is
   never fetched by any engine — so the hidden theme's art costs
   nothing, and a visitor who never switches never pays for the other
   theme at all. Verified on the running page by
   scripts/check-theme.js, which watches the network at both OS
   settings and fails if a file from the inactive directory is
   requested. It is checked rather than asserted because "lazy images
   in a hidden subtree are not fetched" is exactly the kind of claim
   that is true until a browser changes its mind.

   WHY NOT <picture> WITH A prefers-color-scheme <source>. Because a
   <source media> query CANNOT see [data-theme] — media queries read
   the viewport and the OS, never an attribute on <html>. A <picture>
   would therefore follow the OS and silently ignore an explicit
   Light or Dark choice, which is most of the point of having the
   switch. Two elements plus these rules honour both.

   THE NO-JS CASE IS THE LAST PAIR. With JavaScript off nothing sets
   data-theme, so :root[data-theme="dark"] never matches and the day
   art would show on a dark OS. The :not([data-theme]) + media query
   pair below covers exactly that: it applies ONLY when the attribute
   is absent, so it cannot fight theme-init.js when the attribute is
   present.

   EVERY PAIR IS PIXEL-IDENTICAL — verified file by file on the
   shipped bytes, all eight: hero-home-wide 1942x809, hero-home-phone
   1024x768, mission-people-wide 1448x1086, mission-people-phone
   1024x768, context-pedestrian / context-following 1279x720. A swap
   therefore cannot shift the layout, which is why no rule here sets
   a size.

   NO FILTERS. Never invert, never hue-rotate. The night art is a
   separately drawn picture, not a processed day picture, and a
   filtered version would look like a photograph at dusk rather than
   a scene at night. The brief is explicit and so is the canvas note. */
.art-night { display: none; }
:root[data-theme="dark"] .art-day { display: none; }
:root[data-theme="dark"] .art-night { display: block; }
@media (prefers-color-scheme: dark) {
  :root:not([data-theme]) .art-day { display: none; }
  :root:not([data-theme]) .art-night { display: block; }
}

/* ====================================================================
   THE HONEYPOT FIELD — the advisory-quote email gate. 8 Nov 2026.
   ==================================================================== */

/* A field a human never fills and a form-filling bot usually does.
   docs/tasks/advisory-quote-email-gate.md asks for one; the handler
   treats a non-empty value as success and writes nothing.

   ⚠️ NOT .visually-hidden, WHICH ALREADY EXISTS AND IS THE WRONG
   TOOL. That class is documented, in terms, for "screen-reader-only
   text that must STAY IN THE ACCESSIBILITY TREE". A honeypot wants
   the exact opposite: out of the tree, out of the tab order, and
   announced to nobody. Reusing it would put a field called "Website"
   in front of every screen-reader user and invite them to fill the
   trap.

   ⚠️ AND NOT display:none, WHICH IS THE OBVIOUS CHOICE AND IS WORSE.
   Some bots skip display:none fields precisely because it is the
   obvious choice, so the trap stops catching anything. Off-canvas
   positioning keeps the field in the layout — and therefore fillable
   by anything working from the DOM — while putting it where no eye
   reaches.

   The template pairs this with aria-hidden="true" on the wrapper and
   tabindex="-1" on the input, so all three routes to it are closed:
   sight, keyboard and the accessibility tree. */
.honeypot {
  position: absolute;
  left: -9999px;
  width: 1px;
  height: 1px;
  overflow: hidden;
}

/* ====================================================================
   "WE'VE EMAILED YOU A COPY" — the result page's confirmation box.
   8 Nov 2026, from the refreshed ResultV2 boards.
   ==================================================================== */

/* IT REPLACED .emailcard, which was a FORM (heading, field, button).
   This is a statement, so it is a different shape: a tick in a disc
   beside two lines of text. The old rules are left in place because
   quote_join.html and the form_page shell still use parts of them.

   MEASURED OFF THE BOARD, ResultV2.dc.html:133 —
     padding       18px 22px        -> 1.125rem 1.375rem
     radius        20px             -> 1.25rem
     max-width     720px            -> 45rem, same as .emailcard
     gap           14px             -> 0.875rem
     disc          36px, 50%        -> 2.25rem
     tick          18px             -> set on the <svg>
     h2            17px/700         -> 1.0625rem
     body          15px/1.5
     fine          13px, --ink-soft
   Converted to rem per the quote-layout contract (R5: "In rem, not
   px"); 16px base, so the numbers above divide exactly.

   THE BORDER IS --line, 1px, NOT .emailcard's 2px --accent. The board
   draws a 1px hairline here, and it is right: --accent's heavy edge
   marks something you ACT on, and there is nothing to act on in a
   confirmation. The two links inside it keep their own focus rings. */
.emailsent {
  display: flex;
  align-items: flex-start;
  gap: 0.875rem;
  margin-block: 2rem;
  padding: 1.125rem 1.375rem;
  max-width: 45rem;
  background: var(--paper);
  border: 1px solid var(--line);
  border-radius: 1.25rem;
}

/* The tick disc. --ok-bg / --ok-ink are the tokens the step indicator
   already uses for a completed step, so a completed thing looks the
   same wherever it appears. */
.emailsent-tick {
  flex: none;
  display: grid;
  place-items: center;
  width: 2.25rem;
  height: 2.25rem;
  border-radius: 50%;
  background: var(--ok-bg);
  color: var(--ok-ink);
}

/* min-width: 0 so a long masked address wraps instead of stretching
   the flex item past the box — the classic flex overflow. */
.emailsent-body { min-width: 0; }

.emailsent-h {
  margin: 0 0 0.25rem;
  font-size: 1.0625rem;
  font-weight: 700;
  line-height: 1.3;
}

.emailsent-lede {
  margin: 0;
  font-size: 0.9375rem;
  line-height: 1.5;
}

/* The spam-folder line, the resend and the change-address link. It is
   a flex row that WRAPS rather than a paragraph, because it contains a
   <form> — see the markup note in quote_result.html about why a form
   cannot live inside a <p>. */
.emailsent-fine {
  display: flex;
  flex-wrap: wrap;
  align-items: baseline;
  gap: 0.25rem;
  margin-top: 0.375rem;
  font-size: 0.8125rem;
  line-height: 1.55;
  color: var(--ink-soft);
}

/* The resend form is inline so its button sits in the sentence. */
.emailsent-resend { display: inline; margin: 0; }

.emailsent-ok { font-weight: 700; color: var(--ok-ink); }
