/* ============================================================================
   SEEK — landing page stylesheet.  rev 13 — literal reconstruction.

   The architecture is the reference's, measured rather than remembered:

     · ONE fixed full-viewport atmosphere layer behind everything (never
       scrolls, never parallaxes, never crossfades, no blur filter).
     · A load-time curtain that consumes zero scroll, never blocks a click and
       holds the page still only while its sheet is up.
     · Opaque panels that ride 45px over the previous panel on a negative
       margin. The #moment and #information overlaps are scroll ranges
       (page.js §4b); the rest are static CSS.
     · EXACTLY ONE position:sticky element on desktop — the product story's
       left column — with four ordinary flow cards scrolling past it.
     · One parallax (the Why SEEK lanes, page.js §3b); the hero photograph is a
       static crop and the band's scales on a range.
     · Motion is MOTION-SPEC's vocabulary: one-shot entrances, the story cards'
       reversible entrance, scroll ranges and one footer spring.

   Every pinned rule is scoped to [data-live], written by the runtime, so the
   page is complete and readable with the script removed.
   ========================================================================= */

/* NUNITO is the SEEK app's interface face — 500/600/700/800, no 400 anywhere.
   It is used ONLY inside the product mockups (--f-app): the page's own editorial
   voice stays Newsreader + Instrument Sans, and the app surfaces inside it are
   set in the app's own type.

   IT IS NO LONGER LOADED FROM HERE. An @import is the one font-loading form the
   browser cannot begin early: it is discovered only after this stylesheet has
   been fetched and parsed, so the request is serialised behind it and behind
   the two preconnects it cannot use. Nunito now travels in the SAME head <link>
   as Newsreader and Instrument Sans, emitted by tools/build-v2.mjs, which the
   preload scanner sees in the first bytes of the document. Nothing about the
   type changed — only when the bytes are asked for. */

/* ---------------------------------------------------------------- 1. TOKENS */
:root {
  color-scheme: dark;

  /* — SEEK palette, dark (default) — taken from the app's frozen seekTokens.
     The hierarchy is the app's, not the old one: CORAL leads, TEAL is reserved
     for SEEK itself speaking, LAVENDER is the border base. --accent is teal
     because the only thing it marks on this page is SEEK's own voice. */
  --ink:        #F7F2F6;
  --accent:     #1FD3AE;   /* teal — SEEK AI, and nothing else */
  --accent2:    #4FE0C4;
  --coral:      #FF8E7A;   /* primary */
  --lavender:   #A99BFF;
  --gold:       #E8C79A;
  --plum:       #C78FB0;
  --on-accent:  #06231C;

  /* EDITORIAL is the calm reading canvas; DEEP is the cinematic canvas and
     the page's own ground. The light theme repaints EDITORIAL to cream and
     keeps DEEP dark, so the light/dark rhythm survives the theme flip.

     41% of the reference's scroll is a #EAEAEA panel, and that tonal cut is
     how its 40px panel corners read as STRUCTURE rather than as a shadow.
     SEEK cannot go grey, so the separation has to be carried by amplitude
     instead: EDITORIAL is pushed from #171226 (a 12-point step off DEEP,
     which read as one continuous surface) to a 30-point step, so the panel
     boundaries land as a cut in both themes. */
  --editorial:      #251C40;
  --page: #100F18;
  --editorial-fg:   #F7F2F6;
  --editorial-fg2:  #C9C1D6;
  --editorial-fg3:  #9A92AE;
  --editorial-tint: rgba(169,155,255,.09);
  --editorial-line: rgba(169,155,255,.22);

  /* DEEP is now the app's own canvas, #100F18, and DEEP-2 its surfaceRaised.
     The page's old #0B0810/#120D19 were a cooler, flatter near-black than
     anything the product ships. */
  --deep:      #100F18;
  --deep-2:    #1A1520;
  --deep-fg:   #F7F2F6;
  --deep-fg2:  #C9C1D6;
  --deep-fg3:  #9A92AE;
  --deep-tint: rgba(0,0,0,.22);
  --deep-line: rgba(169,155,255,.22);

  /* THE ATMOSPHERE — FOUR FIELDS BY ROLE, NOT BY HUE.

     The tokens are named for the job each field does, because the two themes
     put DIFFERENT hues in the same slot: in dark the KEY and CORE fields LIFT
     the floor and the WASH pulls the right side down; in light the floor is
     already the brightest value, so the same three fields all TINT DOWNWARD
     and the composition survives unchanged. One geometry, one rule, two
     palettes — see §5.

       1  KEY    the broad side light, core off the left edge at 52% vh
       2  CORE   the one small core inside the frame, low left — the warmth
       3  BREATH SEEK's mint, bottom right, barely there
       4  WASH   the dominant field, core past the right edge — TOPMOST

     Channels stay raw so alpha is free. */
  --glow-1: 146,96,148;    /* plum   — the key light            */
  --glow-2: 255,142,122;   /* coral  — the warm core, low left  */
  --glow-3: 31,211,174;    /* teal   — the mint breath          */
  --glow-4: 10,8,18;       /* near-black — the wash that kills the right */
  /* the fixed layer's own floor. It is never seen unmixed — every field
     reaches every pixel — so it is a MID value the fields work either side
     of, not the darkest tone on screen. Measured on the built page: the
     darkest open-field sample is #151220 in the WASH corner and the brightest
     is #694B64 over the KEY, an 84-unit spread on the strongest channel
     against the 6 units this replaces. */
  --atmos-base: #281F38;
  --atmos-1: .52; --atmos-2: .18; --atmos-3: .07; --atmos-4: .66;

  /* DARK'S OWN TRANSLUCENT MATERIAL. The same three roles as light's --glass
     set, expressed entirely in tones the dark theme already uses: the card
     fill is --deep-tint's black wash lifted to .34, the hairline is the
     approved lavender rule, and the shadow is black. No light value and no new
     hue enters dark through this — it is the existing dark card, made frosted
     so the Why SEEK panel is the same material in both themes. */
  --glass:        rgba(10,8,18,.34);
  --glass-hi:     rgba(10,8,18,.46);
  --glass-line:   rgba(169,155,255,.20);
  --glass-fx:     blur(20px) saturate(112%);
  --glass-shadow: 0 1px 2px -1px rgba(0,0,0,.30), 0 24px 46px -26px rgba(0,0,0,.46);

  /* — geometry, measured — 1440 first, 390 second — */
  --pad-x:   clamp(20px, 4.167vw, 60px);      /* 60 / 20  */
  --r-panel: clamp(24px, 2.78vw, 40px);       /* 40 / 24  */
  --r-card:  24px;

  /* THEME TRANSITION. 500ms is the brief's starting point; the curve is a
     decelerate, so most of the change lands early and the tail settles. */
  /* ONE SHARED CLOCK, AND A CURVE THAT ACTUALLY TRACKS IT.

     WHAT WAS WRONG. cubic-bezier(.22,.61,.36,1) is a decelerate: measured, it
     is .594 of the way through at a QUARTER of the run and .873 at the half.
     On a whole-page colour change that reads as a flash — the page arrives
     almost immediately and then the tail creeps. Shortening it to 420ms made
     that worse, not better. Progress against the clock:

                                 t=10%  t=25%  t=50%  t=75%   worst gap
       cubic-bezier(.22,.61,.36,1) .271   .594   .873   .976     .373
       cubic-bezier(.4,0,.2,1)     .026   .237   .776   .959     .276
       cubic-bezier(.37,0,.63,1)   .024   .145   .500   .855     .105
       cubic-bezier(.42,0,.58,1)   .020   .129   .500   .871     .121
       THIS  .35,.18,.6,.85        .066   .214   .536   .831     .081

     The two symmetric curves land the midpoint correctly but leave the first
     tenth of the run almost motionless — .024 and .020 — and open at 0.01x
     average velocity, which is the dead frame on click the brief rules out.
     This one opens at .52x, so the blend is already moving on the first frame,
     and it stays within .081 of the clock the whole way: a quarter through it
     looks a quarter done, halfway it looks halfway, three quarters through it
     is mostly there and still visibly moving. Night to day, not a flash.

     600ms because smoothness is the priority now. A colour blend has no
     momentum, so length costs nothing as long as the movement is continuous —
     and continuity is exactly what the old curve did not have. */
  --theme-dur:   700ms;
  --theme-blend: linear;                     /* the snapshot dissolve      */
  --theme-ease:  cubic-bezier(.35,.18,.6,.85);  /* the no-VT fallback only */
  --r-frame: 24px;
  --r-pill:  12px;
  --r-chip:  8px;
  --overlap: clamp(24px, 3.125vw, 45px);      /* 45 from 1440 up; phones are pinned to 45 below */
  --sec-y:   clamp(100px, 11.11vw, 160px);    /* 160 / 100 */
  --sec-y-b: clamp(100px, 8.33vw, 120px);     /* 120 / 100 */

  /* — type — Newsreader for display, Instrument Sans for UI — */
  --f-display: 'Newsreader', Georgia, 'Times New Roman', serif;
  --f-ui: 'Instrument Sans', system-ui, -apple-system, 'Segoe UI', sans-serif;
  /* the app's face, used only inside the product mockups */
  /* ── THE APP'S PALETTE ROLES, BOTH THEMES ──────────────────────────────────
     Verbatim from matching-app-mobile/src/theme/seekTokens.ts. Named --ap-*
     and declared HERE rather than per-section because every product mockup on
     the page — the Full Profile phone, the four story screens, the interview
     demo — is the same product and must answer to one palette. Previously the
     roles lived on #understanding only and .ui hard-coded its own copies, so
     the two could drift and only one of them could ever follow a theme.

     LIGHT IS NOT AN INVERSION. It is the app's own approved light treatment:
     porcelain canvas, cards that lean faintly LAVENDER (the temperature step
     between page and card is what separates them, not a heavy border), plum
     shadows rather than black, and teal that drops to #06614F wherever it has
     to carry small text — #0A8E77 fails AA below 18.66px on porcelain. The
     cream/beige direction is recorded in the token file as tried and REJECTED
     and appears nowhere here. */
  --ap-canvas: #100F18;
  --ap-surface: #141020;
  --ap-ink: #F7F2F6;
  --ap-ink2: #C9C1D6;
  --ap-muted: #9A92AE;
  --ap-chip-fill: rgba(169,155,255,.09);
  --ap-chip-border: rgba(169,155,255,.22);
  --ap-rule: rgba(247,242,246,.16);
  --ap-teal: #1FD3AE;
  --ap-teal-text: #1FD3AE;
  --ap-teal-fill: rgba(31,211,174,.14);
  --ap-teal-border: rgba(31,211,174,.42);
  --ap-teal-glow: rgba(31,211,174,.5);
  --ap-on-teal: #06231C;
  --ap-verify-fill: rgba(31,211,174,.14);
  --ap-verify-border: rgba(31,211,174,.4);
  --ap-verify-text: #4FE0C4;
  --ap-lav: #A99BFF;
  /* the device shell needs no lift on a dark page; on porcelain it does */
  --ap-surface2: #1A1520;                  /* dark.surfaceRaised */
  --ap-floating: rgba(26,22,38,.86);       /* dark.floating      */
  --ap-veil: rgba(5,4,10,.74);             /* the in-app sheet scrim */
  --ap-coral: #FF8E7A;                     /* dark.coral         */
  --ap-on-coral: #2B120B;                  /* dark.onCoral       */
  /* the WHY I INTRODUCED YOU card — lavender-tinted in both themes, but the
     tint has to be the theme's OWN lavender or it turns milky on porcelain. */
  --ap-why-fill: rgba(169,155,255,.12);
  --ap-why-border: rgba(169,155,255,.3);
  --ap-shell-shadow: 0 0 0 0 rgba(0,0,0,0);
  --ap-atmos:
    radial-gradient(78% 42% at 14% 0%,   rgba(132,88,140,.72) 0%, rgba(132,88,140,0) 62%),
    radial-gradient(60% 30% at 94% 10%,  rgba(169,155,255,.20) 0%, rgba(169,155,255,0) 64%),
    radial-gradient(80% 34% at 50% 104%, rgba(255,142,122,.16) 0%, rgba(255,142,122,0) 66%),
    #100F18;
  --f-app: 'Nunito', system-ui, -apple-system, 'Segoe UI', sans-serif;

  /* ---- THE MICRO-INTERACTION CLOCK ---------------------------------------
     What survives of the old token set: --dur-fast and --ease-enter spell the
     buttons' hover, the skip link and the theme toggle. Section motion is
     MOTION-SPEC's (page.js §3, <style data-claude-opening>, <style data-flip>).

     EVERY NUMBER BELOW IS A MEASUREMENT, not a preference. They are the
     reference's own one-shot beats, read frame by frame off a 30fps capture:

       167 ms   its smallest one-shot (an accordion image swap)
       267 ms   its three-icon row, staggered
       567 ms   its FAQ group
       700 ms   its opening curtain — one full viewport of travel
       967 ms   its masked display-line reveal
       533-767  its four product-story card reveals

     so the ladder was 267 / 700 / 967. --dur-fast is the one surviving rung
     (267); every other clock on the page is MOTION-SPEC §0's.

       --ease-enter  THE SHAPE OF THE REFERENCE'S SETTLE. Its scroll and its
                     entrances decay geometrically at k = 0.8462 per 30fps
                     frame — lambda 5.01/s, tau 200 ms — which normalises on a
                     700 ms beat to y(u) = (1-e^-3.507u)/(1-e^-3.507).
                     cubic-bezier(.23,.84,.56,.95) fits that curve with
                     rmse 0.0009; the previous curve fitted it at 0.0188, 21x
                     worse, because it reached 95% at 0.66 of its duration
                     against the reference's 0.725 and 99% at 0.83 against
                     0.92. Same role, same direction — a longer tail. */
  --ease-enter: cubic-bezier(.23,.84,.56,.95);
  --dur-fast:   .27s;   /* the reference's staggered three-icon row, 267 ms   */

  /* what is left of the accordion clock: both drop-downs are the FLIP's now
     (<script data-flip>), and this curve only dresses the demo's stepper. */
  --acc-ease: cubic-bezier(.33,1,.68,1);

  /* WHY THIS CURVE. The thing that read as mechanical was never the speed, it
     was the VELOCITY PROFILE. Velocity as a multiple of the run's own average,
     sampled at t = 0 / 25 / 50 / 75%:

       cubic-bezier(.2,.8,.2,1)    4.00  1.42  0.32  0.08   <- the old one
       cubic-bezier(.22,.8,.25,1)  3.64  1.60  0.38  0.10
       cubic-bezier(.22,1,.36,1)   4.54  1.56  0.31  0.04
       cubic-bezier(.16,.84,.32,1) 5.25  1.35  0.37  0.10
       cubic-bezier(.33,1,.68,1)   3.03  1.67  0.75  0.20   <- this one

     The old curve collapses 12x in velocity between a quarter and three
     quarters of the way through and is down to a third of average by the
     midpoint: the box lunges, then visibly stops, then creeps. Measured on a
     65px panel, it reaches its last whole pixel at 196ms and spends the
     remaining 89ms moving less than one pixel. That dead tail IS the
     mechanical feeling — an arrival followed by a crawl.

     All three candidate curves sit in the same family and do not fix it; two
     of them start HARDER (the last reaches .274 of full travel in a single
     60Hz frame, which is the snap the brief rules out). This curve is the
     different point in the space: the gentlest start that is still
     unmistakably immediate (.166 of travel in the first 60Hz frame, three
     times the old FAQ's), more than double the velocity still running at the
     midpoint, and the shortest sub-pixel tail of the set at 69ms. It reaches
     .955 at 65% of the run, which is the brief's own shape — quick through the
     first two thirds, soft settle through the last third. */

  /* the ground the page sits on, until a panel says otherwise */
  --fg:   var(--deep-fg);
  --fg-2: var(--deep-fg2);
  --fg-3: var(--deep-fg3);
  --tint: var(--deep-tint);
  --line: var(--deep-line);
}

/* 45px AT EVERY PANEL BOUNDARY ON PHONES (MOTION-SPEC §4). The reference's
   panels overlap 45px whatever the width; SEEK's clamp only reached 45 at
   1440 and left 24 on a phone. Only the phone end of the clamp moves —
   760-1439px keeps the ladder it has always had. */
@media (max-width: 759.98px) { :root { --overlap: 45px; } }

/* ============================ THE LIGHT THEME ===============================
   NOT a de-saturated copy of the dark one, and NOT "everything white".

   The reference this page is measured against is a LIGHT-DOMINANT document —
   roughly 55% light field by area — and its tonal argument is a rhythm:
   light -> photograph -> light -> atmospheric -> dark -> photograph -> light ->
   dark ending. SEEK's sections already sit on that meter, so the light theme's
   job is to let the beats land in the same places:

     hero        photograph        (dark)
     understanding  EDITORIAL      -> PORCELAIN, the light beat
     story       the atmosphere    -> the app's own LIGHT atmosphere, four
                                      radials on porcelain: the "atmospheric"
                                      beat, luminous rather than black
     orb         DEEP              stays plum-black: the dark beat
     moment      photograph        (dark)
     information + closing  EDITORIAL -> porcelain again
     footer      DEEP              the dark ending

   Two things do NOT flip with the theme:
     · the plum-black bands (#orb, the footer card, the hero) — they are the
       rhythm's dark beats and a light page needs them,
     · every product mockup — an app screen is dark because the app is dark.
       Recolouring a screenshot to match a website theme is the one thing that
       would make these read as decoration instead of as the product.

   The palette is the app's light treatment verbatim. There is no cream in it:
   seekTokens records #F4EDE5 / #F7F0E8 / #FAF5EF as tried and REJECTED, so the
   canvas is warm porcelain #F4EFEA and the shadow is plum, never black. */
html[data-theme="light"] {
  color-scheme: light;

  /* ══ THE REFERENCE-2 MATERIAL ═════════════════════════════════════════════
     ONE neutral canvas, ONE darker fill step, white chrome, and ONE ink at
     three strengths. No wash, no hue, no hairline, no shadow: chapters are
     separated by luminance and material (dark and photographic chapters
     between the two light ones), never by a sixth near-white.
       canvas   #EAEAE8        the two light chapters, the page ground
       fill     5.5% ink       cards, chips, FAQ rows, the Why SEEK frame
                               (#DEDEDC over the canvas)
       chrome   #FFFFFF        nav, buttons, inline labels
       ink      #111111        15.9:1 on the canvas
       ink-2    ink at .64     5.3:1  — secondary copy, the second tone
       ink-3    ink at .56     4.1:1  — labels only, never body copy
     SEEK teal is spent only where SEEK itself speaks, in small doses. */
  --l-canvas: #EAEAE8;
  --l-fill:   #DEDEDC;
  --l-fill-a: rgba(17,17,17,.055);
  --l-chrome: #FFFFFF;
  --l-ink:    #111111;
  --l-ink-2:  rgba(17,17,17,.64);
  --l-ink-3:  rgba(17,17,17,.56);
  --l-teal:   #0B8A75;
  --l-teal-text: #076253;

  --page:           var(--l-canvas);
  --editorial:      var(--l-canvas);
  --editorial-fg:   var(--l-ink);
  --editorial-fg2:  var(--l-ink-2);
  --editorial-fg3:  var(--l-ink-3);
  --editorial-tint: var(--l-fill-a);
  --editorial-line: transparent;

  /* THE APP'S OWN LIGHT ROLES — only the Why SEEK phone renders light app UI;
     the product screens inside the two dark chapters take the dark roles back
     (see the system layer at the end of this file). Neutral surfaces, the
     app's own teal and its own lavender "why" card, which is product UI. */
  --ap-canvas: #F6F5F3;
  --ap-surface: rgba(255,255,255,.88);
  --ap-surface2: rgba(255,255,255,.94);
  --ap-floating: rgba(255,255,255,.97);
  --ap-ink: #1E1D1F;
  --ap-ink2: #454447;
  --ap-muted: #67666A;
  --ap-chip-fill: rgba(27,27,26,.05);
  --ap-chip-border: rgba(27,27,26,.14);
  --ap-rule: rgba(27,27,26,.11);
  --ap-veil: rgba(20,20,19,.40);
  --ap-teal: #0A8E77;
  --ap-teal-text: #06614F;
  --ap-teal-fill: rgba(10,142,119,.08);
  --ap-teal-border: rgba(10,142,119,.28);
  --ap-teal-glow: rgba(10,142,119,.22);
  --ap-on-teal: #FFFCF8;
  --ap-verify-fill: rgba(10,142,119,.10);
  --ap-verify-border: rgba(10,142,119,.36);
  --ap-verify-text: #06614F;
  --ap-lav: #5F5B96;
  --ap-coral: #B4543C;
  --ap-on-coral: #FFFCF8;
  --ap-why-fill: rgba(92,78,132,.08);
  --ap-why-border: rgba(92,78,132,.20);
  --ap-atmos: #F6F5F3;
}

/* ------------------------------- 1b. LENIS ---------------------------------
   Back for desktop only (final motion pass; page.js §0b builds it only on a
   fine pointer, never on a phone). The four rules lenis/dist/lenis.css ships,
   inlined rather than fetched.

   Lenis does NOT transform the document — it drives the real scroll position
   through window.scrollTo on its own rAF — so position:sticky, the fixed
   atmosphere layer, the fixed nav and the fixed curtain all keep working
   natively and .story__pin still pins against the browser's own scroll box.
   Nothing below is a smooth-scroll effect; it is housekeeping. */
html.lenis, html.lenis body { height: auto; }
.lenis:not(.lenis-autoToggle).lenis-stopped { overflow: clip; }
.lenis [data-lenis-prevent],
.lenis [data-lenis-prevent-wheel],
.lenis [data-lenis-prevent-touch] { overscroll-behavior: contain; }
.lenis.lenis-smooth iframe { pointer-events: none; }
.lenis.lenis-autoToggle { transition-property: overflow; transition-duration: 1ms; transition-behavior: allow-discrete; }

/* ------------------------------------------------------------- 2. RESET/BASE */
*, *::before, *::after { box-sizing: border-box; }
html { -webkit-text-size-adjust: 100%; }
body {
  margin: 0;
  background:  var(--page);
  color: var(--fg);
  font-family: var(--f-ui);
  font-size: 16px;
  line-height: 1.2;
  -webkit-font-smoothing: antialiased;
  /* CLIP, never HIDDEN. overflow:hidden on <body> makes it a scroll container
     and a scroll container kills position:sticky inside it — that is exactly
     what hid the reference's one sticky node from instrumentation. */
  overflow-x: clip;
}
img, svg { display: block; max-width: 100%; }
h1, h2, h3, p, figure, blockquote { margin: 0; }
ul, ol { margin: 0; padding: 0; list-style: none; }
a { color: inherit; text-decoration: none; }
button, input { font: inherit; color: inherit; }

:focus-visible {
  outline: 2px solid #0B0810;
  outline-offset: 2px;
  box-shadow: 0 0 0 4px rgba(255,255,255,.92);
  border-radius: 5px;
}
/* the focus reveal moves on TRANSFORM, never on `top` — no layout property is
   animated anywhere on this page */
.skip {
  position: fixed; left: 14px; top: 14px; z-index: 200;
  padding: 10px 16px; border-radius: var(--r-pill);
  background: #FFF; color: #16111F; font-weight: 600;
  transform: translateY(-90px);
  transition: transform var(--dur-fast) var(--ease-enter);
}
.skip:focus { transform: none; }
.sr {
  position: absolute; width: 1px; height: 1px; padding: 0; margin: -1px;
  overflow: hidden; clip: rect(0 0 0 0); white-space: nowrap; border: 0;
}

/* ---------------------------------------------------- 3. THE TYPE SCALE ----
   Every size below is the measured reference value at 1440 and at 390, with
   a linear ramp between them. Nothing here is a guess and nothing reaches for
   a bigger clamp than the reference used. */
.t-hero {
  font-family: var(--f-display); font-weight: 300;
  font-size: clamp(49px, 31.54px + 4.476vw, 96px);   /* 96 / 49 */
  line-height: .95; letter-spacing: -.03em;
}
.t-section {
  font-family: var(--f-display); font-weight: 300;
  font-size: clamp(50px, 32.91px + 4.381vw, 96px);   /* 96 / 50 */
  line-height: .95; letter-spacing: -.05em;          /* tighter than the hero */
}
.t-closing {
  font-family: var(--f-display); font-weight: 300;
  font-size: clamp(56px, 32.23px + 6.095vw, 120px);  /* 120 / 56 */
  line-height: .94; letter-spacing: -.04em;
}
.t-story {
  font-family: var(--f-display); font-weight: 300;
  font-size: 46px; line-height: .95; letter-spacing: -.03em;  /* does NOT scale down */
}
.t-statement {
  font-family: var(--f-display); font-weight: 300;
  font-size: clamp(28px, 26.51px + .381vw, 32px);    /* 32 / 28 */
  line-height: 1.2; letter-spacing: -.02em;
}
.t-statement--tight { line-height: 1.1; }
.t-card {
  font-family: var(--f-display); font-weight: 400;
  font-size: 24px; line-height: 1.2; letter-spacing: -.02em;
}
.t-q {
  font-family: var(--f-ui); font-weight: 500;
  font-size: 18px; line-height: 1.2; letter-spacing: -.02em;
}
.t-body {
  font-family: var(--f-ui); font-weight: 400;
  font-size: 16px; line-height: 1.2; letter-spacing: -.02em;
}
.t-body--loose { line-height: 1.3; }
/* .eyebrow is DELETED with the last mockup that used it. The only uppercase
   micro-labels left on the page are inside the app surfaces, and those are the
   app's own rungs (.ui__eyebrow 10.5 +1.365px, .ui__ilabel 11 +1.2px) — the
   website ladder no longer has a 14px tracked-negative eyebrow to spend. */
.dim { color: var(--fg-2); }

@media (min-width: 1000px) { .t-closing { line-height: .84; } }
/* RULE 3 fixes the closing rung at 120 (1440) and 56 (390) and says nothing
   below 390. At 320 the single word "understanding." is 319px at 56px against
   a 280px column, so 39px of the page's locked closing copy was sliced off by
   the mask's overflow:clip. This rescues the sub-360 band only and leaves
   every rung RULE 3 actually specifies untouched. */
/* #closing-scoped (1,1,0) so it survives `.t-closing { font-size: var(--t-d1) }`
   in RULE 3 below, which is the same specificity and later in the file. As a bare
   `.t-closing` this rescue never applied: at 320 the line rendered at 56px and
   "understanding." was clipped 39px by the mask's own overflow. */
@media (max-width: 359px) { #closing .t-closing { font-size: 46px; } }

/* the one chip form on the page: an eyebrow with a surface under it */
.chip {
  display: inline-flex; align-items: center;
  padding: 6px 12px; border-radius: var(--r-chip);
  font-family: var(--f-ui); font-weight: 500;
  font-size: 14px; line-height: .95; letter-spacing: -.01em;
  color: var(--fg-3);
  background: var(--tint);
  box-shadow: inset 0 0 0 1px var(--line);
}

/* ------------------------------------------------------ 4. LAYOUT PRIMITIVES */
.wrap {
  width: 100%; max-width: calc(1320px + 120px);
  margin-inline: auto; padding-inline: var(--pad-x);
}
.col670 { width: min(670px, 100%); margin-inline: auto; padding-inline: var(--pad-x); }
@media (min-width: 760px) { .col670 { padding-inline: 0; width: 670px; } }

/* A PANEL is the reference's core device: a full-width rounded opaque surface
   that rides 45px over the previous one on a 0-height negative margin. The
   overlap itself is static; §13/§14 add a scrolled offset on top of it. */
.panel {
  position: relative; z-index: 2;
  border-radius: var(--r-panel);
  margin-top: calc(var(--overlap) * -1);
  padding-block: var(--sec-y);
}
.panel--light {
  background: var(--editorial); color: var(--editorial-fg);
  --fg: var(--editorial-fg); --fg-2: var(--editorial-fg2); --fg-3: var(--editorial-fg3);
  --tint: var(--editorial-tint); --line: var(--editorial-line);
}
.panel--deep, .on-dark, .on-atmos {
  color: var(--deep-fg);
  --fg: var(--deep-fg); --fg-2: var(--deep-fg2); --fg-3: var(--deep-fg3);
  --tint: var(--deep-tint); --line: var(--deep-line);
}
/* .on-atmos is the ONE band whose ground is the fixed atmosphere layer rather
   than an opaque panel — the product story. In the dark theme that ground is a
   plum-black field and the band's ink is light; in the light theme the same
   layer is the app's porcelain canvas, so the band's ink has to flip with it.
   .on-dark (the two photographs) never flips: a scrimmed photograph is dark in
   both themes. */
.panel--deep { background: var(--deep); }
/* the last panel runs to the bottom of the document */
.panel--tail { border-radius: var(--r-panel) var(--r-panel) 0 0; padding-bottom: 0; }

/* the display-line mask. overflow:CLIP with a little padding so the reveal is
   hidden but the glyphs are not — Newsreader at 96/91.2 overshoots its line
   box, and a bare overflow:hidden shaved the ascenders. */
.mask { display: block; overflow: clip; padding-block: .14em; margin-block: -.14em; }
.mask__i { display: block; }

/* ===================================================== 5. THE ATMOSPHERE ====
   CORRECTION B. ONE fixed, full-viewport, non-scrolling layer for the whole
   page. Overlapping radial gradients and nothing else — no scrim, no blur
   filter, no animation, no per-section glows. It is the ground for the whole
   product-story band. */
.atmos {
  position: fixed; inset: 0; z-index: 0;
  width: 100vw; height: 100svh;
  pointer-events: none;

  /* ONE BLEND, NOT FOUR CIRCLES.

     THE DEFECT THIS REPLACES was geometric, and it was in BOTH themes. The
     old stack was four short fields — 78%x42% at 14% 0%, 60%x30% at 94% 10%,
     46%x24% at 99% 40%, 80%x34% at 50% 104% — each of which reached alpha 0
     at 62-66% of its own radius. Their reaches were y<29% and y>82%, so a
     horizontal band 53% of the viewport tall had NO contributor in it at all
     and was filled by the flat base colour. Measured live: 80.4% of the open
     field in light and 81.1% in dark was one single hex, and the horizontal
     ramp across the middle of the frame was exactly 0 units.

     THREE THINGS FIX IT, and they are independent:

     1. REACH. Every field is now large enough to touch every pixel. There is
        no band the fields do not reach, so there is no place where the base
        shows through unmixed and no plateau anywhere in the frame.

     2. FALLOFF. Each field is a nine-stop SMOOTHERSTEP ramp — the multipliers
        1 / .984 / .896 / .725 / .5 / .275 / .104 / .016 / 0 are 1-(6t^5-15t^4
        +10t^3) sampled every 12.5%. That curve has ZERO slope at both ends,
        which is the whole point: a two-stop linear gradient has a slope
        DISCONTINUITY where it hits alpha 0, and a discontinuity in the first
        derivative of a smooth field is exactly what the eye reads as the edge
        of a circle. With smootherstep there is no kink at the core and none
        at the rim, so a reader sees ambience and cannot locate a field.

        MEASURED, at an amplitude where 8-bit quantisation cannot hide it —
        one gradient, this file's own CORE geometry, black at alpha .80 on
        white, rendered both ways and nothing else changed:
          linear 2-stop : 24.07 units/100px of slope right up to the rim, then
                          0 in a single 20px step — a 24-unit derivative step,
                          i.e. a visible circle edge.
          smootherstep  :  3.10 units/100px arriving at the rim. 7.8x smaller,
                          and that 3.10 is only the piecewise-linear chord of
                          the last segment; scaled to the alphas this file
                          actually ships it is 0.35 units/100px for the CORE
                          and 0.17 for the BREATH — a third of one 8-bit level
                          per 100 px, which cannot be rendered, let alone seen.
        The only two fields whose rim crosses the frame at all are CORE and
        BREATH; KEY and WASH have theirs off-screen on every side.

     3. DIRECTION. The composition is the reference's, not a vignette: bright
        and warm toward the lower left, cool and deep toward the right edge,
        with one small warm core inside the frame and every other core pushed
        past an edge. The WASH is listed first because it paints LAST — it is
        what kills the right third in both themes.

     AND THE PLATEAU IS GONE, which is the other half of "one blend". Walking
     six scanlines (three horizontal, three vertical) and asking for the
     longest run over which the profile does not move 0.5 sRGB units:
         light  OLD  680 / 1140 / 1420 / 440 / 480 / 280 px, with 71 steps of
                     EXACTLY zero change on one line — one hex, edge to edge
         light  NOW  140 /  200 /  200 / 120 / 140 / 260 px, zero flat steps
         dark   OLD  640 / 1420 / 1420 / 440 / 480 / 420 px, 71 exact zeros
         dark   NOW  120 /  120 /  100 /  80 / 100 / 120 px, zero flat steps
     There is now no run anywhere on either axis, in either theme, longer than
     a third of a viewport over which the field holds still.

     The two themes share this geometry byte for byte and differ only in the
     eight colour/alpha tokens above. Everything structural — position, size,
     reach, falloff, order — is one declaration serving both. */
  background:
    /* WASH   — dominant, core past the RIGHT edge. Painted last, on top. */
    radial-gradient(140% 180% at 130% 36%,
      rgba(var(--glow-4),var(--atmos-4))                 0%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .984))   12.5%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .896))     25%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .725))   37.5%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .5))       50%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .275))   62.5%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .104))     75%,
      rgba(var(--glow-4),calc(var(--atmos-4) * .016))   87.5%,
      rgba(var(--glow-4),0)                              100%),
    /* BREATH — SEEK's mint, low right, barely there */
    radial-gradient(52% 58% at 86% 100%,
      rgba(var(--glow-3),var(--atmos-3))                 0%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .984))   12.5%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .896))     25%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .725))   37.5%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .5))       50%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .275))   62.5%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .104))     75%,
      rgba(var(--glow-3),calc(var(--atmos-3) * .016))   87.5%,
      rgba(var(--glow-3),0)                              100%),
    /* CORE   — the ONE core inside the frame, low left: the warmth */
    radial-gradient(58% 62% at 20% 90%,
      rgba(var(--glow-2),var(--atmos-2))                 0%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .984))   12.5%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .896))     25%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .725))   37.5%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .5))       50%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .275))   62.5%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .104))     75%,
      rgba(var(--glow-2),calc(var(--atmos-2) * .016))   87.5%,
      rgba(var(--glow-2),0)                              100%),
    /* KEY    — the broad side light, core off the LEFT edge */
    radial-gradient(104% 98% at 6% 52%,
      rgba(var(--glow-1),var(--atmos-1))                 0%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .984))   12.5%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .896))     25%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .725))   37.5%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .5))       50%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .275))   62.5%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .104))     75%,
      rgba(var(--glow-1),calc(var(--atmos-1) * .016))   87.5%,
      rgba(var(--glow-1),0)                              100%),
    var(--atmos-base);
}

/* ================================================= 6. THE OPENING CURTAIN ===
   Not here. Every curtain rule — its static frame, the sentence, the clock
   and the scroll lock — is the head's <style data-claude-opening>, so the
   sheet paints with the document's first bytes. */

/* ============================================================= 7. THE NAV ===
   A SOLID WHITE PILL, ~155x50, radius 12, fixed at top 30 and centred — the
   reference's is 144x51 at exactly that top, and it is the single
   highest-contrast object in its hero. A bare 34x18 brand mark was standing in
   for it, which left the hero with no anchor at all: three soft objects on a
   photograph and nothing to fix the eye. The light/dark tile beside it shares
   the pill's surface and radius, so the two read as one lockup.

   The mark inside is the LIGHT-SURFACE mark. The pill is white in both themes,
   so one mark is correct everywhere and no runtime swap is needed.

   It enters once, −85px → 0 on the reference's nav curve (900ms ENAV), and
   then it is STATIC at every width, as the reference's pill is: no hide on
   scroll, no return on scroll-up, no scrolled state. The entrance uses the
   individual translate property, so the −50% centring is never rewritten.
   The opening (head) parks it at the FROM pose and moves only the delay onto
   its clock; without an opening — reduced motion aside — it runs at load. */
.nav {
  position: fixed; z-index: 8;
  top: clamp(14px, 2.08vw, 30px); left: 50%;
  transform: translateX(-50%);
  display: flex; align-items: center; gap: 8px;
  --nav-lift: 0 10px 30px -12px rgba(0,0,0,.55), 0 2px 6px -2px rgba(0,0,0,.35);
  animation: seekNavIn 900ms var(--enav) .1s both;
}
@keyframes seekNavIn { from { translate: 0 -85px; } to { translate: none; } }
/* On the pale ground a dark cast is the wrong instrument — the pill is white on
   near-white, so the edge does the work and the cast only grounds it. */
html[data-theme="light"] .nav { --nav-lift: 0 8px 24px -14px rgba(17,17,17,.26), 0 0 0 1px rgba(17,17,17,.07); }
.nav__pill {
  display: inline-flex; align-items: center; justify-content: center; gap: 10px;
  min-width: 155px; height: 50px; padding: 0 20px;
  border-radius: var(--r-pill);
  background: #FFF; color: #16111F; text-decoration: none;
  /* THE PILL IS THE ONLY FLOATING OBJECT ON THE PAGE AND IT HAD NO ELEVATION,
     so wherever it passed over type it read as a collision rather than as a
     thing in front. A soft cast and a hairline — no blur behind, no tint, no
     change to its geometry, its position or the decision (documented above)
     that it never hides. */
  box-shadow: var(--nav-lift);
}
.nav__pill img { width: 26px; height: auto; }
.nav__word {
  font-family: var(--f-ui); font-weight: 600;
  font-size: 17px; line-height: 1; letter-spacing: .04em;
}

/* ---- THE LIGHT/DARK CONTROL -----------------------------------------------
   A 50x50 white tile beside the pill, on the pill's own radius and the pill's
   own surface, so the nav reads as one lockup of two objects rather than as a
   pill with a setting bolted inside it. Requirements it is built to:

     · IDENTICAL GEOMETRY IN BOTH THEMES. The tile is white in light and in
       dark, exactly as the pill is; nothing about its box, its position or the
       nav's width depends on the theme, so switching cannot shift the layout.
     · NO LAYOUT SHIFT WHEN IT CHANGES. Both glyphs are stacked in the same
       grid cell and cross-rotate; neither is ever added to or removed from
       flow, and the tile never resizes.
     · IDENTICAL ANIMATION IN BOTH DIRECTIONS. The outgoing glyph rotates -90
       and scales to .6 while the incoming one arrives from +90; swapping the
       theme swaps which is which, so light->dark and dark->light are the same
       motion mirrored, on the same duration and the same curve.
     · It is a real <button> with aria-pressed, so it is keyboard operable and
       announced as a toggle. */
/* The control only exists where it can work. `js` is written by the pre-paint
   resolver in the document head — the same three lines that own the theme —
   so with scripting off the nav is the pill alone and there is no dead button
   in it. No flash either: the class lands during head parsing. */
.nav__theme { display: none; }
html.js .nav__theme {
  flex: 0 0 auto;
  width: 50px; height: 50px; padding: 0;
  display: grid; place-items: center;
  border: 0; border-radius: var(--r-pill);
  background: #FFF; color: #16111F;
  box-shadow: var(--nav-lift);
  cursor: pointer;
  transition: transform var(--dur-fast) var(--ease-enter);
}
/* THE LIFT IS A POINTER HOVER, NOT A TOUCH STATE. On iOS a tap leaves :hover
   stuck on the button until the next tap elsewhere — measured on the user's
   phone, the tile rose 1.5pt on the first theme tap and stayed up for the
   rest of the visit. And Safari's grey press flash (-65 luma for 3-4 frames)
   was the only tap highlight left on the page's own controls. */
.nav__theme { -webkit-tap-highlight-color: transparent; }
@media (hover: hover) and (pointer: fine) {
  .nav__theme:hover { transform: translateY(-1.5px); }
}
.nav__theme svg { grid-area: 1 / 1; width: 20px; height: 20px; }
/* the two glyphs live in the SAME grid cell — the swap can never reflow */
.nav__theme .ico-sun,
.nav__theme .ico-moon {
  transition: opacity var(--dur-fast) var(--ease-enter),
              transform var(--dur-fast) var(--ease-enter);
}
.nav__theme .ico-sun  { opacity: 0; transform: rotate(90deg) scale(.6); }
.nav__theme .ico-moon { opacity: 1; transform: none; }
/* in the light theme the pair simply trades places — same distance, same
   duration, same curve, opposite direction */
html[data-theme="light"] .nav__theme .ico-sun  { opacity: 1; transform: none; }
html[data-theme="light"] .nav__theme .ico-moon { opacity: 0; transform: rotate(-90deg) scale(.6); }

@media (max-width: 759px) {
  .nav__pill { min-width: 132px; height: 44px; padding: 0 16px; }
  .nav__pill img { width: 22px; }
  .nav__word { font-size: 15px; }
  .nav__theme { width: 44px; height: 44px; }
  .nav__theme svg { width: 18px; height: 18px; }
}

/* ------------------------------------------------------------- 8. BUTTONS */
.btn {
  display: inline-flex; align-items: center; justify-content: center;
  height: 42px; padding: 0 20px; flex: 0 0 auto;
  border-radius: var(--r-chip); border: 0; cursor: pointer;
  font-family: var(--f-ui); font-size: 15px; font-weight: 600;
  letter-spacing: -.01em; white-space: nowrap;
  transition: transform var(--dur-fast) var(--ease-enter);
}
.btn:hover { transform: translateY(-1.5px); }
.btn--light { background: #FFF; color: #16111F; }
.btn--sm { height: 40px; padding: 0 18px; font-size: 14px; }

/* ================================================================= 9. HERO ==
   1.06 viewports, full-bleed, radius 0, headline block starting 51% down.
   A STATIC cover crop — no parallax, no sticky plane, no scale settle. */
#hero {
  position: relative; z-index: 2;
  min-height: 106svh;
  overflow: clip;
  padding-bottom: 40px;
  background: #0B0810;
}
.hero__img {
  position: absolute; inset: 0;
  width: 100%; height: 100%;
  object-fit: cover; object-position: 70% 42%;
}
/* THE CLIP. Identical geometry to .hero__img above — same absolute box, same
   cover fit, same object-position — which is what makes the handover invisible:
   two layers showing the same picture at the same crop, only one of them moving.

   AND NO TRANSFORM. There is deliberately no scale() here. The camera push is
   already in the footage; a CSS zoom on top would stack a second push onto it
   and close the frame twice as far. If a transform ever appears on this rule,
   that is the bug.

   Revealed by page.js in one frame once the clip is PLAYING, over a poster that
   is its frame 0, so the pixels swapped in are the pixels already on screen. No
   transition: a fade would be a visible event where none is wanted, and the
   reference's clip has none. */
.hero__vid {
  position: absolute; inset: 0;
  width: 100%; height: 100%;
  object-fit: cover; object-position: 70% 42%;
  opacity: 0; pointer-events: none;
}
.hero__vid[data-on] { opacity: 1; }

.hero__scrim {
  position: absolute; inset: 0;
  background: linear-gradient(180deg,
    rgba(6,4,10,.55) 0%, rgba(6,4,10,.16) 24%,
    rgba(6,4,10,.36) 48%, rgba(6,4,10,.62) 74%, rgba(6,4,10,.84) 100%);
}
/* MEASURED ON THE REFERENCE, settled: its headline+lede block runs 222->501 in
   a 709 viewport, so the block's TOP is 222/709 = 31svh, and the frosted plate
   follows at 602 — a gap of exactly 101. Everything below is spaced on the
   reference's own gaps: 44 to the lede, 100 to the plate.

   31, not the 38 this carried while the plate did not exist: SEEK's lede runs
   to three lines where the reference's runs to two, and at 38svh the plate
   landed below the fold on a 708-tall viewport. At 31svh the whole four-object
   composition is in the first screen, as the reference's is. */
.hero__inner {
  position: relative;
  padding-top: 31svh;
  display: grid; justify-items: center; text-align: center;
  gap: 0;
}
.hero__h { color: #FFF; max-width: 16ch; }
/* The reference's lede is a TIGHT 278px block of two lines, not a full-width
   paragraph. SEEK's locked sentence is roughly twice as long, so 278 renders
   four lines and pushes the CTA bar onto the fold; 360 is the narrowest width
   that keeps the block to three and still leaves the bottom of the hero open,
   and it is 1.29x the reference's block rather than the 2.2x it was. */
.hero__lede {
  color: rgba(255,255,255,.9);
  font-size: 19px; max-width: 470px; line-height: 1.36; letter-spacing: -.015em;
  margin-top: 32px;
}

/* THE FROSTED CTA PLATE — the reference's fourth hero object, restored to its
   measured geometry: 623x50 centred, radius 12, background rgba(0,0,0,.25),
   backdrop-filter blur(10px), padding 4 4 4 16, gap 24, a sentence on the left
   and a white 140x42 button on the right. Deleting it left the hero with two
   objects where the reference has four, and the argument that a white button
   competes with the headline is the reference's own composition — the plate is
   50px tall against a 96px headline and sits 100px below the lede.

   This is the ONLY frosted surface on the page, and it is here because the
   reference puts one exactly here. */
.hero__cta {
  margin-top: clamp(40px, 6.94vw, 100px);
  width: min(623px, 100%); height: 50px;
  display: flex; align-items: center; justify-content: space-between; gap: 24px;
  padding: 4px 4px 4px 16px;
  border-radius: var(--r-pill);
  background: rgba(0,0,0,.25);
  -webkit-backdrop-filter: blur(10px);
  backdrop-filter: blur(10px);
}
.hero__cta-t {
  margin: 0; white-space: nowrap;
  font-family: var(--f-ui); font-size: 15px; line-height: 1.2; letter-spacing: -.01em;
  color: rgba(255,255,255,.86);
}
/* the reference's plate button is 140x42 on radius 12, not the page's 8 */
.hero__cta-b { width: 140px; padding: 0; border-radius: var(--r-pill); }

@media (max-width: 999px) {
  #hero { min-height: 100svh; padding-bottom: 60px; }
  .hero__inner { padding-top: 42svh; }
  .hero__lede { margin-top: 24px; }
  /* AT 390 THE PLATE STAYS. It was dissolved here on the belief that the
     reference drops it at that width; measured at 390 the reference KEEPS it —
     two lines of text on rgba(0,0,0,.25) with the white button dropped below.
     Dissolving it left the beta line naked on a bright photograph, which is
     the lowest-contrast type on the mobile page, so the desktop repair now
     carries through: same surface, same alpha, same blur, stacked. */
  .hero__cta {
    width: auto; max-width: 296px; height: auto;
    margin-top: 32px;
    display: grid; justify-items: center; gap: 12px;
    padding: 14px 16px 16px;
  }
  .hero__cta-t { white-space: normal; text-align: center; line-height: 1.45; }
  .hero__cta-b { width: 100%; padding: 0 20px; margin-top: 0; height: 48px; }   /* was 42 — under the 44pt minimum, on the page's primary conversion */
}

/* ---- THE PHONE HERO IS A LARGE-VIEWPORT BOX (MOTION-SPEC §4) --------------
   100lvh, as the reference's 100vh hero: the next panel's top sits at the
   LARGE viewport minus 45, so on iOS its peek hides under the toolbar and
   never moves as the bar collapses. PHONES ONLY — iPad Safari keeps the svh
   hero above, where a collapsing bar is not part of the composition.

   81 = 60 + the 21px the phone overlap gained (24 -> 45). When the type is
   taller than 100lvh the hero is content-sized, so without this the extra
   overlap would come straight out of the frosted plate's clearance (36 -> 15).
   Where 100lvh wins, the padding is inside the box and changes nothing.

   THE CROP IS THE ONE IT WAS AT 100svh. object-position 42% is 42% of the
   overflow of the CURRENT box, so growing the box svh -> lvh slid the faces
   down behind the fixed headline (17-42px on a real iPhone; headless cannot
   show it, svh = lvh there). Subtracting 42% of the growth puts them back.
   clamp() bounds it by the overflow itself — 100% IS the box minus the image —
   so a height-limited crop, or an engine where lvh = svh, lands on plain 42%
   and no band can open under the picture. */
@media (max-width: 759.98px) {
  #hero { min-height: 100lvh; padding-bottom: 81px; }
  .hero__img,
  .hero__vid { object-position: 70% clamp(100%, calc(42% - (100lvh - 100svh) * .42), 0%); }
}

/* ======================================================= 10. LIGHT PANEL ====
   Four beats over 2.63 viewports: statement · media · statement · three parts.
   Internal offsets are the reference's own: 160 padding, +60 to the media
   frame, +160 to the closing block, 60 between its parts, 160 padding out. */
.statement {
  width: min(656px, 100%); margin-inline: auto;
  display: grid; grid-template-columns: 1fr; gap: 24px;
}
.statement > .chip { justify-self: start; }
.statement__body { display: grid; gap: 18px; }
.statement__body .dim { color: var(--fg-2); }
@media (min-width: 760px) {
  .statement { grid-template-columns: 181px 475px; gap: 0; align-items: start; }
}

/* ---- THE CONTRAST BEAT ----------------------------------------------------
   WHAT THIS REPLACES, AND WHY. This beat was a 1416x577 café photograph of TWO
   PEOPLE with a translucent caption floating on it. Three ideas competed in one
   frame: copy about profiles being incomplete, a picture of a couple who had
   plainly already met, and a subtitle about SEEK's interview. The photograph
   also argued against the section — this beat happens BEFORE SEEK introduces
   anyone, so a couple cannot illustrate it.

   THE BEAT IS NOW ONE EDITORIAL SPREAD, and the composition is the argument:
     LEFT   one person, alone, mid-thought, at 47% of the wrap — a main object
            of the viewport, not an illustration beside text
     RIGHT  a single reading axis: the statement, then the seven-fact version
            of a person, then — under a hairline — what a conversation gives
            you instead

   THE STATEMENT MOVED INTO THE RIGHT COLUMN. That is what closed the ~220px
   dead zone the previous pass left: with the statement in its own full-width
   row above, the copy column was ~430px of content beside a 596px photograph
   and had to be vertically centred, which opened a hole between the statement
   and the first label. One column, one axis, no hole, and the photograph can
   now be much larger without the two halves drifting apart.

   THE ARGUMENT IS MADE BY WEIGHT, NOT BY EXPLANATION. The fragments are 22px
   display face at --fg-2 with no container at all; the understanding line is
   34px display face at full --fg. Nothing has to say "these are shallow" —
   the type ladder already said it. */
/* the chip is a restrained label for the whole section */

/* THE TOP MESSAGE IS SHORT ON PURPOSE. The annotated profile below does the
   explaining now, so this is two statements and one supporting line, held to a
   readable measure rather than run across the full 1320 wrap. */
/* The measure goes on the CHILDREN, not on .why__lead itself. .why__lead also
   carries .wrap, and .wrap is `max-width + margin-inline:auto` — so a
   max-width here made the whole block auto-centre, which left the message
   floating in the middle of the page while the chip above it and the closing
   beat below it both sat at the gutter. */


/* THE SEQUENCE, IN THE PAGE'S OWN EDITORIAL VOICE.

   WHY IT LOOKED LOUD. The three steps were set in .cal__label — Instrument
   Sans 600 — immediately under a Newsreader 300 statement at 32px. A small
   semibold sans carries more ink per character than a large light serif, so
   the supporting line was visually heavier than the headline it supports:
   the hierarchy read upside down, and three of them in a row with three
   arrows read as a checklist rather than as an explanation.

   THE FIX IS WEIGHT AND FACE, NOT SIZE. The steps keep their full brightness
   and grow from 18px to 22px — they are MORE readable, not less — but at
   Newsreader 300 they now sit a clear step below the statement instead of
   fighting it, and they read as the sentence that follows it. Nothing was
   removed and nothing was hidden; the section simply stopped shouting.

   The note is pulled up close underneath so the two read as one block with a
   quiet tail, rather than as two separate messages stacked in the middle of
   the section. */
/* SECONDARY, AND NOW UNMISTAKABLY SO. The face and weight were already right;
   what was still wrong was the distance from the statement above. At 22px in
   full --fg the row sat close enough to the 32px statement to read as a second
   headline. Down a size and onto --fg-2 — the same ink the statement's own
   second line uses — puts a clear step between them: 32 --fg, 32 --fg-2, then
   17/19 --fg-2, then the note at 15.5 --fg-3. Four rungs, each quieter than
   the last, and the eye lands on the phone. */




/* DESKTOP ONLY: the top message centres over the composition it introduces.
   Below 1000px the annotated diagram becomes a stacked list and the page's
   ordinary left-aligned rhythm is the right one, so this does not apply there.
   The reason it applies here: everything under this message — the two column
   heads, the connectors, the phone — is symmetric about the page centre, and a
   left-aligned lead over a centred object leaves the top-right quarter of the
   section visibly empty. Centring the lead is what closes that gap; it adds no
   copy and changes no words. */


/* ---- THE SEEK FULL PROFILE, REBUILT FROM THE RUNNING APP -------------------
   VERIFIED IN THE ANDROID EMULATOR, not read off the source. Pixel 6 AVD,
   1080x2400 at density 420 (2.625 px/dp), signed in against the local backend,
   navigated Discover -> For You -> "i" -> Full profile. Measurements taken off
   that screenshot and reconciled with FULL_PROFILE in discoverDesign.ts:

     photo/body boundary   1156px  = 440.4dp   -> photoHeight 440      OK
     chip height            108px  =  41.1dp   -> chipHeight  42       OK
     chip left inset         63px  =  24.0dp   -> sideInset   24       OK
     pager dash          x516-563  =  17.9dp   -> pagerDashW  18       OK
     canvas                       #100F18      -> dark.canvas          OK
     why-card fill                #232034      -> rgba(169,155,255,.12)
                                                 over canvas = #222034 OK
     chip fill                    #1E1C2D      -> borderFaint over
                                                 canvas = #1D1C2D      OK
     chip text                    #C9C1D6      -> dark.inkSecondary    OK
     connect button               #1FD3AE      -> dark.teal            OK

   THE FRAME IS THE DEVICE VIEWPORT: 411 x 914dp is 1080 x 2400px at this
   density. The content is TALLER than that and is clipped by the screen, which
   is exactly what the emulator shows — its first screen ends part-way through
   the "Looking for / Height" row. That clip is the app's own, not a device
   invented to make the content fit. */
.pscr {
  --pw: 411;                       /* device viewport width, dp  */
  --psh: 914;                      /* device viewport height, dp */
  /* 0.76. This is the ONLY number that changes the device, and it scales the
     whole frame uniformly: the internals are laid out at the app's dp values
     inside .pscr__frame and are transformed as one object, so no font size,
     inset, radius or chip is individually resized. App fidelity is
     arithmetically unaffected by any value here.
     History: .90 -> .68 (too small) -> .76, i.e. +11.8% on the centred
     composition. MUST STAY IN SYNC with --ps on .annot below, which is what
     the connector arithmetic reads. */
  --ps: .76;                       /* scale; mobile steps it down */
  width: calc(var(--pw) * var(--ps) * 1px);
  /* CONTENT-BOX: the page sets border-box globally, which made the bezel eat
     into the screen width and pushed the screen 8px past the shell's right
     edge - a phone with a bezel on three sides and none on the fourth. The
     screen is the DEVICE's width and is not negotiable, so the padding has to
     live outside it. */
  box-sizing: content-box;
  /* ═══════ THE SAME DEVICE AS THE PRODUCT DEMO ═══════════════════════════
     What was here was an 8dp black pad, one lavender inset ring and one soft
     drop — the same "rounded rectangle around a UI" the #orb phones were
     rebuilt out of, and the reason this one still read as a frame while they
     read as hardware. It now uses .phone's own construction, token for token:
     a real gradient border for the polished band (border-box/padding-box, so
     the metal IS the border and the bezel IS the padding), the three-layer
     drop, the chamfer, the glass and the island.
       THE DEVICE DOES NOT CHANGE SIZE. The 8dp shell is SPLIT into band +
     bezel at .phone's own ratio (0.8 : 2.3 of device width, so 2.1 : 5.9 of
     the 8), not added to. Total is still 8dp a side, the device is still
     411 + 16 = 427dp wide, and the fitter's PW = 427 in §Why SEEK of the js
     — which drives --ws-ps, the annot connector and the wall geometry — is
     arithmetically untouched. The front bezel is not thickened: 5.9dp of
     black is LESS than the 8 that was there.
       IT STAYS FRONTAL. No rotateY, no perspective: this device exists to be
     read, the profile is the hero, and a turn would resample every glyph in
     it and move the composition the fitter has already been approved on. Its
     physicality is the band, the bezel, the island, the chamfer, the glass
     and the shadow — which is exactly the cue set #orb's own state 1 uses, so
     the two chapters show the same phone rather than two cousins. */
  --rim: calc(3.2px * var(--ps));       /* the polished band */
  --bez: calc(4.8px * var(--ps));       /* black between band and glass */
  padding: var(--bez);
  border: var(--rim) solid transparent;
  border-radius: calc((46px + 8px) * var(--ps));
  background:
    linear-gradient(#000, #000) padding-box,
    linear-gradient(155deg,
      #C4C2C1 0%, #8A8883 12%, #56555A 28%, #2E2D34 44% 58%,
      #5C5A56 78%, #3B3A40 94% 100%) border-box;
  --drop:
    0 2px 5px rgba(0,0,0,.62),
    0 16px 30px -6px rgba(0,0,0,.45),
    0 46px 90px -24px rgba(0,0,0,.62);
  box-shadow: var(--drop);
}
.pscr__screen {
  position: relative;              /* the island and the glass sit on this */
  width: calc(var(--pw) * var(--ps) * 1px);
  height: calc(var(--psh) * var(--ps) * 1px);
  border-radius: calc(46px * var(--ps));
  overflow: hidden;                /* the device clip - see the note above */
  background: var(--ap-canvas);
  /* the chamfer — the hairline where polished metal meets black glass */
  box-shadow: inset 0 0 0 1px rgba(255,255,255,.10),
              inset 0 1px 0 rgba(255,255,255,.13);
}
/* THE DYNAMIC ISLAND, same geometry as #orb's. The app's hero photograph is
   full-bleed here, which is exactly the case a real iPhone shows it in: the
   island sits ON the picture. It lands at 10-47dp, clear of the photo pager
   at 104dp and between the two floating discs, so nothing in the profile UI
   is covered and nothing moves. */
.pscr__screen::before {
  content: ""; position: absolute; z-index: 2;
  top: calc(10px * var(--ps)); left: 50%;
  width: calc(126px * var(--ps)); height: calc(37px * var(--ps));
  transform: translateX(-50%);
  border-radius: 999px;
  background:
    radial-gradient(circle calc(5px * var(--ps)) at 78% 50%,
      #2C2C3A 0 46%, #15151D 47% 74%, #07070B 75% 92%, rgba(0,0,0,0) 93%),
    #000;
  box-shadow: inset 0 0 0 1px rgba(255,255,255,.045);
}
/* THE GLASS. One wide key sheen off the top-left and a weak edge return along
   the foot; it never crosses the middle of the screen, so Mia's photograph and
   every line of the profile keep their contrast. */
.pscr__screen::after {
  content: ""; position: absolute; inset: 0; border-radius: inherit;
  pointer-events: none; z-index: 3;
  background:
    linear-gradient(158deg, rgba(255,255,255,.10) 0%, rgba(255,255,255,.035) 16%,
      rgba(255,255,255,0) 38%),
    linear-gradient(0deg, rgba(255,255,255,.035) 0%, rgba(255,255,255,0) 22%);
}
/* LIGHT THEME: this app screen is cream, so a white sheen over it is
   arithmetically invisible. What glass over a bright screen actually shows is
   the room — a narrow specular at the corner, and a dark reflection rolling
   off the opposite one. Identical treatment to #orb's phones. */
html[data-theme="light"] .pscr__screen::after {
  background:
    linear-gradient(158deg, rgba(255,255,255,.30) 0%, rgba(255,255,255,.07) 9%,
      rgba(255,255,255,0) 22%),
    linear-gradient(212deg, rgba(16,14,22,.13) 0%, rgba(16,14,22,.045) 17%,
      rgba(16,14,22,0) 40%);
}
.pscr__frame {
  width: calc(var(--pw) * 1px);
  transform: scale(var(--ps));
  transform-origin: 0 0;
  font-family: var(--f-app);       /* Nunito - the app's own face */
  letter-spacing: 0;
}

/* ---- hero: full-bleed photograph, floating chrome ------------------------- */
.pscr__hero { position: relative; height: 440px; overflow: hidden; }
.pscr__photo {
  display: block; width: 100%; height: 100%;
  object-fit: cover; object-position: 50% 30%;
  /* THE APP'S HERO IS A PORTRAIT FRAMING, not a wide scene. cover alone fits
     the master's full 1493 width into the 411dp hero, which leaves the subject
     small in a lot of cafe; the emulator's hero is filled by the person. The
     extra 1.5x shows ~64% of the master's width instead of all of it. It is
     still a DOWNSCALE at DPR 2 - the box wants 822 device px and the visible
     crop supplies ~995 - so nothing is upscaled to buy the framing. */
  transform: scale(1.5);
  transform-origin: 50% 28%;
}
/* seekTokens.scrim / scrimStops - four stops, BYTE-IDENTICAL IN BOTH THEMES.
   That is what the app's `photo` surface role guarantees, and it is why the
   name below stays #FFFCF8 even when the page around it is porcelain. */
.pscr__scrim {
  position: absolute; inset: 0; pointer-events: none;
  background: linear-gradient(180deg,
    rgba(26,18,38,.44)  0%,
    rgba(26,18,38,0)    26%,
    rgba(30,20,44,.16)  46%,
    rgba(22,16,34,.9)   86%);
}
.pscr__disc {
  position: absolute; top: 58px;
  width: 46px; height: 46px; border-radius: 23px;
  display: grid; place-items: center;
  background: rgba(0,0,0,.4);              /* onPhotoDisc       */
  border: 1px solid rgba(255,255,255,.24); /* onPhotoDiscBorder */
  color: #FFFCF8;                          /* onPhoto           */
}
.pscr__disc--back   { left: 22px; }
.pscr__disc--report { right: 22px; }
.pscr__disc--back svg   { width: 17px; height: 17px; display: block; }
.pscr__disc--report svg { width: 15px; height: 15px; display: block; }

/* ONE mark, because the screen renders Math.max(1, photoIds.length) and this
   person has one photograph. Six dots was the previous guess; the emulator
   pixels show a single 18dp dash centred at x516-563, and disprove it. */
.pscr__pager {
  position: absolute; top: 104px; left: 0; right: 0;
  display: flex; justify-content: center; align-items: center; gap: 5px;
}
.pscr__pager i { width: 18px; height: 5px; border-radius: 3px; background: #FFFCF8; }

.pscr__identity { position: absolute; left: 24px; right: 24px; bottom: 16px; }
.pscr__namerow { display: flex; align-items: center; gap: 9px; margin: 0; }
.pscr__name {
  font-weight: 800; font-size: 29px; line-height: 1.16;
  letter-spacing: -.696px;                  /* the app's ABSOLUTE tracking */
  color: #FFFCF8;
}
.pscr__verified {
  height: 28px; border-radius: 14px; border: 1px solid var(--ap-verify-border);
  background: var(--ap-verify-fill); color: var(--ap-verify-text);
  padding: 0 10px; display: inline-flex; align-items: center; gap: 5px;
  font-weight: 800; font-size: 10.4px; line-height: 1; letter-spacing: .832px;
  text-transform: uppercase;
}
.pscr__verified svg { width: 13px; height: 13px; display: block; }
.pscr__occ {
  margin: 5.5px 0 0;
  font-weight: 600; font-size: 13px; line-height: 1.3;
  color: #FFFCF8; opacity: .82;
}

/* ---- scrolled content ----------------------------------------------------- */
.pscr__body { padding: 16px 24px 24px; }
/* blockGap 30 - "nearly double sectionGap, because these are separate subjects
   rather than fields of one record" (discoverDesign.ts). */
.pscr__block { margin-top: 30px; }

/* SEEK'S REASONING CARD. Missing entirely before this pass. */
.pscr__why {
  border-radius: 20px; padding: 14px 16px;
  background: var(--ap-why-fill);
  border: 1px solid var(--ap-why-border);
}
.pscr__whylabel {
  margin: 0; display: flex; align-items: center; gap: 8px;
  font-weight: 800; font-size: 10.4px; line-height: 1; letter-spacing: 1.664px;
  color: var(--ap-lav);
}
/* the orb at 12px - SEEK's identity mark, a small sphere rather than an icon */
.pscr__orb {
  width: 12px; height: 12px; border-radius: 50%; flex: 0 0 auto;
  background: radial-gradient(circle at 34% 30%, #7FEBD3 0%, #1FD3AE 42%, #0C6E5C 100%);
}
/* the master's editorial serif, which is where SEEK's own voice is set */
.pscr__whybody {
  margin: 8px 0 0;
  font-family: var(--f-display); font-weight: 400;
  font-size: 16px; line-height: 23.2px; color: var(--ap-ink);
}

/* HER PROMPT - the largest type on the screen after her name. The question
   stays in the serif; the answer is interface type at full ink weight. */
.pscr__promptq {
  margin: 0; font-family: var(--f-display); font-weight: 400;
  font-size: 19px; line-height: 26px; color: var(--ap-muted);
}
.pscr__prompta {
  margin: 10px 0 0;
  font-weight: 800; font-size: 15.4px; line-height: 24px; color: var(--ap-ink);
}

.pscr__h {
  margin: 0 0 9px; font-weight: 800; font-size: 15.5px; line-height: 1.3;
  color: var(--ap-ink);
}
.pscr__text {
  margin: 0; font-weight: 600; font-size: 13.2px; line-height: 21.4px;
  color: var(--ap-ink2);
}

/* CHIPS - r999 IS the app's radius (borderRadius = half of chipHeight 42), and
   the measured fill / border / text are borderFaint / border / inkSecondary. */
.pscr__chips {
  list-style: none; margin: 14px 0 0; padding: 0;   /* sectionGap 14 */
  display: flex; flex-wrap: wrap; gap: 7px;
}
.pscr__chips li {
  height: 42px; border-radius: 21px;
  padding: 0 15px; display: inline-flex; align-items: center;
  border: 1px solid var(--ap-chip-border);
  background: var(--ap-chip-fill);
  font-weight: 700; font-size: 12.6px; line-height: 1;
  color: var(--ap-ink2);
}

/* THE DETAIL PAIRS. "NOT a table: no rules, no 44pt rows, no full-width
   space-between. A muted label beside its value, wrapping as a normal line of
   text would." The previous build had exactly the full-width ruled rows that
   comment forbids. */
.pscr__facts {
  margin: 0; display: flex; flex-wrap: wrap;
  column-gap: 22px; row-gap: 10px;         /* detailGapX / detailGapY */
}
.pscr__pair { display: flex; align-items: baseline; gap: 7px; }
.pscr__pair dt { font-weight: 600; font-size: 12.6px; color: var(--ap-muted); }
.pscr__pair dd { margin: 0; font-weight: 800; font-size: 13.4px; color: var(--ap-ink); }

/* THE TWO BEATS, AND THE DISTANCE BETWEEN THEM.
   Measured before this pass: the right column was 471px tall against an 837px
   phone, so centring left 165px of dead canvas above the heading and 201px
   below the payoff — the "text floating in a huge empty column" problem. The
   fix is NOT to stretch the gaps until the two columns match; it is to give the
   column enough real mass that the slack left over reads as a margin rather
   than a hole. Mass comes from the payoff (bigger type, below) and from a
   separation large enough to READ as the transition between two ideas: the
   claim above the rule, what SEEK arrived at below it. */


/* ---- THE SECTION'S OWN ATMOSPHERE -----------------------------------------
   The panel is a flat --editorial fill, which in the light theme is one cream
   hex across the whole band and made this section the plainest on the page.
   This is the SAME device the fixed atmosphere layer uses — overlapping
   radials on the --glow tokens, nothing else, no blur and no animation — and
   because those tokens are themed, ONE rule gives both themes: plum/champagne/
   sage over porcelain in light, and the existing plum environment lifted very
   slightly in dark. Alphas are deliberately in the .05-.10 range: at these
   values it reads as light in a room rather than as gradient blobs.
   It is behind everything in the section and takes no pointer events.

   THE ALPHAS ARE PER-THEME, which is the convention --atmos-1..4 already uses,
   and for the same reason: the light glows sit only a few steps from the
   porcelain ground (--glow-1 is 214,168,166 against 252,248,245), so the dark
   theme's alphas are invisible there. Light therefore runs roughly 3x the
   opacity to land the SAME perceived amount of colour. Sage is held lower than
   the other two in both themes — it is the one hue far enough from the ground
   to go from 'light in a room' to 'green wash' quickly.

   THE GROUND IS THIS SECTION'S OWN, and that is the point of the pass. The
   .panel--light fill is --editorial, which in the dark theme is a flat #251C40:
   one mid-plum rectangle behind a warm photograph, which is what read as dull
   and dated. This section drops to a deep near-black aubergine so the
   photograph is the brightest thing in the frame and the plum becomes an
   ambience rather than a block. #information and #closing keep --editorial —
   they are out of scope, and nothing places them beside this section, so the
   two grounds are never compared directly. Light goes the other way, from cream
   #F4EFEA to a cleaner warm porcelain, so the atmosphere has something to sit
   on that is not already beige. */
#understanding {
  position: relative; isolation: isolate;
  --und-base: #120C1B;
  --und-1: .17; --und-2: .085; --und-3: .04;
  /* THE PROFILE CARD'S OWN PALETTE, per theme. The dark values are the app's
     literal interview-card values (#141020 on a rgba(169,155,255,.22) hairline)
     so the mockup matches the real screens further down the page. Light cannot
     reuse them — a near-black card on porcelain reads as a hole — so it
     inverts: an almost-white surface lifted off the ground by a plum-tinted
     hairline and a soft shadow, which is how the same card looks in the app's
     light theme. */
  /* The app's palette roles are declared once, for both themes, at :root — see
     the --ap-* block there. They used to be redeclared here in dark only, which
     is why an ID selector could out-specify a theme override and the phone
     could never follow the page. */
  --annot-line: rgba(247,242,246,.26);
  --annot-dot:  rgba(247,242,246,.62);
  background: var(--und-base);
}
#understanding::before {
  content: ''; position: absolute; inset: 0; z-index: 0;
  border-radius: inherit; pointer-events: none;
  /* PINNED. The dark values are written out rather than read from --glow-*/
  /* --und-*, because those tokens are theme-dependent and reading them is
     exactly what made this layer swap its image mid-transition. */
  background:
    radial-gradient(72% 48% at 10% 4%,  rgba(146,96,148, .17)  0%, rgba(146,96,148, 0) 72%),
    radial-gradient(60% 42% at 94% 26%, rgba(255,142,122, .085) 0%, rgba(255,142,122, 0) 74%),
    radial-gradient(66% 46% at 78% 98%, rgba(31,211,174, .04)  0%, rgba(31,211,174, 0) 76%);
  opacity: 1;
}
html[data-theme="light"] #understanding::before { opacity: 0; }
#understanding > * { position: relative; z-index: 1; }

/* THE CLOSING BEAT OF THE LIGHT PANEL IS THREE OBJECTS, NOT ONE.
   A previous pass deleted the kicker and the three-part row on the recorded
   grounds that the row was "a card grid the reference does not have". That was
   wrong: the row is there, measured live at three 432x350 cards, radius 24,
   padding 40, rgba(0,0,0,.05), at x60 / x504 / x948 — gap 12 inside the 1320
   wrap. Restoring both returns the panel from 1.90 to ~2.6 viewports, which is
   the single largest geometry repair on the page.

   The kicker is the reference's 32/35.2 dim two-line block at the gutter,
   ~424 wide. It is NOT a second display size stacked on the display line — it
   is the same 32px statement rung the panel opens on, used dim. */
/* 160px was the reference's own offset to this closing beat, and it was
   correct when the beat above it was a full-bleed 577px photograph that ran the
   whole width. It is NOT correct now: the spread above ends with a 690px
   photograph on the left and a copy column that stops 115px higher, so the
   reference's 160 sat on top of that overhang and read as one ~275px hole
   under the last sentence. 100px keeps the two beats clearly separate while
   closing the run to ~215px, and the full-width empty band below the
   photograph to 100px. */

/* ====================================================== 11. PRODUCT STORY ===
   CORRECTION C. The left column is pinned; four ordinary flow cards scroll
   past it. No cross-fade, no stacking, no state machine, no scroll controller.
   The chip index follows the cards' own entrance lines (page.js §3 -> §4). */
#story {
  position: relative; z-index: 2;
  margin-top: calc(var(--overlap) * -1);
  padding-block: var(--sec-y) var(--sec-y-b);
  /* transparent: the fixed atmosphere IS this section's ground */
}
.story__grid { display: grid; grid-template-columns: 1fr; gap: 40px; }

/* The pinned column is TWO ANCHORS WITH AIR BETWEEN THEM, not a list: the
   headline and index at the top, the CTA on the very bottom edge, and roughly
   250px of deliberate void in between. A flex column is what makes
   margin-top:auto on the CTA real. */
.story__pin { display: flex; flex-direction: column; align-items: flex-start; }
/* the section eyebrow. Every other beat on the page opens with one; this one
   did not, so the process section had to be recognised from its headline
   alone. Same .chip object, same 24px to the headline the other beats use. */
.story__eyebrow { margin-bottom: 24px; }
.t-story { color: var(--fg); }

/* the index — four pills, no numbers, no menu, no buttons. Orientation only. */
/* THE STAGE LABELS, against the reference measured pill for pill. Everything
   here already matched and is left alone: one translucent fill shared with the
   card (--tint), NO border, NO shadow, width hugging the text, flush with the
   headline and the CTA on the same 60px text edge, and a state carried by the
   INK — the reference's inactive label is 3.06-3.60:1 against its own fill and
   its active one 10.20-11.29:1, and that jump, not the box, is what reads.
   Two numbers were off by a couple of pixels and are corrected: the gap is 9
   (measured 8.5-9.0, was 8) and the corner is 6 (measured 5.2-6.5 with the
   straight asymptote reached at 6.5, was --r-chip's 8). It is emphatically not
   a full pill — a 35-tall full pill would be r=17.5. */
.story__index {
  display: flex; flex-direction: column; align-items: flex-start;
  gap: 9px; margin-top: 40px;
}
.story__chip {
  display: inline-flex; align-items: center;
  height: 35px; padding: 4px 12px;
  border-radius: 6px;
  background: var(--tint);
  font-family: var(--f-ui); font-size: 16px; letter-spacing: -.01em;
  color: var(--fg-3);
}
/* THE HAND-OVER. The active pill grows 36 -> 42 and its label brightens and
   steps 16 -> 18 while the outgoing pill does the reverse, simultaneously,
   the moment page.js §4 moves data-on (a card's top crossing 0.5 of the
   fold). The reference runs height and label on one Framer spring
   {duration .4, bounce .2}: damping ratio .8 and, as Motion solves it,
   w0 22.49/s — 1.5% overshoot at 233ms, settled at ~420ms. Its measured
   series (25/68/87/96/100.3/101.5% at 68-251ms from the scroll jump) is this
   curve to within a frame. linear() is that spring sampled every 10ms;
   engines without linear() keep the cubic above it. The chip's own entrance
   is a WAAPI tween on translate/scale/opacity and never meets this list. */
#story .story__chip {
  transition-property: color, height, font-size;
  transition-duration: .42s;
  transition-timing-function: cubic-bezier(.3, 1.1, .45, 1);
  transition-timing-function: linear(0, 0.022, 0.079, 0.158, 0.249, 0.344, 0.438, 0.527, 0.609, 0.683, 0.748, 0.804, 0.852, 0.891, 0.923, 0.949, 0.969, 0.984, 0.995, 1.004, 1.009, 1.013, 1.014, 1.015, 1.015, 1.014, 1.013, 1.012, 1.01, 1.009, 1.008, 1.006, 1.005, 1.004, 1.003, 1.002, 1.002, 1.001, 1.001, 1.001, 1, 1, 1);
}
.story__chip[data-on="1"] { color: var(--fg); }   /* size step: §11 desktop rules (36 -> 42, 16 -> 18) */
/* TEAL WAS TRIED HERE AND MEASURED OUT. The active step is the obvious home
   for a "selected state" accent, and on the built page it does not work: this
   pill's whole design is a CONTRAST JUMP — the reference's inactive label is
   3.06-3.60:1 against its own fill and its active one 10.20-11.29:1, and the
   comment on the rule above says in as many words that "that jump, not the
   box, is what reads". Measured live on the new ground, idle sits at 4.88:1;
   the ink active state sits at 11.38:1, and the deepest teal that still reads
   as teal at 18px sits at 5.07:1. Teal here would take the state signal from
   2.3x to 1.04x — the two pills become the same weight and the only thing
   left carrying the hand-over is the 5px size step.
   So the state stays ink, and the two teal points this chapter is owed are
   spent below where they cost no hierarchy: the eyebrow that names it, and
   the contact dots where the annotations land. */

.story__cta { margin-top: 40px; display: grid; justify-items: start; gap: 16px; width: 100%; }
.story__cta p { color: var(--fg-2); max-width: 34ch; }

/* IMAGE-LED, ONE VISUAL PER CARD. The reference's four cards are not text
   cards: each holds a title, a 300x274 product screenshot occupying 54% of the
   card's height, and a paragraph — the card is full top to bottom. SEEK's held
   text alone, so 55% of card 1 and 75% of card 4 was empty dark rectangle and
   the section's dominant object was a void.

   What goes in is ONE screen per card and nothing else — not the six-part
   composite an earlier pass rightly deleted. Four moments of the same product
   surface: the question, the answer, what SEEK understood, the five layers. */
.story__cards { display: grid; gap: 12px; }
.story__card {
  display: flex; flex-direction: column; justify-content: space-between; gap: 24px;
  padding: 24px;
  border-radius: var(--r-card);
  background: var(--card-bg);
  box-shadow: inset 0 0 0 1px var(--card-line);
  --card-bg: rgba(0,0,0,.2);
  --card-line: rgba(169,155,255,.16);
}
.story__card .t-card { color: var(--fg); }
.story__card > .t-body { color: var(--fg-2); }

/* THE 40x40 MARK, NOW BESIDE THE TITLE RATHER THAN ABOVE IT.
   The reference stacks its 40x40 mark above the card title as an 89px-tall
   head block, and that is what shipped here. It is being laid on its side for
   one reason, stated plainly: a REAL app screen needs the height. The card's
   inner box is 481px on the 561-tall cards, and a stacked head spends 89 of it
   before the product is reached — which is what forced the old mockup down to
   a 274px rectangle holding three lines of type. As a row the head costs 40,
   and the 89px saved is exactly what lets a true 375-wide app screen sit in
   the card at 0.9 scale. The mark is unchanged; only its axis is. */
.card__head { display: flex; align-items: center; gap: 14px; }
.card__ico { display: block; flex: 0 0 auto; color: var(--fg); }
.card__ico svg { display: block; }

/* ═══════════════════════════ THE SEEK APP SURFACE ═══════════════════════════
   Everything under `.ui` is THE PRODUCT, not the website. It is the app's own
   frozen design language — Nunito 500/600/700/800, Newsreader for SEEK's own
   voice, coral as primary, teal reserved for SEEK speaking, LAVENDER borders
   (never white-alpha), the app's radii, and the app's var(--ap-canvas) canvas under its
   own three-radial atmosphere.

   HOW IT IS DRAWN — and why this is not eyeballed.
   Every mockup is authored at TRUE APP PIXELS on a 375-wide screen, which is
   the app's own base test width, and the whole block is then scaled as one
   unit by --s. So 25px really is 25px of Newsreader on the app's measure, the
   send pill really is 52 tall on radius 26, the progress rail really is 3, the
   question eyebrow really is 10.5 at +1.365px of tracking, and the voice card
   really is radius 22 with a 1px lavender border. Nothing inside was re-guessed
   at website scale; one number, --s, converts the whole surface.

   THE SCREENS STAY DARK IN BOTH THEMES. A product mockup shown inside a product
   section is a photograph of the app, and the app is dark. Recolouring it to
   follow the website's theme would make it read as decoration rather than as
   the thing being sold. The light theme changes the CARD around it, never the
   screen inside it. */
.scr {
  --sw: 330;                        /* rendered width, in px  */
  --sh: 400;                        /* authored height, in APP px */
  --s: calc(var(--sw) / 375);       /* 375 = the app's base test width */
  flex: 0 0 auto; margin-inline: auto;
  width: calc(var(--sw) * 1px);
  height: calc(var(--sh) * var(--s) * 1px);
  border-radius: 22px;              /* the app's own card radius */
  overflow: hidden;
  box-shadow: inset 0 0 0 1px var(--ap-chip-border);
  background: var(--ap-canvas);
}
/* The atmosphere is painted at the size of a WHOLE 812-tall screen and the
   fragment windows part of it, so a 400px crop shows the top of the app's
   atmosphere rather than the whole of it compressed into 400px. */
.scr > .ui { width: 375px; height: calc(var(--sh) * 1px); transform: scale(var(--s)); transform-origin: 0 0;
             background-size: 375px 812px; background-repeat: no-repeat; background-position: 0 0; }
.scr > .ui--say { background-position: 0 100%; }

/* the app canvas + the app's dark atmosphere, three radials, barely visible by
   design — never a gradient poster */
.ui {
  position: relative;
  display: flex; flex-direction: column;
  font-family: var(--f-app);
  color: var(--ap-ink);
  letter-spacing: 0;                /* the app puts ZERO tracking on body */
  background: var(--ap-atmos);
}

/* ---- app atoms ---------------------------------------------------------- */
/* the interview header rail: "QUESTION 3 OF 12" + "Save & exit" */
.ui__head { display: flex; align-items: baseline; justify-content: space-between; }
.ui__eyebrow { font-weight: 800; font-size: 10.5px; line-height: 1; letter-spacing: 1.365px; text-transform: uppercase; color: var(--ap-muted); }
.ui__exit    { font-weight: 600; font-size: 13px; line-height: 1; color: var(--ap-muted); }
/* progress: 3 tall, radius 2, hairline track, TEAL fill */
.ui__rail    { height: 3px; border-radius: 2px; background: var(--ap-rule); overflow: hidden; }
.ui__rail i  { display: block; height: 100%; width: var(--p, 25%); border-radius: 2px; background: var(--ap-teal); }
/* the orb — the real one, inlined once as a <symbol> and used at ladder sizes */
.ui__orb     { display: block; margin-inline: auto; width: calc(var(--d) * 1px); height: calc(var(--d) * 1px); }
/* the question: Newsreader 400/25, leading 33.5, NO tracking, centred */
.ui__q       { font-family: var(--f-display); font-weight: 400; font-size: 25px; line-height: 33.5px; letter-spacing: 0; text-align: center; color: var(--ap-ink); text-wrap: balance; }
/* the voice card: radius 22, 1px lavender border, app surface, padding 16 */
.ui__card    { border-radius: 22px; border: 1px solid var(--ap-chip-border); background: var(--ap-surface); padding: 16px; display: grid; gap: 12px; }
.ui__state   { font-weight: 800; font-size: 9.5px; line-height: 1; letter-spacing: 1.235px; text-transform: uppercase; color: var(--ap-teal-text); }
.ui__timer   { font-weight: 800; font-size: 9.5px; line-height: 1; letter-spacing: 1.235px; color: var(--ap-muted); }
/* 10 bars, w3 r2 gap3 in a 44 band, tallest 38, one level through static
   weights — never per-bar randomness, and never animated */
.ui__wave    { height: 44px; display: flex; align-items: center; justify-content: center; gap: 3px; }
.ui__wave i  { width: 3px; border-radius: 2px; background: var(--ap-teal); height: calc(var(--h) * 1px); opacity: var(--o); }
.ui__tlabel  { font-weight: 800; font-size: 9px; line-height: 1; letter-spacing: 1.08px; text-transform: uppercase; color: var(--ap-muted); }
.ui__tbody   { font-weight: 600; font-size: 14px; line-height: 22.4px; color: var(--ap-ink); }
.ui__tlabel + .ui__tbody { margin-top: 8px; }   /* the app's own 8px gap */
/* THE PRIMARY ACTION. One solid teal pill 52 tall on radius 26 carrying the
   mic glyph and one word — not an inert placeholder bar. The app is voice
   first and the affordance has to say so. */
.ui__pill    { height: 52px; border-radius: 26px; background: var(--ap-teal); color: var(--ap-on-teal); display: flex; align-items: center; justify-content: center; gap: 8px; font-weight: 800; font-size: 15.5px; line-height: 1; }
.ui__pill svg { width: 22px; height: 22px; flex: 0 0 auto; }
/* and ONE secondary text link. No border, no fill, no radius. */
.ui__link    { padding: 6px 0; text-align: center; font-weight: 700; font-size: 12.5px; line-height: 1; color: var(--ap-muted); }
/* the reflection stack: a teal GlassPanel with a 2.5px teal rail */
.ui__serif   { font-family: var(--f-display); font-weight: 400; font-size: 26px; line-height: 34px; letter-spacing: 0; color: var(--ap-ink); }
.ui__insight { border-radius: 24px; border: 1px solid var(--ap-teal-border); background: var(--ap-teal-fill); padding: 16px; display: flex; gap: 12px; overflow: hidden; }
.ui__insight > i { flex: 0 0 auto; width: 2.5px; align-self: stretch; min-height: 30px; border-radius: 2px; background: var(--ap-teal); box-shadow: 0 0 5px var(--ap-teal-glow); }
.ui__ilabel  { font-weight: 800; font-size: 11px; line-height: 1; letter-spacing: 1.2px; text-transform: uppercase; color: var(--ap-muted); }
.ui__ibody   { font-weight: 600; font-size: 14.5px; line-height: 22px; color: var(--ap-ink2); margin-top: 8px; }
.ui__stack   { display: grid; gap: 12px; }
/* the five-layer rows, on the reflection stack's own treatment: a teal fill,
   a teal rail, and the app's most common radius (field 16) */
.ui__layer   { display: flex; align-items: center; gap: 12px; height: 40px; padding: 0 14px; border-radius: 16px; border: 1px solid var(--ap-teal-border); background: var(--ap-teal-fill); }
.ui__layer > i { flex: 0 0 auto; width: 2.5px; height: 18px; border-radius: 2px; background: var(--ap-teal); }
.ui__layer span { font-weight: 700; font-size: 13.5px; line-height: 1; color: var(--ap-ink); }
/* the app's own "why introduced" label row — a 12px orb, gap 8, lavender */
.ui__why     { display: flex; align-items: center; gap: 8px; }
.ui__why .ui__orb { margin: 0; }
.ui__why span { font-weight: 800; font-size: 10.4px; line-height: 1; letter-spacing: 1.664px; text-transform: uppercase; color: var(--ap-lav); }

/* ---- the four card screens, each a real fragment of a real screen -------- */
/* 01 — the top of the interview: header, rail, orb, question */
.ui--ask   { padding: 18px 22px 22px; }
.ui--ask .ui__rail { margin-top: 22px; }
.ui--ask .ui__orb  { margin-top: 30px; }
.ui--ask .ui__q    { margin-top: 22px; }
.ui--ask .ui__pill { margin-top: auto; }
/* 02 — the foot of the same screen while it is listening */
.ui--say   { padding: 18px 22px 22px; }
.ui--say .ui__orb  { margin-top: 10px; }
.ui--say .ui__card { margin-top: 18px; }
/* 03 — the reflection stack */
.ui--know  { padding: 24px 22px; }
.ui--know .ui__serif { margin-top: 14px; }
.ui--know .ui__stack { margin-top: 16px; }
/* 04 — the five layers */
.ui--fit   { padding: 24px 22px; }
.ui--fit .ui__serif { margin-top: 14px; }
.ui--fit .ui__stack { margin-top: 18px; gap: 8px; }

/* the orb sprite lives in the document but is never drawn directly */
.orbdefs { position: absolute; width: 0; height: 0; overflow: hidden; }

/* Narrower cards get a smaller --sw and the whole app screen scales with it.
   Nothing inside is re-laid-out and nothing rewraps: it is always the same
   375-wide app screen, drawn smaller. The three rungs are sized to the card's
   inner width at 384 / 352 / 316 — the narrowest viewport each one serves. */
@media (max-width: 999px) { .scr { --sw: 296; } }
@media (max-width: 383px) { .scr { --sw: 264; } }
@media (max-width: 351px) { .scr { --sw: 228; } }

@media (min-width: 1000px) {
  .story__grid {
    grid-template-columns: 464fr 556fr;
    column-gap: clamp(60px, 20.83vw, 300px);
    align-items: stretch;
  }
  /* THE ONE STICKY ELEMENT ON THE PAGE. Scoped to [data-live] so a page with
     no runtime is four ordinary cards beside an ordinary column. 605 tall is
     the reference's measured pin height, and the CTA sits on its bottom edge
     so the void between the index and the CTA is real. */
  #story[data-live] .story__pin { position: sticky; top: 100px; }
  /* 605 IS THE REFERENCE'S PIN HEIGHT AND IT STAYS THE HEIGHT — but it can no
     longer be a floor that costs the reader the whole section. As a flat 605 it
     forced top(100)+605 = 705px of fold, and the guard below then unpinned
     anything shorter, so a 1440x761 window (≈674px of viewport once Chrome's
     chrome is taken) silently lost the sticky column AND the card lifecycle and
     fell back to four permanently-lit cards. That is a real laptop size.
     min() keeps 605 wherever it fits and shrinks it to the fold where it does
     not. The column's own content is 421px (94 headline + 172 index + 115 CTA
     + margins), so the CTA stays reachable; only the deliberate void between
     the index and the CTA gives way, which is the right thing to spend.
     116 = the 100px sticky offset + 16px of clearance. The break-even is
     100svh-116 >= 605, i.e. 721px: at or above that this is EXACTLY 605 and the
     approved composition is untouched. */
  .story__pin { min-height: min(605px, calc(100svh - 116px)); }
  .story__cta { margin-top: auto; padding-top: 40px; }
  .story__cards { gap: 40px; }
  .story__card { padding: 40px; }
  .story__card:nth-child(odd)  { min-height: 580px; }
  .story__card:nth-child(even) { min-height: 561px; }
}
/* RULE 4 — at narrow widths the sticky is dropped entirely and the index is
   deleted. One 350-wide stack: headline, CTA, then the four cards. */
@media (max-width: 999px) {
  .story__index { display: none; }
  .story__pin { position: static; }
}

/* .quote is DELETED. It set the device's question at the website's own
   card-title rung, 24/1.2/-.02em. The question inside a SEEK screen is
   Newsreader 400/25 on 33.5 leading with NO tracking, and that is now .ui__q —
   the app's number, not the page's nearest neighbour to it. */

/* ========================================================= 12. DARK PANEL ===
   1.45 viewports. A butt joint, not an overlap. A 24px-radius card inside,
   the device visual at 37.6vw entering once from the opposite side. */
#orb {
  margin-top: 0;
  padding-block: var(--sec-y) var(--sec-y-b);
}
/* The head is the SAME chip-left / text-left row the light panel uses — the
   one composition the reference repeats at every section opening. It was
   centred here, which broke the pattern one section after SEEK established it,
   so it now reuses .statement verbatim. */
.orb__card {
  position: relative;                 /* the pager is positioned against it */
  margin-top: 60px;
  min-height: 700px;
  padding: 40px;
  border-radius: var(--r-card);
  background: var(--deep-2);
  box-shadow: inset 0 0 0 1px var(--deep-line);
  display: grid; grid-template-columns: 1fr; gap: 40px; align-items: center;
  overflow: hidden;
}
/* ═══════════════ THE LEFT COLUMN IS NOW A REAL CONTROL ═════════════════════
   WHAT WAS HERE: `.orb__rows` as a <dl> of hairline-separated static rows,
   with the comment "They are static rows, not a second accordion: no
   chevrons, no buttons, no new runtime." That sentence is what this pass
   overturns — the block LOOKED interactive and was not, rows 2 and 3 had no
   description in the document at all, and the pager's active pill was welded
   to :first-child.

   THE REFERENCE'S ROWS, MEASURED: filled cards LIGHTER than the panel behind
   them, ~8 radius, 8px gap, 24px side padding, a rounded-rect chevron button
   at the right vertically centred on the title, and — the most-missed fact —
   ZERO colour spent on state. Active, inactive and hovered all read the same
   fill; the only things that say "this one is open" are the description, the
   box height and the chevron angle. That restraint is reproduced here: there
   is no hover tint and no active tint anywhere below.

   TWO DELIBERATE DEVIATIONS FROM THE REFERENCE, both stated rather than
   hidden:
     · RADIUS 12, not the reference's 8. SEEK's own radius ladder has --r-pill
       12 and --r-card 24 and no 8, and the pager tray sitting directly under
       these rows is 12. Internal consistency wins over a foreign design
       system's number.
     · FULL-WIDTH ROWS, not the reference's shrink-to-fit. Its collapsed rows
       are sized to their titles (245/300/337) and animate width on the same
       curve as the height. CSS cannot interpolate between an intrinsic width
       and a definite one without measuring, and SEEK's titles are far shorter
       than its 426px column, so shrink-to-fit would leave three ragged stubs
       and would need a per-frame measure to animate. The height carries the
       whole motion instead. */
/* align-self: CENTER, not stretch. The card already says align-items:center;
   `stretch` was overriding it and hard top-anchoring the stack, which measured
   0px of space above row 1 and 360px of dead void below row 3 on desktop (0 /
   538 at 390). The reference vertically centres the stack and re-centres it per
   state — 210 above / 209 below in a 700-tall section — and the top-heavy
   column with a hole under it was the single most visible difference in the
   side-by-side. Centring is what the card was already asking for. */
.orb__rows { display: flex; flex-direction: column; gap: 8px; max-width: 40ch; align-self: center; }
.orb__row  { border-radius: var(--r-pill); background: rgba(169,155,255,.06); }

/* THE WHOLE ROW IS THE CONTROL — a real <button>, not the chevron and not a
   div with a click handler. Everything a <button> gives for free is the
   point: Tab reaches it, Enter and Space activate it, tap works with no
   hover, and the accessible name is the title itself.

   `button` inherits only font and colour from the reset, so the four
   declarations that make it stop looking like a button are written here,
   exactly as .faq__btn writes them. */
.orb__btn {
  width: 100%;
  display: flex; align-items: center; justify-content: space-between; gap: 20px;
  padding: 19px 24px;                 /* collapsed row = 21.6 + 38 = 59.6 */
  background: transparent; border: 0; cursor: pointer; text-align: left;
  color: var(--deep-fg);
}
/* the page's one focus treatment, unaltered except for the corner radius,
   which is matched to the card it rings */
.orb__btn:focus-visible { border-radius: var(--r-pill); }

/* 32 x 22 rounded rect, NOT a circle — measured constant on every row in
   every frame of every transition. Only the glyph inside rotates, and it is
   one shape reused, so this is a plain reversible rotation between two angles
   on the same clock as the box. 180deg clockwise: the reference is
   inconsistent per row and clockwise is its 2-of-3 majority. */
.orb__ico {
  flex: 0 0 auto; width: 32px; height: 22px;
  display: grid; place-items: center;
  border-radius: 8px;
  background: rgba(169,155,255,.10);
  color: var(--deep-fg);
  /* the rotation's clock is the FLIP's (<style data-flip> §2) */
}
[data-orb-live] .orb__btn[aria-expanded="true"] .orb__ico { transform: rotate(180deg); }
/* no script, no operable control, so no affordance — the FAQ's own rule */
[data-orb]:not([data-orb-live]) .orb__ico { display: none; }

/* THE HEIGHT IS A GRID-ROWS STATE, 0fr <-> 1fr, and it CUTS: the row's motion
   is the FLIP's (<script data-flip>), which measures the before and after
   boxes itself and draws the travel on transforms. Nothing here animates, so a
   click stays a single attribute write and 1->2->3->1 hammering has no queue
   to get out of order and no inline style that can be left stranded.

   The panel's padding lives on the inner element, because only that element
   is the collapsing grid item. */
.orb__panel   { display: grid; grid-template-rows: 1fr; overflow: hidden; }
.orb__panelin { min-height: 0; }
/* the padding belongs to the CONTENT of the collapsing grid item, not to the
   item itself: padding on the item survives a 0fr track and would leave every
   collapsed row 22px too tall. */
.orb__panelin > p { padding: 0 24px 22px; }
/* the collapse belongs to the ACCORDION. Below 760 the rows are one caption
   card whose panels all stay open (§ THE MOBILE DEMO below). */
@media (min-width: 760px) {
  [data-orb-live] .orb__btn[aria-expanded="false"] + .orb__panel { grid-template-rows: 0fr; }
}

/* the description's clock (in after 200ms over 350ms, out in one frame) is the
   FLIP's: <style data-flip> §2 */
@media (min-width: 760px) {
  [data-orb-live] .orb__btn[aria-expanded="false"] + .orb__panel .orb__panelin { opacity: 0; }
}
[data-orb-live] .orb__btn[aria-expanded="true"]  + .orb__panel .orb__panelin { opacity: 1; }

/* REDUCED MOTION. The page-wide rule at the foot of this file collapses every
   transition-DURATION to .001ms, which is enough for anything that only has a
   duration — but the description's fade-in is sequenced with a
   transition-DELAY, and a delay survives that rule. Left alone, a reader who
   asked for less motion would click a row and wait for its description to
   appear. The control must WORK under reduced motion, not merely stop moving.
   (The stage, the pager and the mobile caption carry no delays.) */
@media (prefers-reduced-motion: reduce) {
  .orb__panelin, .orb__ico { transition-delay: 0s !important; }
}

/* ══════════════════════ THE DEVICE STAGE — THREE COMPOSITIONS ══════════════
   THE FINDING THAT SHAPES THIS WHOLE BLOCK: the reference does not swap a
   screenshot inside a fixed phone. The DEVICE ARRANGEMENT ITSELF changes —
   state 1 is two upright phones side by side, state 2 is ONE much larger
   phone in perspective cropped by the stage's bottom edge, state 3 is two
   angled phones crossing. Four independent proofs say each state's stage is
   one pre-rendered layer rather than per-device nodes: the two devices ramp
   opacity in lock-step to within 1.6 points, their separation scales by the
   same factor as their widths, their silhouettes are not affine-consistent,
   and the whole composition appears atomically. The practical consequence is
   the rule this obeys: ANIMATE ONE WRAPPER PER STATE, never the individual
   devices.

   All three compositions are authored in the document and stacked in one box.
   Nothing is created, destroyed, measured or positioned at runtime; a state
   change is one attribute on the card and CSS picks the layer.

   THE STATE CHANGE IS THE REFERENCE'S, VERBATIM (MOTION-SPEC §3). The
   outgoing composition is unmounted there, so here it is CUT: the idle pose
   below carries no transition, and a layer that stops being the active one
   lands on opacity 0 on the very frame the state changes — no fade, no move,
   so two compositions are never on screen together. The incoming is a
   CONSTANT rightward entry whatever the step's direction: opacity 0 -> 1,
   translateX +70px -> 0 and scale .95 -> 1 over 550ms on the reference's
   in-out E2 (--e2); below 760 it is 15px, no scale, 700ms
   (§ THE MOBILE DEMO). A second click mid-entrance cuts the half-faded layer
   the same way and the next one starts again from its idle pose.
   The entrance lives on the individual `translate` and `scale` properties,
   so no per-state pose written as `transform` is ever touched by it. */
.dev {
  position: relative;
  height: 430px;
  /* the composition's dimensions, taken as one unit by the size rungs below,
     so every ratio the reference frame gave survives at every width */
  --pw1: 158;      /* state 1 + state 3 phone screen width            */
  --pw2: 250;      /* state 2's single larger phone                   */
  --dev-gap: 20px; /* state 1 separation                              */
  --dev-t1: 20px;  /* state 1 near phone drop                         */
  --dev-t2: 44px;  /* state 1 far phone drop                          */
  --dev-t3: 130px; /* state 3 lower-left phone drop                   */
  --dev-lap: 44px; /* state 3 overlap                                 */
  --dev-x: 70px;   /* entrance travel, the reference's 70 at all rungs */
}
.stage {
  position: absolute; inset: 0;
  display: flex; align-items: flex-start; justify-content: flex-end;
  pointer-events: none;
  /* IDLE = THE ENTRY POSE, AND NO TRANSITION. The after-change style decides
     the clock, so the layer that just stopped being active takes this rule's
     none: its opacity is 0 on the frame of the click and a half-run entrance
     is cancelled rather than reversed. */
  opacity: 0;
  translate: var(--dev-x) 0;
  scale: .95;
  /* THE PRODUCT DEMO's FIRST-ENTRY STALL. An opacity-0 stage still carries its
     phones' 3D transforms, so WebKit built and PAINTED a layer for every
     phone of every stage — 27.9 MB for the three nobody can see — all in the
     one frame before the card scrolled in: 60-79 ms, measured in Simulator
     Safari, right inside #moment's scrub range. visibility:hidden gives those
     layers no backing store. Nothing on screen changes: the idle stage is
     already opacity 0 and cut in one frame, and visibility has no clock. */
  visibility: hidden;
}
[data-orb-state="1"] .stage--1,
[data-orb-state="2"] .stage--2,
[data-orb-state="3"] .stage--3 {
  visibility: visible;
  opacity: 1;
  translate: none;
  scale: none;
  z-index: 1;
  transition: opacity   .55s var(--e2),
              translate .55s var(--e2),
              scale     .55s var(--e2);
}

/* ═══════════════ THE DEVICE IS AN OBJECT, NOT A FRAME ══════════════════════
   WHAT WAS HERE: an 8px black pad, one 1px lavender inner ring, one 1px black
   outer ring and a single soft drop. That is a bordered rectangle. It reads as
   a UI mockup at every angle because a real phone is identified by three things
   this had none of — a POLISHED CHASSIS BAND that catches light, THICKNESS that
   becomes visible the moment the device turns, and a GLASS surface sitting a
   little proud of the screen beneath it.
   All three are built here, and they are built as geometry rather than as
   decoration, so they stay correct at every rotation the three states use:
     · THE BAND is a real gradient border — border-box/padding-box clipping, so
       the metal is the border and the bezel is the padding. Six stops around
       155deg: two bright (the machined flats catching the key light), two mid,
       two near-black (the chamfer rolling away). Nothing chrome, nothing
       mirrored; this is the brushed-titanium read, not a car bumper.
     · THICKNESS is a real face. ::before and ::after are the two side rails,
       each rotated 90deg out of the picture plane and extruded BACKWARD in the
       phone's own 3D space. At 0deg they are edge-on and invisible; as a state
       turns, the rail that should appear appears, at exactly the width the
       geometry demands. The depth is the real ratio: an iPhone is 8.25mm on
       71.5mm, so --pt is 11.5% of the device's own width. Nothing is faked with
       a painted stripe, which is why the silhouette stays honest.
     · THE BUTTONS are bands in the rail's own gradient — volume pair and the
       action button on the left, the side button lower on the right, at the
       real proportions. At 158-250px of device width a raised band IS the
       button; modelling them as separate boxes would add geometry nobody can
       resolve.
   LIGHT: one key from the upper left, one weak fill from the lower right. The
   drop is three shadows, not one — a tight contact shadow that sits the device
   ON something, a mid body shadow, and a wide ambient. That is what separates
   "photographed object" from "png with a blur behind it". */
.phone {
  position: relative; flex: 0 0 auto;
  --pw: var(--pw1);                       /* rendered screen width, px */
  --ps: calc(var(--pw) / 375);            /* 375 = the app's base test width */
  /* HARDWARE PROPORTIONS, NOT HARDWARE PIXELS. The device is drawn from 134px
     wide (1024 desktop) to 380px wide (1440 desktop), and a fixed 7px bezel is
     5.2% of the narrow one against 1.8% of the wide one — so the small rungs
     grew a fat black surround that no phone has while the large ones went thin.
     All three are now ratios of the device's own width, at the real article's
     numbers: bezel 1.6mm/71.5mm, the visible polished band ~0.55mm, and the
     chassis 8.25mm/71.5mm. The floors keep them from disappearing entirely on
     the narrowest rung. */
  /* THE POLISHED BAND. Two corrections. Chrome FLOORS a border-width to whole
     pixels (1.928px was rendering as 1px), so at every rung below 1440 the
     band was drawing at roughly half its authored width. And the reference's
     frontal phone — which shows no side wall at all — carries a graded edge
     specular 9-11px wide on a 611px face, i.e. 1.5-1.8% of width, against the
     0.8% here. Both point the same way: 1.3%, floored at 2px. This is METAL,
     not bezel; the black between band and glass is untouched below. */
  /* THE BAND IS NOW AN EDGE, NOT A CHASSIS. 1.3% was sized to read as polished
     titanium — a material you are meant to notice. The direction is the
     opposite: the frame is a thin dark line and the screen is the subject, so
     the width comes back to 0.9% (4.2px -> 2.9px on the widest device) and the
     pigment below goes graphite. Chrome still floors border-width to whole
     pixels, so the floor stays. */
  /* 1.1%, FLOORED AT 2px. 0.9% computed to 1.89px on a 210px mobile device and
     Chrome FLOORS a border-width to whole pixels, so the chassis line was
     rendering at 1px — half its authored width, and the reason the edge read as
     absent rather than slim. 1.1% + the 2.3% bezel is 3.4% of device width,
     against the live Fixa AI asset's measured 3.24% / 3.37%. */
  --rim: max(2px, calc(var(--pw) * 0.011px));   /* a slim chassis line, not a chassis */
  --bez: max(3px, calc(var(--pw) * 0.023px));   /* black between band and glass */
  /* CHASSIS DEPTH, AND IT IS DELIBERATELY OVER-SCALE. The true article is
     8.25mm on 71.5mm = 11.5% of width. What the viewer actually reads is not
     the depth but the RATIO of visible rail to visible face, and that is
     (depth/width) * tan(angle) — so the reference photographs win on angle,
     not on thickness: a real phone at 28-45deg shows 6-12% of its face width
     as rail, while these states are locked at 7-16deg by the approved
     composition. At 13.5% and 16deg state 2 showed 3.9%, which is a sliver.
     Angle is not available, so depth is the only free variable: 18.5% brings
     state 2 to 5.3%, i.e. the same visible rail a real 11.5%-thick phone gives
     at 25deg. That is a 61% overstatement of the ONE dimension nothing else in
     frame can be checked against; the bezel, the band and the screen all stay
     at their real ratios, so the device's proportions are unchanged and only
     the wall it stands on gets thicker. State 1 is frontal, so this is
     invisible there. */
  /* 18.5% was an OVERSTATEMENT, authored when the brief asked for readable
     hardware thickness. The brief now asks for the opposite — a slim, quiet
     side — so the drawn depth goes BELOW the real article's 11.5%: 10.5%.
     Measured at 1440: state 2's visible wall 13.0px -> 7.4px, state 3's
     17.6 -> 10.0. The wall still exists, so a turned device still has a side
     and the silhouette stays honest; it simply stops being the loudest thing
     in the frame. */
  /* 9%, MEASURED NOT GUESSED. The lens is now short enough for the turn to read
     (see .stage--2), and a short lens works AGAINST a visible wall — the near
     edge is magnified outward over the far one. Sweeping the drawn depth at the
     shipped lens and turn gave 6% -> 1 device px of wall, 9% -> 3, 11.5% -> 3,
     14% -> 4. 9% is the first value that renders a wall you can see, and it is
     still below the real article's 11.5%, half the 18.5% that read as an iPhone
     ad, and under the 10.5% of the pass before this one. On mobile that is 1.5
     CSS px of wall beside a 2px chassis line: a side, not a rail. */
  --pt: calc((var(--pw) + 18) * 0.09px);        /* 11.5% real -> 9% drawn */
  transform-style: preserve-3d;
  padding: var(--bez);
  border: var(--rim) solid transparent;
  border-radius: calc(46px * var(--ps) + var(--bez) + var(--rim));
  background:
    linear-gradient(#000, #000) padding-box,
    /* GRAPHITE, NOT TITANIUM. The old six stops ran 77% -> 18% luminance with
       two near-white flats: on a near-black stage that is the brightest object
       in the composition after the screen, which is what read as an Apple
       chassis. These run 43% -> 8%, so the frame reads as a dark edge with one
       restrained catch at the key-light corner and the screen is the only
       bright thing left. Same six-stop structure, same --band angle, so every
       state's light direction is unchanged. */
    /* THE SIDE IS A CONTINUOUS LINE, NOT A CORNER CATCH. Re-zoomed on the live
       Fixa AI asset at 3x: both side edges carry a slim mid-grey chassis line
       that runs the WHOLE height of the device, plus three button slabs proud
       of the silhouette. The angled wash below could only light one corner —
       measured 95 at the key corner and 12-21 down the rest of the edge, which
       is why the side read as gone. These two strips are the chassis line: one
       gradient across the rim's own width, repeated down the full height, at
       the reference's measured specular (peak 93-109, never white). The angled
       wash stays underneath so the top and bottom edges keep their direction. */
    linear-gradient(90deg, #2A2D30 0%, #5E6266 40%, #6A6E72 60%, #303336 100%) border-box,
    linear-gradient(90deg, #303336 0%, #6A6E72 40%, #5E6266 60%, #2A2D30 100%) border-box,
    linear-gradient(var(--band, 155deg),
      #5C6064 0%, #2C2F33 9%, #17191C 22%, #0D0E10 42% 60%,
      #212427 82%, #101113 95% 100%) border-box;
  background-origin: padding-box, border-box, border-box, padding-box;
  background-size: auto, var(--rim) 100%, var(--rim) 100%, auto;
  background-position: 0 0, left top, right top, 0 0;
  background-repeat: repeat, no-repeat, no-repeat, repeat;
  --drop:
    0 2px 5px rgba(0,0,0,.62),
    0 16px 30px -6px rgba(0,0,0,.45),
    0 46px 90px -24px rgba(0,0,0,.62);
  box-shadow: var(--drop);
}
/* THE SIDE RAILS, SLIM. Deleting them outright over-corrected: the reference's
   device is frontal so it shows no WALL, but it plainly shows a side — a slim
   grey chassis line the full height and three button slabs proud of the
   silhouette. On a turned state the honest analogue of that is a thin wall, so
   the rails come back at 6% of device width: barely over half the 10.5% of the
   last pass, a third of the 18.5% that read as an iPhone ad, and half the real
   article's 11.5%. Pigment tops out at 109 of 255, the same value as the
   chassis line, so the wall can never out-shine the edge it belongs to. */
.phone::before,
.phone::after {
  content: ""; position: absolute; top: 7.5%; height: 85%;
  width: var(--pt);
  border-radius: calc(var(--pt) * .5);
}
/* RIGHT rail — visible when the state turns the right edge away (rotateY < 0) */
.phone::before {
  right: calc(var(--rim) * -1);
  transform-origin: 100% 50%;
  transform: rotateY(-90deg);
  background:
    /* THE SIDE BUTTON IS A RECESSED SLOT. Measured off the reference: a dark
       groove 11% of the rail wide, INSET from both rail edges, with no bright
       cap — the bright-capped band that was here read as a painted stripe.
       background-size/position is what insets it across the rail's width. */
    linear-gradient(180deg, transparent 0 26%, rgba(0,0,0,.42) 26% 37%,
      rgba(255,255,255,.06) 37% 38.2%, transparent 38.2%),
    /* THE CROSS-SECTION OF A MACHINED WALL, read from the glass edge (0%)
       backward (100%). The reference's rail is NOT a ramp: it is ~38% chamfer
       complex, then a genuinely FLAT plateau across the middle half at one
       constant value, then a short roll-off and a hard cut at the silhouette.
       What was here ramped continuously and put its brightest pixel ON the
       outer edge, which is backwards — a lit cylinder is brightest a quarter
       of the way in and darkest where it turns away. So: chamfer, groove,
       specular at 22%, groove, plateau, roll-off, hard cut. */
    linear-gradient(90deg,
      #5A5E62 0%, #3C4044 7%, #2B2E32 14%,
      #6A6E72 22%, #44484C 30%, #33363A 36%,
      #393C40 40%, #3B3E42 62%,
      #2A2D30 84%, #191B1E 94%, #08090A 100%);
  background-size: 62% 100%, 100% 100%;
  background-position: 24% 0, 0 0;
  background-repeat: no-repeat, repeat;
}
/* LEFT rail — visible when the state turns the left edge away (rotateY > 0) */
.phone::after {
  left: calc(var(--rim) * -1);
  transform-origin: 0 50%;
  transform: rotateY(90deg);
  background:
    /* action button, then the volume pair — recessed slots, same as the right */
    linear-gradient(180deg, transparent 0 20%, rgba(0,0,0,.40) 20% 27%,
      rgba(255,255,255,.06) 27% 28.2%,
      transparent 28.2% 33%, rgba(0,0,0,.40) 33% 44%,
      rgba(255,255,255,.06) 44% 45.2%,
      transparent 45.2% 47%, rgba(0,0,0,.40) 47% 58%,
      rgba(255,255,255,.06) 58% 59.2%, transparent 59.2%),
    /* mirrored: this rail is read from the back edge (0%) to the glass (100%) */
    linear-gradient(90deg,
      #08090A 0%, #191B1E 6%, #2A2D30 16%,
      #3B3E42 38%, #393C40 60%,
      #33363A 64%, #44484C 70%, #6A6E72 78%,
      #2B2E32 86%, #3C4044 93%, #5A5E62 100%);
  background-size: 62% 100%, 100% 100%;
  background-position: 14% 0, 0 0;
  background-repeat: no-repeat, repeat;
}
/* THE DYNAMIC ISLAND. The single strongest "this is an iPhone" cue there is,
   and it costs nothing: every app screen already reserves 58-62px of status-bar
   space at the top (the clone's own padding), so the island lands in room that
   was empty — measured first ink at 68px in ui--reflect and 35-36px in the
   interview screens, against an island foot at 44. It is DEVICE chrome drawn by
   the phone, not app UI: nothing inside .ui moves, and it scales with --ps so it
   holds its proportion at every rung. Pure black with a lens glint at the right,
   under the glass sheen rather than over it. */
.phone__screen::before {
  content: ""; position: absolute; z-index: 2;
  top: calc(10px * var(--ps)); left: 50%;
  width: calc(126px * var(--ps)); height: calc(37px * var(--ps));
  transform: translateX(-50%);
  border-radius: 999px;
  background:
    radial-gradient(circle calc(5px * var(--ps)) at 78% 50%,
      #2C2C3A 0 46%, #15151D 47% 74%, #07070B 75% 92%, rgba(0,0,0,0) 93%),
    #000;
  box-shadow: inset 0 0 0 1px rgba(255,255,255,.045);
}
.phone__screen {
  position: relative;
  width: calc(var(--pw) * 1px);
  height: calc(812 * var(--ps) * 1px);
  border-radius: calc(46px * var(--ps));
  overflow: hidden;
  background: var(--ap-canvas);     /* the app's canvas: #100F18 dark, #F4EFEA light */
  /* the glass sits a hair proud of the bezel */
  /* THE CHAMFER. In the references the single clearest hardware cue on a
     frontal device is the hairline where polished metal meets black glass —
     brighter than either. It is one pixel and it must stay one pixel. */
  box-shadow: inset 0 0 0 1px rgba(255,255,255,.07),
              inset 0 1px 0 rgba(255,255,255,.09);
}
/* THE GLASS. One wide key sheen off the top-left corner and a weak edge return
   along the foot — 10% and 3.5%, which is a lit sheet of glass and not a
   highlight painted onto a picture. It never touches the middle of the screen,
   so nothing in the app loses contrast to it. */
.phone__screen::after {
  content: ""; position: absolute; inset: 0; border-radius: inherit;
  pointer-events: none; z-index: 3;
  background:
    linear-gradient(158deg, rgba(255,255,255,.10) 0%, rgba(255,255,255,.035) 16%,
      rgba(255,255,255,0) 38%),
    linear-gradient(0deg, rgba(255,255,255,.035) 0%, rgba(255,255,255,0) 22%);
}
/* LIGHT THEME: the app screen inside the phone is cream, so a white sheen over it
   is arithmetically invisible and the glass disappeared. What glass over a bright
   screen actually shows in a dark room is the ROOM — so the key stays as a narrow
   specular at the corner and the material is carried by a dark reflection rolling
   off the opposite corner instead. */
html[data-theme="light"] .phone__screen::after {
  background:
    linear-gradient(158deg, rgba(255,255,255,.30) 0%, rgba(255,255,255,.07) 9%,
      rgba(255,255,255,0) 22%),
    linear-gradient(212deg, rgba(16,14,22,.13) 0%, rgba(16,14,22,.045) 17%,
      rgba(16,14,22,0) 40%);
}
.phone__screen > .ui { width: 375px; height: 812px; transform: scale(var(--ps)); transform-origin: 0 0; }

/* STATE 1 — two upright phones, identical size, the second 29px lower, on the
   measured 40/554 separation. Rotation 0.0deg and no perspective: the
   reference's left and right bezel edges hold a constant x across 480px of
   height. This is the composition the section already had, and it is the
   no-JS resting state. */
.stage--1 { gap: var(--dev-gap); }
/* STATE 1 STAYS DEAD FRONTAL. A 4deg turn was tried and removed: at that angle
   the rail still projects INSIDE the silhouette (see the perspective note on
   state 2), so it bought no thickness, and a non-zero rotateY resamples the app
   text — which is the one thing this state exists to let you read. Its
   object-ness comes from the band, the bezel, the island and the shadow, which
   is exactly what a straight-on product shot has. */
.stage--1 .phone { margin-top: var(--dev-t1); }
.stage--1 .phone--drop { margin-top: var(--dev-t2); }

/* STATE 2 — ONE phone, 1.58x wider than either state-1 phone, in perspective,
   running past the stage's foot so the card's own overflow crops it exactly as
   the section edge crops the reference's. rotateY is the honest CSS analogue
   of what the reference renders: it slopes the top edge while leaving the
   side edges vertical, which is precisely the silhouette measured (top edge
   rising 13.0deg to the right, left edge 0.26deg off vertical, right edge
   0.5deg). A 2D rotate cannot do that; a Y-rotation can, because vertical
   lines stay parallel to the image plane. */
/* WHY THE PERSPECTIVE GOT LONGER, NOT SHORTER. A rail is only visible if its far
   edge PROJECTS outside the near edge, and perspective works against that: the
   near edge sits w/2 off the optical axis and is magnified outward by
   P/(P - z), while the rail's far edge is pushed away from the viewer and
   shrinks back toward the axis. Solving x_far*s_far > x_near*s_near for these
   exact geometries gives a floor on P — 723px for state 2 and 820px for state 3
   — and the shipped 420/520px were far below it, which is why 45px of real
   chassis depth rendered zero visible pixels at every size. Lengthening the
   lens is also the photographically correct move: product shots are taken long
   precisely to avoid the near-edge distortion a short lens gives. */
/* STATE 2 KEEPS ITS APPROVED ANGLE AND GETS A LONGER LENS. -16deg is the
   composition and it is not touched. The rail it showed at perspective(1600)
   measured 10.4px of a 250px face; the same 16deg at 3400 measures ~15px,
   because the near edge stops being magnified outward over the far one. This
   is also the photographically correct move — product shots are taken long
   precisely to keep the near edge from swallowing the side. */
/* THE LENS SCALES WITH THE DEVICE. A fixed px perspective is a different lens on
   a 190px mobile device than on a 340px desktop one, and at 4200px it was so long
   the turn stopped reading at all: MEASURED at 390, the top edge sloped 0.40deg —
   a device that is mathematically turned and optically frontal, which is exactly
   "State 2 feels too similar to State 1". Sweeping the lens at a fixed -12deg gave
   4200 -> 0.40deg, 2400 -> 0.81, 1600 -> 1.61, 1200 -> 2.22, 900 -> 3.02, 700 ->
   4.26. 4.6x the device's own width lands every breakpoint on ~3deg: a top edge
   that visibly slopes, a near-side wall that is actually a wall, and nothing like
   the short lens the client rejected as dramatic. */
.stage--2 .phone { --pw: var(--pw2); transform: perspective(calc(var(--pw) * 4.6px)) rotateY(-12deg); }
/* ONE TURN, ONE RAIL. At -16deg only the right wall can face the viewer, but at
   the long lens the LEFT rail's far edge just clears the left silhouette and
   painted a 0.5-1.0px sliver at 1440 and 1512 — a wall on the side that is
   turning away. The state is a fixed single-direction turn, so the other rail
   simply does not exist here. (State 3 keeps both: its two devices turn
   opposite ways and each shows the correct one.) */
.stage--2 .phone::after { content: none; }   /* moot now: no rails anywhere */

/* STATE 3 — two phones tilted in OPPOSITE senses and crossing, the lower-left
   one cropped by the stage's foot. The reference's pair splay apart (one edge
   slope -0.28, the other +0.196) with the upper-right device occluding the
   other's top corner; DOM order puts the right one last so it paints on top. */
.stage--3 { gap: 0; }
.stage--3 .phone--left {
  margin-top: var(--dev-t3);
  margin-right: calc(var(--dev-lap) * -1);
  /* -9.94deg, not -5. The 2D rotate maps 1:1 onto the measured off-vertical
     (5deg shipped read 5.00, 6deg read 5.98), and the reference's pair measure
     9.94 and 13.76 — so the shipped splay was half of it and the two phones
     read as gently offset instead of crossing. */
  --band: 161deg;   /* 155 + the 6 this phone is rolled, so the key stays put in page space */
  transform: perspective(calc(var(--pw) * 4.6px)) rotateY(8deg) rotate(-6deg);
}
.stage--3 .phone--right {
  --band: 147deg;   /* 155 - the 8 this phone is rolled */
  transform: perspective(calc(var(--pw) * 4.6px)) rotateY(-8deg) rotate(8deg);
  z-index: 2;
  box-shadow: var(--drop), -34px 38px 62px -12px rgba(0,0,0,.66);
}

/* THE PAGER — 98 x 46 tray on radius 12 at the CARD's horizontal centre, 40
   above its bottom edge, three items for three rows. Verified against the
   reference frames rather than assumed: at t=0.4 it reads [pill · ·] and at
   t=4.4 it reads [· pill ·], so the active slot IS the accordion index and
   the tray itself never moves or resizes.

   It changes the way the reference changes it: the two affected items animate
   their WIDTH — the leaving one 30 -> 6, the arriving one 6 -> 30 — and
   because the run's total width is invariant the neighbours reflow, which is
   what reads as a pill sliding. Both are mid-width in the same frame there, so
   it is genuinely not a hard swap and not a separate sliding element. The
   width runs on the ROWS' clock, not the stage's: the reference's no-bounce
   row spring (critically damped, w = 16/s, 650ms, sampled into linear()),
   starting on the click with no delay. The colour is asymmetric like the
   reference's: the arriving pill lights in 367ms, the leaving one dims over
   660ms, both on the in-out curve. Below 760 both run 600ms in-out
   (§ THE MOBILE DEMO).

   IT IS NOT INTERACTIVE AND IS NOT MADE SO. In 261 frames the cursor never
   enters the tray, never comes within ~150px of it, and no state change ever
   originates there; there is no hover pixel change on it either. That is zero
   evidence of clickability, so it stays a passive indicator and stays
   aria-hidden — the three buttons above already carry the state accessibly. */
.dev__pager {
  display: flex; align-items: center; justify-content: center; gap: 8px;
  width: 98px; height: 46px;
  margin-top: 40px;
  border-radius: var(--r-pill);
  background: rgba(169,155,255,.06);
}
.dev__pager i {
  display: block; height: 6px; width: 6px; border-radius: 49px;
  background: rgba(169,155,255,.22);
  /* the LEAVING pill's clock (the after-change style picks it) */
  transition: width .65s var(--rows-nobounce),
              background-color .66s var(--e2);
}
[data-orb-state="1"] .dev__pager i:nth-child(1),
[data-orb-state="2"] .dev__pager i:nth-child(2),
[data-orb-state="3"] .dev__pager i:nth-child(3) {
  width: 30px; background: var(--deep-fg);
  transition: width .65s var(--rows-nobounce),
              background-color .367s var(--e2);
}
/* an index for a control that cannot be operated is noise, and with every
   panel open there is no single index to show */
[data-orb]:not([data-orb-live]) .dev__pager { display: none; }

/* ---- the whole-screen layouts inside the phones — real app spacing -------- */
/* THE RAIL GAP WAS INVERTED. InterviewShell puts 14 ABOVE the track (the
   header row's margin-bottom, headerRowGap) and 22 BELOW it (trackGap). */
.ui--interview { padding: 58px 22px 30px; }
.ui--interview .ui__rail { margin-top: 14px; margin-bottom: 22px; }
.ui--interview .ui__centre { flex: 1; display: flex; flex-direction: column; align-items: center; justify-content: center; gap: 22px; padding-bottom: 72px; }
.ui--interview .ui__card { margin-top: auto; }
/* the reflection screen: head gap 12 (Banner -> heading), cards block 16 */
.ui--reflect { padding: 62px 24px 30px; }
.ui--reflect .ui__serif { margin-top: 12px; }
.ui--reflect .ui__stack { margin-top: 16px; }
/* InsightCard's text column is gap 5, not the 8 the story cards use. Scoped,
   so #story's own .ui--know geometry is untouched. */
.ui--reflect .ui__ibody { margin-top: 5px; }

/* ---- atoms that only the demo panel's screens use ------------------------ */
/* Banner, tone="info": a plain raised surface, because information is not an
   accent. c.surface fill, c.hairline border, radius input 14, 12/16 padding,
   600/13 on 19 leading in inkSecondary. */
.ui__banner {
  border: 1px solid var(--ap-rule); border-radius: 14px;
  background: var(--ap-surface); padding: 12px 16px;
  font-weight: 600; font-size: 13px; line-height: 19px; color: var(--ap-ink2);
}
/* InterviewTranscriptDrawer — a bottom sheet over the question screen */
.ui--sheet .ui__scrim { position: absolute; inset: 0; background: var(--ap-veil); }
.ui__drawer {
  position: absolute; left: 0; right: 0; bottom: 0; max-height: 85%;
  border-radius: 24px 24px 0 0;
  border-top: 1px solid var(--ap-rule);
  background: var(--ap-floating);
  padding: 16px 24px 24px;
  overflow: hidden;
}
.ui__dhead   { display: flex; align-items: center; justify-content: space-between; }
.ui__dtitle  { font-weight: 800; font-size: 24px; line-height: 1.2; letter-spacing: -.48px; color: var(--ap-ink); }
.ui__dclose  { width: 44px; height: 44px; margin-right: -10px; display: grid; place-items: center; color: var(--ap-ink2); }
.ui__dclose svg { width: 20px; height: 20px; }
.ui__turn    { border-top: 1px solid var(--ap-rule); padding: 12px 0; }
.ui__dlabel  { display: block; font-weight: 800; font-size: 11px; line-height: 1; letter-spacing: 1.3px; text-transform: uppercase; color: var(--ap-ink2); }
.ui__dlabel--sp { margin-top: 10px; }
.ui__dbody   { font-weight: 600; font-size: 15px; line-height: 22px; color: var(--ap-ink); margin-top: 2px; }
.ui__dacts,
.ui__composer { display: grid; gap: 8px; padding-top: 16px; }
.ui__dh2     { font-weight: 800; font-size: 15px; line-height: 1.2; color: var(--ap-ink); }
.ui__dnote   { font-weight: 600; font-size: 13px; line-height: 22px; color: var(--ap-ink2); }
.ui__dnote--dim { color: var(--ap-muted); }
.ui__field {
  min-height: 88px; border-radius: 14px;
  border: 1px solid var(--ap-rule); background: var(--ap-surface); padding: 16px;
  font-weight: 600; font-size: 15px; line-height: 22px; color: var(--ap-ink);
}
.ui__btnp, .ui__btng {
  min-height: 52px; display: flex; align-items: center; justify-content: center;
  padding: 0 16px; border-radius: 14px;
  font-weight: 800; font-size: 15px; line-height: 1.2; text-align: center;
}
.ui__btnp { background: var(--ap-coral); color: var(--ap-on-coral); }
.ui__btng { background: var(--ap-surface2); border: 1px solid var(--ap-rule); color: var(--ap-ink); }

@media (min-width: 1000px) {
  .orb__card {
    padding: 40px clamp(40px, 6.94vw, 100px);
    /* 446 text / 100 gutter / 560 device. The device track is the measured
       cluster span (261 + 40 + 259) and nothing more, so the 100px the old
       608-wide track was holding for an overlap goes to the text column —
       the reference's own text column is 337-397 wide. */
    grid-template-columns: minmax(0, 446fr) minmax(0, 560fr);
    column-gap: clamp(40px, 6.94vw, 100px);
    align-items: center;
  }
  /* the tray floats at the card's centre, 40 above its bottom edge, and above
     the stage — the reference's sits over its devices too */
  .dev__pager { position: absolute; left: 50%; bottom: 40px; transform: translateX(-50%); margin-top: 0; z-index: 3; }

  /* ═══ THE FRAME NOW FITS THE OBJECT ═════════════════════════════════════
     WHAT WAS WRONG: the stage box was sized from the reference's FLAT device
     art, and the realism pass then turned the phones and lengthened the lens —
     which makes a projected silhouette much taller than the element it is laid
     out in. Measured at 1440: state 2's phone projected 862.5px into a 620px
     box and lost 195.4px (22.7% of the device) off the card's foot, with its
     top edge only 7px from the card's edge — cut at BOTH ends, which is what
     reads as "not fully rendered" rather than as a crop. State 3's lower-left
     device lost another 89.7px, straight through the coral Save button.
     THE RULE NOW: the stage's height equals the card's content height at every
     rung (see the rungs below), every device is fully inside the frame, and the
     three compositions are CENTRED in it rather than top-anchored:
       · state 2 is one object, so it centres on both axes.
       · state 3 keeps flex-start — its two phones are staggered on purpose, and
         align-items:center would centre them individually and flatten the
         crossing (measured: the 190px drop collapses to ~100). It is seated
         with --s3-top instead, which moves the pair without touching the pose.
       · state 1 is centred by its own --dev-t1/--dev-t2 at each rung, so the
         29px stagger between the two phones survives exactly.
     Nothing here changes an angle, a perspective, a rail or a shadow. */
  .stage--2 { align-items: center; }
  .stage--2, .stage--3 { justify-content: center; }
  .stage--3 { padding-top: var(--s3-top, 0px); }
}

/* THE COMPOSITION IS SCALED AS ONE UNIT AT THREE RUNGS: every phone width,
   the separation, the drops and the overlap take the same factor (the
   entrance offset does not — it is motion, and the reference's 70px holds at
   every desktop width), which keeps every ratio the frame gave — 40/554 separation,
   23/620 and 52/620 drops — at every width. Only the top rung is the literal
   measurement.

   THE RUNGS ARE min-width ONLY, ASCENDING, AND THAT IS DELIBERATE. Written as
   (min-width:1000px) and (max-width:1199px) they left an uncovered band, and
   it was not theoretical: at a browser zoom of 110% the media width is
   fractional — 1199.09 at a 1199px viewport — so max-width:1199px is false and
   min-width:1200px is also false, and the phones dropped back to the base
   sizes with no drop at all. Measured live at exactly 1199 and 1399 before
   this was written as three overlapping min-width rungs that simply override
   each other. */
/* THE STAGE HEIGHT IS THE CARD'S CONTENT HEIGHT. `.orb__card` has a 700px
   min-height and 40px of vertical padding, so a stage of H makes the card
   max(700, H + 80) tall and the stage then owns the whole interior — which is
   what lets every composition be centred in the frame the reader actually
   sees. 620 / 660 / 760 are the smallest heights at which all three states fit
   with air: state 2's projected phone is 1.39x the nominal box at every rung,
   and state 3's pair spans 1.24x. --pw2 comes down with them (380 -> 326 at
   the top rung, 13% smaller) because that is the only lever that shortens the
   projection without touching the approved -16deg turn. --pw1 is UNCHANGED at
   every rung: states 1 and 3 keep the device size they were approved at. */
@media (min-width: 1000px) {                                  /* factor 0.68 */
  .dev {
    height: 620px;
    --pw1: 163; --pw2: 258;
    --dev-gap: 27px; --dev-t1: 118px; --dev-t2: 137px;  /* centres the pair in 620 */
    --dev-t3: 128px; --dev-lap: 38px;
    --s3-top: 64px;
  }
}
@media (min-width: 1200px) {                                  /* factor 0.85 */
  .dev {
    height: 660px;
    --pw1: 205; --pw2: 291;
    --dev-gap: 34px; --dev-t1: 90px; --dev-t2: 114px;
    --dev-t3: 160px; --dev-lap: 48px;
    --s3-top: 21px;
  }
}
@media (min-width: 1400px) {                 /* THE MEASURED ONE. factor 1.00 */
  .dev {
    height: 760px;
    --pw1: 241; --pw2: 326;
    --dev-gap: 40px; --dev-t1: 96px; --dev-t2: 125px;
    --dev-t3: 190px; --dev-lap: 56px;
    --s3-top: 16px;
  }
}
@media (max-width: 999px) {
  .orb__card { min-height: 0; padding: 24px; margin-top: 40px; }
}
/* (these two rungs now serve only the script-less stacked layout below 760;
   the live mobile demo sizes itself in § THE MOBILE DEMO) */
@media (max-width: 419px) {
  .dev {
    height: 372px;
    --pw1: 134; --pw2: 210;
    --dev-gap: 14px; --dev-t1: 16px; --dev-t2: 36px;
    --dev-t3: 112px; --dev-lap: 36px;
  }
}
@media (max-width: 351px) {
  .dev {
    height: 320px;
    --pw1: 112; --pw2: 176;
    --dev-gap: 12px; --dev-t1: 14px; --dev-t2: 30px;
    --dev-t3: 96px; --dev-lap: 32px;
  }
}

/* ═══════════════ THE MOBILE DEMO — ONE CINEMATIC STAGE (≤759px) ════════════
   Below 760 the demo stops being the desktop card stacked in a column and
   becomes one immersive dark chapter: THE SECTION IS THE STAGE. The plate
   keeps its box for layout only — it is full-bleed, has no fill, no hairline
   and no radius, so nothing reads as "a card inside a section inside a page".
   What is left is a lit ground, one device floating in it, and one glass
   caption + control dock over the device's foot.
     · THE GROUND. Two very soft radial blooms — plum high behind the device,
       teal low — on the section's own --deep. They exist to separate the
       device from a near-black field, nothing more: peak alpha .11 / .06.
     · THE DEVICE IS THE HERO, BUT IT IS A DEVICE — 66-68% of the viewport
       width, not 77%. At 77% the silhouette ran off both edges and state 1
       read as a full-bleed screenshot of an app rather than a phone standing
       in a dark room; at 67% both device edges, the corner radius and the
       foot are in frame on every rung (61-71px of ground each side), the
       screen's own controls stop being cut by the caption, and the device
       still runs down past the plate's foot, where the mask dissolves it
       behind the glass instead of cutting it. Each state gets its OWN
       composition, not the same pose re-filled.
     · THE DOCK. The caption card and the stepper share one width, one radius
       and one material, 6px apart, so they read as one control system in two
       parts. The stepper's bar is the card's ::after — a grid item, so no
       markup is duplicated for it — and the keys and pager sit on it with no
       fill of their own.
   MOTION (the reference's phone showcase). The leaving layer is cut in one
   frame; the arriving one fades in and slides 15px from the right, whichever
   key was pressed, over 700ms in-out; the caption text swaps in one frame
   and the pager eases over 600ms. Nothing bounces, nothing spins, and under
   prefers-reduced-motion every one of them lands at once.
   Everything is gated on [data-orb-live]: without the runtime there is no
   selection, so a script-less reader keeps the stacked list. */
.orb__step { display: none; background: rgba(169,155,255,.06); }
@media (max-width: 759.98px) {
  .orb__card[data-orb-live] {
    /* full-bleed: the plate spans the viewport whatever the wrap's padding is */
    margin-inline: calc(50% - 50vw);
    /* THE STAGE IS SIZED FROM THE SPACE IT IS ALLOWED TO HAVE, not from a fixed
       number. 132 = 92 above + 40 below, so when the stage is framed (its foot
       40px off the viewport floor) its head lands at a CONSTANT 92px on every
       phone — 28px of air under the floating nav's 64px band — and the dock's
       foot at a constant 60px above it. svh is the toolbar-VISIBLE height, so
       the geometry is identical whether Safari's bottom bar is expanded or
       collapsed; collapsing it only adds room. The old `100svh - 48` capped at
       700 made the stage 93% of a real iPhone's safe viewport: nothing was left
       for the header above it and the dock sat in the browser chrome. */
    height: clamp(520px, calc(100svh - 132px), 820px);
    /* the dock's own inset from the screen edges; the device layer ignores it.
       +8 over the inset keeps the keys off the home indicator when Safari's bar
       collapses and the inset becomes non-zero. */
    padding: 0 16px max(24px, calc(env(safe-area-inset-bottom) + 8px));
    border-radius: 0;
    /* THE CONTAINER DISAPPEARS: no fill, no hairline — only the lighting */
    /* THE LIGHT MUST DIE BEFORE THE BOX DOES. Each bloom's visible radius is
       its extent times its zero-stop, so bloom 1 reached 20% - 34%*.72 = -4.5%
       and bloom 3 reached 94% + 18%*.76 = 107.7%: both were CUT by the plate's
       own edges, and a cut gradient on a flat ground is a hairline. Measured at
       390x844 the top cut was a single-row luminance step of .00130 on a .00518
       field — a 25% jump, which is exactly the "rectangular stage boundary" you
       could still see. Re-centred to 26% and 86% the blooms reach 1.5% and
       99.7%, so the light fades out INSIDE the plate and the plate's edges paint
       nothing at all. The section is the stage because the box is now invisible,
       not because the box was removed. */
    background:
      radial-gradient(64% 34% at 12% 26%, rgba(169,155,255,.12), rgba(169,155,255,0) 72%),
      radial-gradient(58% 32% at 90% 66%, rgba(75,182,168,.08), rgba(75,182,168,0) 70%),
      radial-gradient(70% 18% at 50% 86%, rgba(169,155,255,.10), rgba(169,155,255,0) 76%);
    box-shadow: none;
    grid-template-columns: auto 1fr auto;
    grid-template-rows: 1fr auto auto;
    grid-template-areas: ". . ." "cap cap cap" "prev pager next";
    gap: 6px 0;
    align-items: end;
  }
  /* THE DOCK'S BAR — the card's own ::after, placed in the stepper row and
     spanning it, so the two keys and the pager sit ON one surface instead of
     being three separate pills. Same material as the caption, a little
     thinner, so they read as one system in two parts. */
  .orb__card[data-orb-live]::after {
    content: ""; grid-row: 3; grid-column: 1 / -1; align-self: stretch;
    border-radius: 22px;
    background: color-mix(in srgb, var(--deep-2) 34%, transparent);
    -webkit-backdrop-filter: blur(20px) saturate(155%) contrast(104%);
            backdrop-filter: blur(20px) saturate(155%) contrast(104%);
    box-shadow: inset 0 1px 0 rgba(255,255,255,.08), inset 0 0 0 1px rgba(255,255,255,.05);
  }

  /* the device layer fills the plate edge to edge behind the dock and
     dissolves just above it, so the phone keeps its full brightness through
     the glass and never shows a cut line under the controls */
  .orb__card[data-orb-live] .dev {
    position: absolute; inset: 0; z-index: 0; height: auto;
    --pw1: 250; --pw2: 268;
    /* STATE 3 IS SIZED ON ITS OWN. States 1 and 2 are height-limited (one
       device in a tall stage) and state 3 is WIDTH-limited (two devices across
       a 390px screen), so one number cannot serve both: at the shared size the
       pair measured 460px wide on a 390px screen — the front phone 10px off the
       left edge, the back one 85px off the right, no ground on either side.
       --pw3 / --lap3 / --top3 are the state-3 composition, scaled per rung. */
    --pw3: 174; --lap3: 61px; --top3: 20px;
    --dev-gap: 0px; --dev-t1: 0px; --dev-t2: 0px; --dev-t3: 0px;
    -webkit-mask-image: linear-gradient(#000 calc(100% - 150px), rgba(0,0,0,.62) calc(100% - 110px), rgba(0,0,0,.24) calc(100% - 74px), transparent calc(100% - 40px));
            mask-image: linear-gradient(#000 calc(100% - 150px), rgba(0,0,0,.62) calc(100% - 110px), rgba(0,0,0,.24) calc(100% - 74px), transparent calc(100% - 40px));
  }
  /* NO MOBILE SHADOW OVERRIDE. What was here painted a 28% LAVENDER inset ring
     straight over the titanium band and a 120px lavender bloom behind the
     device — written for the old flat frame, where that ring was the only thing
     standing in for a bezel edge. At (0,3,0) it also outranked .phone's own
     three-layer drop, so the contact/body/ambient shadow never reached mobile at
     all. The device now inherits one drop everywhere, and the only colour near
     it is the stage's own lighting. */

  /* THE THREE COMPOSITIONS. Same device, three poses: frontal, angled and
     raised, then the crossing pair. The stage-level pose is the motion (see
     below); the per-state geometry is these three rules. */
  .orb__card[data-orb-live] .stage {
    justify-content: center;
    /* 56, NOT 28. Pulling the device back to 67% shortened it, and at 28 its
       foot stopped ABOVE the glass: the caption went from sitting on a lit
       screen to sitting on empty ground, and its measured environmental spread
       collapsed .090 -> .011 — the exact "flat grey card" the material exists
       to avoid. Seating the device 56 down puts 82% of the caption back over
       the screen. Measured at 390x844 dark, sweeping 28/44/56/68/80 gave spread
       .011/.091/.162/.109/.089 at contrast 10.4/10.3/10.3/10.2/10.0, so 56 is
       the peak of the light AND still 1.1 points of contrast above the
       previous candidate's 9.2. The air it adds above the device is what makes
       state 1 the calmest of the three. */
    /* 22, NOT 56. The 56 seat was chosen to push the device's foot far enough
       behind the caption glass to light it — and it worked, but it buried 90px
       (state 1), 92px (state 2) and 119px (state 3) of the device under the
       dock, taking the Stop key, the last insight card and the Save bar with
       it. That is the "text card covers too much" complaint, and it is the
       device's position, not the caption's size. At 22 the device rises 34px
       and the overlap falls to 20-40px — still the device's own bezel foot
       under the glass, so the material keeps the screen's light, but every
       control the screen is showing is now above it. */
    padding-top: 22px;
    /* THE PHONE ENTRY POSE: 15px to the right and nothing else — no scale, and
       the same pose for both keys. Opacity 0 and the missing transition come
       from the shared idle rule, so the leaver is still cut in one frame. */
    translate: 15px 0;
    scale: none;
  }
  .orb__card[data-orb-live][data-orb-state="1"] .stage--1,
  .orb__card[data-orb-live][data-orb-state="2"] .stage--2,
  .orb__card[data-orb-live][data-orb-state="3"] .stage--3 {
    translate: none;
    transition: opacity   .7s var(--e2),
                translate .7s var(--e2);
  }
  /* STATE 1 — frontal and calm: the listening screen alone, square to the
     viewer, the question and the orb sitting in the upper half of the frame.
     (The idle twin is desktop's second voice and stays out of this one.) */
  .orb__card[data-orb-live] .stage--1 .phone:not(.phone--drop) { display: none; }
  /* THE DEVICE IS TRIMMED, NOT SHRUNK. Each state gives back a few per cent so
     that what is left is COMPLETELY in frame: measured across 375/390/393/402/
     430/440 the single device now reads 63-71% of the viewport (was 67-70% with
     its foot cut), its top edge clears the plate by 25-27px on every rung, and
     the horizontal ground is symmetric. */
  /* STATE 1 IS THE COMPLETE DEVICE. At .95 the phone stood 526px tall in the
     524px the plate gives above the caption, so its foot was always under the
     glass and it read as a magnified crop of the app rather than as a phone.
     .86 puts the WHOLE device in frame — top edge 25px inside the plate, foot
     15-53px clear of the caption on every rung — at 57-64% of the viewport,
     which is still the largest single object on the screen. */
  /* THE PLATE IS 100svh - 132px AND ON iOS svh IS THE TOOLBAR-SHOWN VIEWPORT.
     Headless Chrome resolves svh to the full window, so every previous fit was
     made against a plate ~100px taller than the device gives: simulated at the
     real small viewports the state-1 device ran 27-66px behind the caption
     glass, state 2 39-77px, and state 3's front phone overflowed the plate's
     TOP by 7-71px. That is the crop. All three states are re-sized against the
     small plate, and state 1 — the calm, complete, frontal one — is the state
     that clears the glass entirely. */
  /* SEATED FROM THE DOCK, NOT FROM THE TOP. A fixed top seat cannot guarantee the
     foot clears the caption, because the room above the dock varies with the
     plate: measured at the real Safari viewports the device had 18 / 13 / 21 /
     23 / 37 / 59px of slack at 375 / 390 / 393 / 402 / 430 / 440, so a 22px seat
     ran 9px behind the glass at 390 and left 15px of dead air at 430. Anchoring
     the foot 5px above the dock makes the clearance CONSTANT at every size and
     hands all the remaining slack to the top, which is where the breathing room
     belongs. Scale is untouched — this is seating only. */
  .orb__card[data-orb-live] .stage--1 {
    align-items: flex-end;
    padding-top: 0;
    padding-bottom: calc(193px + 5px);
  }
  /* STATE 2 IS A DIFFERENT CHAPTER, NOT THE SAME PHOTOGRAPH. State 1 is the
     calm one: centred, frontal, seated 22 down, its whole foot clear of the
     caption. State 2 is the presentation: WIDER than state 1, turned -11deg so
     its slim side shows, seated high at 8, and the one state whose foot is
     deliberately allowed to run behind the caption glass — which is also what
     lights the glass. Two different framings of the same device, which is what
     "distinct chapter" means when the section only owns one object. */
  /* SAME SEATING AS STATE 1, SO THE TWO CHAPTERS SHARE A BASELINE. State 2 is the
     shorter device, so anchoring both feet 5px above the dock leaves it more air
     above than state 1 — which is exactly the "pulled back and turned" reading
     the 40px top seat was buying, now held at every viewport instead of one. */
  .orb__card[data-orb-live] .stage--2 {
    align-items: flex-end;
    padding-top: 0;
    /* 6px more than state 1: flex-end aligns the LAYOUT box, and state 2's turn
       projects its lowest corner ~6px below that box, so the same 5px of padding
       measured 0 to -0 of real clearance at 402-440. */
    padding-bottom: calc(193px + 11px);
  }
  /* THE NEAR-SIDE WALL, DRAWN FLAT. Same width the 3D rail measured at this size
     (2 CSS px at 390), same pigment, and it inherits the phone's own turn, so it
     still leans with the device. It sits OUTSIDE the chassis line, which is what
     makes the edge read as a side rather than a thicker line. */
  .orb__card[data-orb-live] .stage--2 .phone::before {
    top: 6.5%; height: 87%;
    width: max(1.5px, calc(var(--pw) * 0.011px));
    right: calc(var(--rim) * -1 - max(1.5px, calc(var(--pw) * 0.011px)) + 0.5px);
    transform: none;
    border-radius: 0 2px 2px 0;
    background:
      linear-gradient(180deg, transparent 0 26%, rgba(0,0,0,.42) 26% 37%,
        rgba(255,255,255,.06) 37% 38.2%, transparent 38.2%),
      linear-gradient(90deg, #4E5256 0%, #3B3E42 55%, #191B1E 100%);
    background-size: 100% 100%, 100% 100%;
    background-position: 0 0, 0 0;
  }
  .orb__card[data-orb-live] .stage--1 .phone { --pw: calc(var(--pw1) * .73); }
  .orb__card[data-orb-live] .stage--2 .phone { --pw: calc(var(--pw2) * .63); }
  /* STATE 2 — the device steps forward and turns. It stays the closest of the
     three (measured 70-72% of the viewport against state 1's 66-68%, so it
     still reads as the more immersive state) but it no longer crops its own
     top away: raised 6px instead of 44, its top edge lands 4-9px inside the
     stage on every rung, so the top bezel and both upper corners are in frame
     and the pose reads as a phone held at an angle rather than as a screenshot
     enlarged past its own silhouette. The right edge keeps 42-50px of ground
     (was 22). Same rotation, same 420px perspective. */
  /* THE SCREEN STAYS CRISP. transform-style: preserve-3d renders the whole phone
     into a 3D context and rasterises it as ONE texture, which is then mapped by
     the turn — so the app screen, already scaled by .ui's own transform, gets
     resampled. MEASURED on the state-2 card copy at 390: Laplacian variance 144
     under preserve-3d against 1326 flat, and 191 vs 1067 at 430; peak white 161
     vs 195. Layer hints do nothing (translateZ, will-change and
     backface-visibility all measured 144). Desktop is unaffected — its --ps is
     0.85 against mobile's 0.47, so the same texture carries twice the pixels and
     measures 2110 — which is why this override is mobile-scoped. Flattening
     collapses the rotateY(-90deg) rail, so the near-side wall is redrawn below
     as a 2D strip at the width the 3D one actually rendered. */
  .orb__card[data-orb-live] .stage--2 .phone {
    transform-style: flat;
    margin-top: 0;   /* the 22px seat is the whole offset now; -34 against it clipped the top */
    /* the -12deg turn throws the projected silhouette 2.5px right of the row it is
       laid out in; 5px of margin puts it back on the caption's own centre line
       (measured gutters were 103/98 at 375 and 100/94 at 430). */
    margin-right: 5px;
    /* THE 16px CORRECTION AND THE 1.5deg ROLL ARE BOTH GONE. Both were written
       for the -18deg turn: the turn threw the projected silhouette right of the
       row it was laid out in (gutters 64/42 ... 76/49), and 16px of margin put
       it back on the centre line. At -5deg the throw is gone and that same 16px
       was pushing the device 7.8px LEFT of a caption and dock that are exactly
       symmetric underneath it — measured at all five widths. The roll went with
       it because desktop state 2 has none: the -5deg lean is now the whole
       difference from state 1, and it is the SAME lean at every breakpoint. */
    /* NO scale() HERE, AND NO LOCAL TRANSFORM. A fractional scale on top of
       .phone__screen > .ui's own transform: scale(var(--ps)) rasterises the app
       screen once and then resamples it: MEASURED, state 2's in-screen text came
       back at a Laplacian variance of ~350 against state 1's ~2500, so the one
       thing that has to be the hero was the softest thing in the section. The
       .96 is folded into --pw, and the state inherits the shared lens and turn
       from .stage--2 .phone so mobile and desktop are posed identically. */
  }
  /* STATE 3 — the layered one: the composer ("Change this answer") leads and
     bleeds off the left, the saved-answers list crosses behind it, dropped far
     enough that only its header reads before it leaves the right edge. */
  /* on mobile the LEFT device is the front one, so it is the one that casts */
  /* padding-right, not a margin: rotateY throws BOTH phones' projected boxes to
     the right of the row they are laid out in, so a centred row is not a centred
     picture — measured gutters were 42 left / -12 right. Half of 50 brings the
     pair back onto the screen's own centre line and puts the back device fully
     inside the frame. */
  /* BOTTOM-ANCHORED, NOT TOP-SEATED. The plate is 100svh - 132px, so its height
     grows faster than the composition does: a fixed top seat left the back device
     3-11px off the caption glass at 359-402 and 67-88px at 430-440 — the same
     state reading crowded on one phone and stranded on the next. Anchoring the
     pair to the dock makes that gap a CONSTANT 38px everywhere. 193 is the dock
     (caption + pager) below 420 where the caption runs three lines; 173 above it
     where it runs two. --top3 is now purely the stagger. */
  .orb__card[data-orb-live] .stage--3 {
    align-items: flex-end;
    padding-top: 0;
    padding-bottom: calc(193px + 34px);
    padding-right: 50px;
  }
  .orb__card[data-orb-live] .stage--3 .phone { --pw: calc(var(--pw3) * .88); }
  /* SAME FLATTENING AS STATE 2, SAME REASON. Measured on the front device's
     "Your answers" copy at 390: Laplacian variance 263 under preserve-3d against
     1687 flat (304 -> 1481 at 430). The two devices keep their turn and their
     roll — transform-style only governs how the CHILDREN are placed — and each
     one's near-side wall is redrawn flat, on the side its own yaw actually
     turns toward the viewer. */
  .orb__card[data-orb-live] .stage--3 .phone { transform-style: flat; }
  .orb__card[data-orb-live] .stage--3 .phone::before,
  .orb__card[data-orb-live] .stage--3 .phone::after {
    top: 7.5%; height: 85%;
    width: max(1.2px, calc(var(--pw) * 0.009px));
    transform: none;
    background: linear-gradient(90deg, #4A4E52 0%, #35383C 55%, #16181B 100%);
  }
  /* the front device is turned +8deg: its LEFT side faces the viewer */
  .orb__card[data-orb-live] .stage--3 .phone--left::before { content: none; }
  .orb__card[data-orb-live] .stage--3 .phone--left::after {
    left: calc(var(--rim) * -1 - max(1.2px, calc(var(--pw) * 0.009px)) + 0.5px);
    border-radius: 2px 0 0 2px;
    background: linear-gradient(90deg, #16181B 0%, #35383C 45%, #4A4E52 100%);
  }
  /* the back device is turned -8deg: its RIGHT side faces the viewer */
  .orb__card[data-orb-live] .stage--3 .phone--right::after { content: none; }
  .orb__card[data-orb-live] .stage--3 .phone--right::before {
    right: calc(var(--rim) * -1 - max(1.2px, calc(var(--pw) * 0.009px)) + 0.5px);
    border-radius: 0 2px 2px 0;
  }
  /* THE FRONT DEVICE IS WHOLE, THE BACK ONE IS DELIBERATELY CROPPED. Measured
     at 390: the front phone now stands 13px inside the left edge with its
     silhouette, rail, buttons and the whole Save/Cancel stack in frame, and its
     foot lands on the caption rather than 119px under it; the back phone keeps
     its top, its island and 80% of its width and leaves through the right edge
     by ~50px. One object reads as complete and the other as the layer behind
     it — which is a composition, where two half-devices with no ground on
     either side was not. */
  .orb__card[data-orb-live] .stage--3 .phone--left  {
    margin-left: calc(var(--pw3) * .26px);
    margin-right: calc(var(--lap3) * -1);
    z-index: 3;
    box-shadow: var(--drop), 30px 34px 56px -12px rgba(0,0,0,.64);
  }
  .orb__card[data-orb-live] .stage--3 .phone--left { margin-bottom: var(--top3); }

  /* THE CAPTION — smoked glass, not a UI card: the fill is 34% of the panel
     colour and the blur 20px, so the device behind it genuinely shows through
     — measured, the luminance spread across the caption nearly doubles in
     states 1 and 3 (.046 -> .090, .050 -> .087), which is the phone's own light
     reaching the material. It does NOT get brighter doing it: the plate reads
     slightly darker and the title/body contrast rises in all three states
     (8.9/9.4/8.5 -> 9.2/9.5/8.9). Saturate and a 4% contrast lift keep the
     colour the device throws from going milky. The only
     edges are a 1px top highlight and a 1px inner ring, and the drop is soft
     enough to read as lift rather than as a border. All three rows share one
     grid cell, so the dock never moves between states. */
  .orb__card[data-orb-live] .orb__rows {
    grid-area: cap; position: relative; z-index: 1;
    display: grid; max-width: none; align-self: end;
    border-radius: 22px;
    background: color-mix(in srgb, var(--deep-2) 34%, transparent);
    -webkit-backdrop-filter: blur(20px) saturate(155%) contrast(104%);
            backdrop-filter: blur(20px) saturate(155%) contrast(104%);
    box-shadow:
      inset 0 1px 0 rgba(255,255,255,.08),
      inset 0 0 0 1px rgba(255,255,255,.05),
      0 22px 44px -30px rgba(0,0,0,.55);
  }
  /* LIGHT THEME PAYS FOR THE SHOW-THROUGH. The stage stays dark in light, but
     the app UI inside the phone does not — so the same 34% fill that lets the
     dark theme's device through also lets a bright screen through. A brightness
     trim on the BACKDROP (not the fill) takes the screen's brightness out
     without closing the material. The trim went .84 -> .72 with this pass: once
     the device was seated 56 down, far more bright screen sits behind the glass
     and .84 left state 3 (the white composer and the Save bar) at 3.96, WORSE
     than the 4.41 it used to have. Swept .84/.78/.72/.66 the three states read
     4.99-6.52-3.96 / 5.39-6.87-4.30 / 5.83-7.20-4.67 / 6.25-7.57-5.09: .72 is
     the first rung where every state beats the previous candidate (4.11 / 4.16
     / 4.41) and the worst case clears 4.5, and it still keeps .13-.16 of
     luminance spread — the device's own light — in the material. */
  /* THE RECTANGLE. `html[data-theme="light"] .orb__card` carries an inset
     1px ring so the DESKTOP panel keeps an edge on a light page — and at
     (0,2,1) it outranked the mobile plate's own `box-shadow: none` at (0,2,0).
     On desktop the plate is a rounded card and the ring is right; here the
     plate is FULL-BLEED with radius 0, so the same ring painted a hairline
     across the top of the stage and straight down both screen edges. That is
     the "rectangular stage boundary" — a real border, not an illusion. Mobile
     light gets it taken back off; desktop is untouched. */
  html[data-theme="light"] .orb__card[data-orb-live] { box-shadow: none; }

  html[data-theme="light"] .orb__card[data-orb-live] .orb__rows,
  html[data-theme="light"] .orb__card[data-orb-live]::after {
    background: color-mix(in srgb, var(--deep-2) 46%, transparent);
    -webkit-backdrop-filter: blur(20px) saturate(155%) contrast(104%) brightness(.64);
            backdrop-filter: blur(20px) saturate(155%) contrast(104%) brightness(.64);
  }
  /* AND THE SUPPORTING LINE IS LIFTED, in light only. Measured properly — the
     ink hidden, the 99th-percentile backdrop luminance sampled inside the text's
     own bounding box — the caption's 15px body copy sat at 2.80 / 2.95 / 2.35
     against the brightest patch beneath it, i.e. below AA in all three states,
     because in light the device behind the glass is a WHITE app screen. Closing
     that on the material alone would have meant a ~60% fill, which is the flat
     grey card this plate exists not to be. 46% + .64 plus one step of ink does
     it while the material keeps .08-.09 of luminance spread. */
  html[data-theme="light"] .orb__card[data-orb-live] .orb__panelin > p { color: #E4E4E0; }
  /* THE CAPTION SWAPS, IT DOES NOT CROSSFADE — the reference changes its text
     between two frames. Visibility has no transition, so the outgoing row is
     gone and the incoming one is there on the frame of the tap. The rows share
     one grid cell, so the card's height is the tallest caption's in every
     state and there is no height change to animate. */
  .orb__card[data-orb-live] .orb__row {
    grid-area: 1 / 1; display: grid; align-content: center;
    padding: 13px 18px; visibility: hidden;   /* type untouched */
    background: none; border-radius: 0;
  }
  .orb__card[data-orb-live][data-orb-state="1"] .orb__row:nth-child(1),
  .orb__card[data-orb-live][data-orb-state="2"] .orb__row:nth-child(2),
  .orb__card[data-orb-live][data-orb-state="3"] .orb__row:nth-child(3) { visibility: visible; }
  .orb__card[data-orb-live] .orb__btn { padding: 0; cursor: default; }
  .orb__card[data-orb-live] .orb__ico { display: none; }
  .orb__card[data-orb-live] .orb__row .t-q { font-size: 16px; line-height: 1.25; }
  .orb__card[data-orb-live] .orb__panelin > p { padding: 0; margin-top: 4px; font-size: 15px; line-height: 1.35; }

  /* THE KEYS AND THE PAGER — no fill of their own now: the dock's bar behind
     them is the material, and these are just the marks on it. */
  .orb__card[data-orb-live] .dev__pager {
    grid-area: pager; position: relative; z-index: 1;
    margin-top: 0; width: auto; height: 52px; background: none;
  }
  /* the phone showcase's pager clock: width and colour together, 600ms in-out,
     no delay, the same for the arriving and the leaving pill */
  .orb__card[data-orb-live] .dev__pager i {
    transition: width .6s var(--e2),
                background-color .6s var(--e2);
  }
  .orb__card[data-orb-live] .orb__step {
    position: relative; z-index: 1;
    display: grid; place-items: center;
    width: 52px; height: 52px; padding: 0; border: 0;
    border-radius: 22px; background: none;
    color: var(--deep-fg); opacity: .82;
    cursor: pointer;
    -webkit-tap-highlight-color: transparent;
    transition: transform var(--dur-fast) var(--acc-ease), opacity var(--dur-fast) var(--acc-ease);
  }
  .orb__card[data-orb-live] .orb__step:active { transform: scale(.9); opacity: 1; }
  .orb__step--prev { grid-area: prev; }
  .orb__step--next { grid-area: next; }

  /* THE SECTION OPENING — nav, then air, then the eyebrow, then the line,
     then the stage. The extra top pad is the air under the floating nav; the
     tighter foot ties the headline to the stage as one chapter. */
  #orb { padding-top: calc(var(--sec-y) + 28px); padding-bottom: 64px; }
  #orb .statement { gap: 14px; }
  /* 96, NOT 44. The section is taller than any phone, so the header and the
     stage are two compositions, not one — and the old 44 put the headline
     exactly in the floating nav's 14-64 band at the rest position where the
     stage is framed (measured: headline 3-60 at 390x844, the nav sitting on the
     words). 96 is the distance that carries it fully off the top instead: at
     the framed rest the headline's foot is at -4 and the eyebrow's at -43. The
     stage's own top edge is invisible — no fill, no hairline — so this reads as
     air under the headline, not as a gap above a box. */
  .orb__card[data-orb-live] { margin-top: 96px; }

  @media (prefers-reduced-motion: reduce) {
    .orb__card[data-orb-live] .stage,
    .orb__card[data-orb-live] .dev__pager i,
    .orb__card[data-orb-live] .orb__step { transition: none !important; }
  }
}
/* one composition, four device rungs: screen width, then the state-3 drop and
   overlap at the same factor. Every rung was pulled back by the same 0.873 so
   the framing is identical on every phone — measured state 1 at 67.7 / 67.7 /
   67.2 / 65.7 / 67.0 % of the viewport at 375, 390, 393, 402 and 430. */
/* FIVE RUNGS, NOT FOUR. The plate's height is 100svh - 132px, so the room above
   the caption grows FASTER than the viewport's width does — and it steps again
   at 420, where the caption's body stops wrapping to three lines and hands back
   another 20px. Measured with four rungs the device's foot sat 34px (430) and
   55px (440) ABOVE the glass while it still overlapped by 28px at 390: the same
   composition read differently on every phone. With these five the overlap is
   8-28px (state 1), 20-40px (state 2) and 0-25px (state 3) everywhere. */
@media (max-width: 389px) {
  .orb__card[data-orb-live] .dev { --pw1: 228; --pw2: 252; --pw3: 164; --lap3: 57px; --top3: 16px; }
}
@media (max-width: 359px) {
  .orb__card[data-orb-live] .dev { --pw1: 210; --pw2: 232; --pw3: 150; --lap3: 52px; --top3: 14px; }
}
@media (min-width: 402px) and (max-width: 759.98px) {
  .orb__card[data-orb-live] .dev { --pw1: 262; --pw2: 281; --pw3: 183; --lap3: 64px; --top3: 26px; }
}
@media (min-width: 420px) and (max-width: 759.98px) {
  .orb__card[data-orb-live] .dev { --pw1: 302; --pw2: 344; --pw3: 208; --lap3: 73px; --top3: 38px;
    /* the caption drops to two lines here, so the dock is 20px shorter */
    }
  .orb__card[data-orb-live] .stage--1 { padding-bottom: calc(173px + 5px); }
  .orb__card[data-orb-live] .stage--2 { padding-bottom: calc(173px + 11px); }
  .orb__card[data-orb-live] .stage--3 { padding-bottom: calc(173px + 40px); }
}

/* ==================================================== 13. THE EMOTIONAL BAND
   A FULL-BLEED PHOTOGRAPHIC MOMENT. 107.4vw wide, 1.06 viewports tall, and the
   photograph fills the whole band — it is the only thing in it.

   MEASURED ON THE REFERENCE, settled (its entrance transforms have to finish
   before any rect is worth reading, which is what corrupted the earlier pass's
   numbers): band section 1440x752 in a 709 viewport = 106vh; the dark panel
   above ends 45px into it — an ordinary --overlap, the SAME 45px every other
   panel on this page rides — and the image is 1546x782 at x=-53, filling the
   band edge to edge. The reader therefore sees ~707px of photograph in a 709px
   viewport: ONE band owning the screen.

   WHAT WAS WRONG: this section pulled itself up by --band-pull (450px, ten
   times the reference's overlap) and then inset .moment__img by that same
   450px to keep the crop out of the hidden strip. Both together left a 410px
   image — 0.58 of the viewport — with the dark panel above it and the
   information panel below it sharing the screen: three bands in one viewport.
   The strip was never "emerging" either; a static overlap covers what it
   covers for good, so those 450px were simply thrown away.

   So: the pull is --overlap like every other panel, and the image fills the
   box. No inset and nothing sticky; what moves is the reference's scroll
   grammar, registered just below. */
/* THE BAND CARRIES THREE OBJECTS, NOT ONE, AND IT IS ONE VIEWPORT TALL.
   The reference's band is three compositional ROLES stacked in one frame:

     1. an oversized photograph        1546x782 in a 1440x752 clip: it bleeds
                                       53px off each side and ~15 top and
                                       bottom — 107.4% wide AND 104% tall
     2. a right-aligned display line   487x198, upper right, right edge flush
                                       to the 60px gutter
     3. an ACTION                      a 140x42 solid-white pill on radius 12,
                                       right edge aligned to the headline,
                                       30px below it

   SEEK carried 1 and 2 and stopped. The third role is filled truthfully rather
   than decoratively: SEEK has a real waitlist and a real anchor for it, so the
   pill is the same object the reference's is — the same 140x42 geometry, and
   as it happens the same words. Nothing was invented to reach three; the
   reference's fourth group (a sub-headline over five feature chips) is NOT
   added, because those are Fixa's features and SEEK does not have them.

   HEIGHT. 752 of a 764 viewport is 0.984. SEEK was at 1.060 — one viewport
   plus 55px, which is the amount that pushes the headline's own air off the
   fold. 98svh puts the band on the reference's ratio. */
/* ---- THE THREE DRIVEN PROPERTIES OF THIS JOIN, REGISTERED ----------------
   page.js §4b writes them, each a Fixaplan scroll range (MOTION-SPEC §4):

     --mt   #moment's offset        -0.412 x H(#orb) -> 0px   ref #orb, thr .5
     --mz   the photograph's scale   1 -> 1.2                 ref #moment, thr 1
     --lp   #information's offset   -120px -> 0px            ref #moment, thr .5

   Registered so a malformed write can never invalidate a transform and drop
   the photograph's mirror, and so each INITIAL VALUE is the resting, static
   composition: with no JavaScript, and under prefers-reduced-motion (where §4b
   returns before it writes anything), the page renders the plain 45px overlaps
   — no reduced-motion override block exists or is needed. None inherits: each
   is read only by the element it is written to, and #information carries the
   whole FAQ, which an inheriting property would restyle on every frame. */
@property --mt { syntax: '<length>'; inherits: false; initial-value: 0px; }
@property --mz { syntax: '<number>'; inherits: false; initial-value: 1; }
@property --lp { syntax: '<length>'; inherits: false; initial-value: 0px; }

#moment {
  position: relative; z-index: 1;
  margin-top: calc(var(--overlap) * -1);
  margin-bottom: 0;
  min-height: 98svh;
  overflow: clip;
  /* THE CHAPTER WAITS UNDER THE DARK PANEL. z-index 1 under #orb's 2, lifted
     by --mt, so it emerges at 0.588x the scroll as #orb's bottom edge wipes up
     over it. Whole pixels (it carries type). #information's range starts with
     18.5px of this travel left and still hangs 120px itself, so no strip opens
     between the two on screen (page.js §4b). */
  transform: translate3d(0, var(--mt, 0px), 0);
  will-change: transform;   /* the offset is a compositor scroll animation (page.js 0a) on `transform`; --mt is only the no-scrub fallback. Keeps the tiled layer pre-promoted at a stable contentsScale */
  /* THE BAND'S OWN GROUND — insurance. At rest the photograph covers this box
     edge to edge and this colour is never seen; should a sub-pixel seam ever
     show at an edge it paints the scrim's own colour (rgb(6,4,10) against the
     photograph's rgb(6,5,9)) rather than the page ground. */
  background: #06040A;
  /* THE COMPOSITION IS THE REFERENCE'S, NOT ITS MIRROR IMAGE. The band's
     height was already right; everything inside it was flipped. The reference
     sets its headline RIGHT-ALIGNED IN THE UPPER RIGHT of the band; SEEK had
     it left-aligned on the bottom-left corner — the opposite corner and the
     opposite alignment. Same one object, moved to where the reference puts it.
     The reference's CTA button and its four floating chips are NOT added: one
     object is fewer than six, and fewer is allowed. */
  display: flex; align-items: flex-start; justify-content: flex-end;
  padding-block: clamp(88px, 9.7vw, 140px) 0;
}
/* max-width:none is load-bearing. The base `img { max-width: 100% }` reset
   silently clamps this, so the authored bleed never applies.

   THE BLEED IS SET BY THE SOURCE'S RESOLUTION, NOT BY TASTE. The photograph is
   the kitchen frame at 2000x1116. The band's box is ~891 CSS px tall, so
   object-fit:cover has to resolve one scale for it, and which TERM wins that
   max() is the whole question:
     cover scale = max(boxWidth/2000, boxHeight/1116)
   The two terms cross where boxWidth/boxHeight = 2000/1116 = 1.7921, i.e. at
   890.8 x 1.7921 = 1596.4 CSS px, which is 110.9vw at 1440. Past that the
   WIDTH term wins and the scale climbs with every extra vw of bleed; below it
   the HEIGHT term is fixed at 890.8/1116 = 0.7982 and cannot be beaten at this
   band height. 106vw sits under that knee, so it buys the floor scale AND
   ~70px of horizontal freedom to place the pair.
     Note the sign: 0.7982 is a DOWNSCALE in CSS px, and a deep one. The band
   does not enlarge this photograph at all on a 1440 desktop — the browser's
   remaining upscale is purely the device-pixel one (x DPR, x --mz), and this
   2000px master cut that from 2.00x to 1.69x against the 1672px frame it
   replaced. The 1376x768 frame before THAT needed 1.1944 here, i.e. it was
   being stretched before DPR was even applied.
     A pre-upscaled lanczos variant was generated and tested against a native
   file at DPR 2, cropped to the faces, and NOT kept: it read no sharper,
   because resampling adds no detail and costs a re-encode. Resolution here
   comes from a genuinely larger master, never from resampling. The four
   shipped variants (640/960/1376/2000) are the whole set, and 2000 — the
   master's own width — is the cap.

   object-position 100% 50% — the band's aspect (1.71) is narrower than this
   photograph's (1.79), so cover crops ~70px off the width and the pair sits
   at one end of it. 100% puts both of them fully in frame; anything lower
   walks the man off the left edge. The VERTICAL half of this pair is inert at
   any desktop width, because the height term binds and leaves zero vertical
   slack; the band's vertical framing is the photograph's own, and the only
   thing that trims it is --mz zooming about the element's centre.

   scaleX(-1) is a STATIC transform, not an effect. It matters more with this
   photograph than the last one: unmirrored, the bright window and the pale
   wall sit under the right-aligned headline. Mirrored, the pair moves left and
   the shelf/wall corner moves right, which is the quiet ground the type needs. */
.moment__img {
  /* the band's box bleeds on BOTH axes — the horizontal figure is derived
     above, the vertical 104% + top:-2% is unchanged and is what keeps the
     crop off the band's top and bottom edges. These four lines are still
     constants — the ONE thing that is not is --mz, and page.js §4b owns it. */
  /* THE VERTICAL BLEED IS 1%, NOT 4%, AND THAT IS A ZOOM DECISION.
     cover binds on HEIGHT here (see the knee above), so this box's height is
     the ONLY thing that sets the scale — every extra percent of vertical bleed
     is an extra percent of enlargement, paid on a source that is already being
     stretched by the device pixel ratio. 104% cost 3.9% of enlargement to buy
     insurance against a sub-pixel seam at the band's top and bottom edges;
     101% buys the same insurance (~4px at a 882px band, and #moment's own
     background is #06040A — the scrim's colour — behind it) for 1%.
     Nothing here can expose an edge in any case: --mz is never below 1. */
  position: absolute; left: -3vw; top: -0.5%;
  width: 106vw; height: 101%; max-width: none;
  object-fit: cover; object-position: 100% 50%;

  /* ---- THE PUSH-IN, and why the mirror is written HERE ---------------------
     The reference's photo scales 1 -> 1.2, linear, over the chapter's own
     height, starting when the chapter's layout top reaches the viewport bottom
     (page.js §4b writes --mz). A uniform scale of the IMAGE ELEMENT ALONE, about
     its own centre (the CSS default — do not author transform-origin): the
     type is a sibling in .moment__inner and never scales. --mz is never below
     1, so no edge of the bleed box can be exposed at any progress or width.
     (The 1.2 depth is the reference's; an earlier pass held it to 1.12 to
     limit device-pixel upscaling of this 2000px master.)

     THE MIRROR LIVES IN THE STYLESHEET, NOT IN JS. page.js writes the scalar
     --mz and never writes `transform`, so no code path can drop the flip; a
     bare scale(s) would silently un-mirror the photograph onto the headline.
     The leading translate3d(0,0,0) keeps the layer promoted. */
  transform: translate3d(0, 0, 0) scaleX(-1) scale(var(--mz, 1));
  will-change: transform;   /* the push-in is a compositor scroll animation (page.js 0a) on `transform` (--mz only in the no-scrub fallback); promoted up front so its first frame never re-tiles this 550x788 layer */
}
/* Three gradients, one layer, and the third one is here because the kitchen
   frame is lit the opposite way round from the bench frame it replaced. That
   one had its DARK side — cabinets and shadow — under the right-aligned
   headline; this one has its BRIGHT side there: a daylit window and a pale
   wall. White type on that is not legible, so the right-hand gradient is
   deeper than it was (.86 against .76, and it now carries a mid stop) and a
   corner radial takes down the top-right specifically.
   It is a gradient behind the TYPE, not a wash over the picture: it reaches
   full strength only in the outer ~20% of the width, and the subjects — who
   sit left of centre — are untouched by it. Nothing here desaturates, tints or
   filters the photograph; the couple keeps its own colour.
   The vertical gradient is unchanged in role, seating the band into the panels
   above and below it. */
.moment__scrim {
  position: absolute; inset: 0;
  background:
    radial-gradient(78% 70% at 92% 16%, rgba(6,4,10,.62) 0%, rgba(6,4,10,0) 74%),
    linear-gradient(264deg, rgba(6,4,10,.86) 0%, rgba(6,4,10,.56) 30%, rgba(6,4,10,.14) 58%, rgba(6,4,10,0) 78%),
    linear-gradient(180deg, rgba(6,4,10,.38) 0%, rgba(6,4,10,.10) 34%, rgba(6,4,10,.42) 74%, rgba(6,4,10,.86) 100%);
}
/* 30px is the reference's own measured gap between its headline's bottom edge
   and the top of its pill. */
/* THE TYPE RIDES WITH THE PHOTOGRAPH. On the reference the headline, its lines
   and the CTA pill are rigid with each other AND with the image while the whole
   chapter emerges from under the panel above — one body, one offset, which is
   why the offset (--mt) is on #moment itself. The image's own scale happens
   INSIDE that body, which is why --mz is on the sibling img and can never reach
   these glyphs. */
.moment__inner { position: relative; display: grid; justify-items: end; gap: 30px;
  transform: none; }
/* the reference's band button is 140x42 on radius 12, the same object as the
   hero plate's, and its right edge is flush with the headline's */
.moment__cta { width: 140px; padding: 0; border-radius: var(--r-pill); }
/* the band's supporting line: right-aligned under the headline, on the same
   edge as everything else in this composition. One rung under .t-section and
   one over the body ladder, so it reads as part of the statement rather than
   as a caption. */
.moment__lede {
  margin: 0; justify-self: end; text-align: right; max-width: 430px;
  font-family: var(--f-ui); font-weight: 400;
  font-size: 17px; line-height: 1.34; letter-spacing: -.015em;
  color: rgba(255,255,255,.86);
}
/* the band's line is a SECTION DISPLAY — one of the ladder's own rungs, held
   to the upper right and right-aligned, as the reference's is. */
.moment__inner .t-section { color: #FFF; max-width: 680px; text-align: right; }
/* The 16px sub-line that used to sit under it is DELETED. Cutting the band
   from 1.43vp to 1.06vp leaves ~320px of visible photograph at 1440x765, and
   a headline plus a three-line paragraph does not fit in it. RULE 6: the fix
   for a crowded viewport is to delete, and the band's one idea is the
   photograph. Its content — the five layers — is already carried by the fourth
   story card and by the first FAQ answer. */
/* MOBILE. 82% is re-verified against THIS photograph, not inherited: at
   390x844 the height term binds at 0.9141, which leaves 1041 CSS px of
   horizontal slack — enough for the window to miss a face entirely. Swept at
   62/72/82/92/100%, 62 and 72 hold only the woman, 92 and 100 push her face
   into the right edge, and 82 is the one value that carries BOTH faces whole.
   The vertical is not a choice on a portrait phone: a landscape photograph
   under cover always binds on height there, so object-position's second value
   has nothing to move.
   The SCRIM is overridden, because the desktop one is built around a
   right-aligned headline and mobile does not have one: at 390 the type spans
   nearly the full width and the woman's face sits directly under the
   top-right corner, so the desktop corner radial fell on her face and dimmed
   it. Here the same job is done by a wide, soft band across the TOP of the
   frame instead of a tight corner — it holds the two lines and the pill, and
   it clears her face entirely. */
@media (max-width: 999px) {
  .moment__img { left: -12.5vw; width: 125vw; object-position: 82% 50%; }
  .moment__scrim {
    background:
      radial-gradient(96% 46% at 70% 2%, rgba(6,4,10,.52) 0%, rgba(6,4,10,0) 78%),
      linear-gradient(180deg, rgba(6,4,10,.50) 0%, rgba(6,4,10,.12) 32%, rgba(6,4,10,.42) 72%, rgba(6,4,10,.88) 100%);
  }
}

/* ======================================================= 14. INFORMATION ====
   The exhale. One 670px reading column, under-designed on purpose. */
/* THE PANEL LAGS OVER THE PHOTOGRAPH — the reference's FAQ panel arrives 120px
   HIGH and spends it linearly over the photo chapter's height, starting when
   the chapter's layout top is at half the viewport (page.js §4b writes --lp:
   -120px -> 0). On screen the edge rises at ~0.85x: a heavy slide over the
   photograph, which continues underneath, so the lift only ever covers more of
   it. The one thing it opens is BELOW, over #closing: ::after closes that. */
#information { --lift: 120px; padding-bottom: 0; transform: translate3d(0, var(--lp, 0px), 0); }
/* THE SKIRT. #information and #closing are flush, so the lift opens a strip of
   page ground between them up to 120px tall. The skirt fills it with the same
   fill both neighbours resolve to (var(--editorial) = the light canvas in
   light), so no join shows at any offset. --lift + 6px of margin; page.js §4b
   reads --lift. At rest it overhangs #closing, which is harmless: #closing (z 2,
   later in the document) paints over the whole overhang, type included.
   pointer-events:none so it can never take a click. */
#information::after {
  content: ''; position: absolute; left: 0; right: 0; top: 100%;
  height: calc(var(--lift) + 6px);
  background: var(--editorial);
  pointer-events: none;
}
/* the same chip-left row as the light panel and the dark panel, sized to the
   670 reading column: 115 + 555 is the reference's own split */
.info__head { display: grid; grid-template-columns: 1fr; gap: 24px; }
.info__head > .chip { justify-self: start; white-space: nowrap; }
/* the trust beat now leads with what the reader gets and puts the beta status
   under it, so the column needs a body wrapper — .info__head is a two-column
   grid at >=760 and a second child would otherwise fall into the chip column. */
.info__body { display: grid; gap: 14px; }
.info__sub { max-width: 52ch; }
@media (min-width: 760px) {
  .info__head { grid-template-columns: 136px 534px; gap: 0; align-items: start; }
}
/* THE ACCORDION HAS NO CLOCK IN THIS FILE. Its motion is the FLIP in the
   page's <style data-flip> / <script data-flip> (MOTION-SPEC §5: 650ms
   ECUBIC open, 650ms phone / 500ms desktop close, followers on the same
   clock). Here are only its resting states; without that script (reduced
   motion) every state change is instant. */
.faq { display: grid; gap: 24px; margin-top: 60px; }
.faq__row { border-radius: var(--r-card); background: var(--tint); overflow: hidden; }
.faq__btn {
  width: 100%; display: flex; align-items: center; justify-content: space-between; gap: 20px;
  padding: 24px 28px;
  background: transparent; border: 0; cursor: pointer; text-align: left;
  color: var(--fg);
}
/* a bare chevron. The 22px filled disc behind it was a surface per row, five
   times, for an 11px mark. */
.faq__ico {
  flex: 0 0 auto; width: 22px; height: 22px;
  display: grid; place-items: center;
  color: var(--fg-3);
  /* the rotation's clock is the FLIP's (<style data-flip> §2) */
}
.faq__btn[aria-expanded="true"] .faq__ico { transform: rotate(180deg); }
/* RULE 10 — the COLLAPSED state is owned by the runtime, not by the
   stylesheet. With no script every answer is simply open and readable; the
   script sets data-live and collapses them, which is the only state in which
   a reader can actually operate the control. */
/* THE PANEL IS A 0fr / 1fr GRID TRACK that commits in one frame: the FLIP
   measures that commit and draws the change. VISIBILITY, not display:none,
   takes the closed answer out of the tree, so the panel always has a box to
   measure. overflow:hidden sits on the PANEL, not the track, because an Nfr
   track on an auto-height grid resolves to N^2 x H while the panel box is
   N x H — the clip has to be the box. */
.faq__panel   { display: grid; grid-template-rows: 1fr; overflow: hidden; }
.faq__panelin { min-height: 0; }
/* the padding belongs to the CONTENT of the collapsing grid item, not to the
   item itself: padding on the item survives a 0fr track and would leave every
   collapsed row 24px too tall. */
.faq__panelin > p { padding: 0 28px 24px; }
[data-qa][data-live] .faq__panel { grid-template-rows: 0fr; visibility: hidden; }
[data-qa][data-live] .faq__btn[aria-expanded="true"] + .faq__panel { grid-template-rows: 1fr; visibility: visible; }
/* the answer never translates; on desktop it fades with the box (the FLIP's
   clock), on a phone it is simply revealed by the edge */
[data-qa][data-live] .faq__panelin { opacity: 0; }
[data-qa][data-live] .faq__btn[aria-expanded="true"] + .faq__panel .faq__panelin { opacity: 1; }
[data-qa]:not([data-live]) .faq__ico { display: none; }
/* REDUCED MOTION needs no carve-out here any more: there is no transition and
   no visibility delay to survive the page-wide duration kill, so a toggle is
   instant and the closed answer leaves the tree on the same frame. */

/* =========================================================== 15. CLOSING ====
   THE STATEMENT DOES NOT MOVE, AND IT IS NOT BURIED. Both halves of that
   sentence are measurements.

   THE HEADLINE IS ORDINARY FLOW, RATE 1.000. Tracked against a rigid FAQ
   landmark in both scroll directions: 1.00042 over 262px down, 1.00119 and
   1.00233 over ~118px up, and an independent point check that disagrees by
   0.003 CSS px. It does not fade (the first emerging sliver's per-row ink
   amplitudes match the fully settled state to 0.12%), it has no mask REVEAL
   (a mask edge would stay put while the glyphs moved; the visible ink top
   moves at exactly document rate), it does not scale, and its two lines are
   ONE RIGID BLOCK - their shifts differ by at most 0.025px across every frame
   where both are measurable. So the one-shot data-in="line"/"line2" entrance
   this h2 used to carry is REMOVED: the reference headline has no entrance at
   all, and the brief's rule is that a measured negative result is honoured
   rather than decorated.

   THE STATIC MASK STAYS EXACTLY AS AUTHORED. The reference carries one too -
   a single element-anchored top-to-bottom gradient across BOTH lines, ink
   depth 170 at line 1's cap line falling monotonically to 6 at line 2's
   baseline, byte-identical at three different scroll positions - and it never
   animates. SEEK's is gentler than that (alpha ~0.49 at line 2's baseline
   against the reference's ~0.035), which is why it survives the geometry fix
   below with the statement still legible.

   WHAT ACTUALLY MOVES IS THE FOOTER CARD, and it is scroll-linked. See section
   16 and page.js section 4b. */
#closing {
  position: relative; z-index: 2;
  background: var(--editorial); color: var(--editorial-fg);
  --fg: var(--editorial-fg); --fg-2: var(--editorial-fg2); --fg-3: var(--editorial-fg3);
  --tint: var(--editorial-tint); --line: var(--editorial-line);
  padding: clamp(90px, 9.7vw, 140px) var(--pad-x) 0;
}
#closing .t-closing {
  /* 808 in the reference; SEEK's closing line is longer and 808 broke it into
     three lines, which is not a controlled line break. 1000 gives the two the
     copy was written for. */
  width: min(1000px, 100%); margin-inline: auto;
  text-align: center;
  /* the type dies out where the card comes over it, exactly as the
     reference's does — a mask, so it is the GLYPHS that fade, not a plate
     laid on top of them */
  -webkit-mask-image: linear-gradient(180deg, #000 56%, rgba(0,0,0,.5) 84%, rgba(0,0,0,0) 100%);
  mask-image: linear-gradient(180deg, #000 56%, rgba(0,0,0,.5) 84%, rgba(0,0,0,0) 100%);
}
/* THE STATEMENT'S BOX BOTTOM IS THE CARD'S LAYOUT TOP (gap 0, the reference's
   flex gap). Every .mask gives its .14em of clip room back with a negative
   margin; on the LAST line that margin collapsed through #closing (no bottom
   padding) and pulled the footer .14em up INTO the statement's box — 7.8px on a
   phone, ~16px at 1440. Keeping it here puts the card's untransformed top edge
   exactly on the h2's bottom edge at every width. */
#closing .t-closing .mask:last-child { margin-bottom: 0; }
/* THE SAME DISSOLVE, RE-AIMED FOR THREE LINES. The stops above are measured
   against the desktop break, where the statement sets in two lines and line 2
   sits in the 56-100% band. On a phone the same 56px rung sets THREE lines, so
   the last one lands at 71-100% and the page's final word rendered at .137 alpha
   at its foot — the payoff half of "Less searching. More understanding." was the
   faintest thing on the page. Same gesture, same element, stops moved down so the
   type still dies into the card without taking the word with it. */
@media (max-width: 759.98px) {
  /* ONE OPENING DISTANCE UNDER THE FLOATING NAV. The nav pill occupies y 4..54
     on every phone and never hides, so the air under it is the page's most
     repeated measurement — and it was not one measurement. At each chapter's
     own entry rest the first ink landed at 100 (#understanding, #story,
     #information) but at 82 in #moment and 88 in #closing, i.e. 28px and 34px
     of clearance against everyone else's 46. Both are the page's largest
     display type, so they were the two places the nav crowded hardest. These
     raise only the MOBILE floor — the base clamps are 88/96 with a vw term that
     takes over at ~1072px and ~915px, so 1024 and up are untouched. */
  #moment  { padding-top: 104px; }


  #closing .t-closing {
    -webkit-mask-image: linear-gradient(180deg, #000 72%, rgba(0,0,0,.78) 92%, rgba(0,0,0,.42) 100%);
    mask-image: linear-gradient(180deg, #000 72%, rgba(0,0,0,.78) 92%, rgba(0,0,0,.42) 100%);
  }
}

/* ============================================================ 16. FOOTER ====
   A dark rounded card rising over the closing canvas; the card is the one
   thing in the ending that moves (page.js §4b). */
/* 12px gutters and the page ENDS ON THE CARD. The reference's footer card is
   ~1417 of 1440 wide and sits flush at the document's end; SEEK's was inset
   47px a side and left 162px of dead panel underneath it. */
#footer {
  position: relative; z-index: 2;
  background: var(--editorial);
  padding: 0 12px 12px;
  /* flow-root, so a margin on the card could never collapse through and drag
     the footer's background box with it (the card carries none today) */
  display: flow-root;
  --fg: var(--deep-fg); --fg-2: var(--deep-fg2); --fg-3: var(--deep-fg3);
  --tint: var(--deep-tint); --line: var(--deep-line);
}
.foot__card {
  background: var(--deep); color: var(--deep-fg);
  border-radius: var(--r-card);
  padding: clamp(24px, 3.3vw, 48px);
  /* NO MARGIN: the card's layout top IS the statement's box bottom (gap 0), as
     the reference's is, so the release closes the gap to exactly 0 at its end
     and no band ever opens between the two. */
  position: relative;
  /* THE RELEASE (page.js §4b). The card rests lifted over the statement and
     slides down to its slot over a scroll range of its own height, starting
     when its layout top reaches the viewport bottom, followed by the
     reference's footer spring:
       --fdy   translateY -L -> 0; L = 1.25 x the statement's height on phones
               (the reference's 130/104), 1.04 x at >= 760 (210/202), so the
               statement starts buried and is revealed from behind the card
       --fs    scale .95 -> 1 at >= 760 only; 1 on phones
     transform-origin is the CSS default, the centre, as the reference's is:
     the scale also pulls the top edge ~2.5% of the card's height down, which
     is exactly the -199 top edge measured on the reference's -210 translate.
     Both default to the resting composition (@property below), so no JS,
     reduced motion or the first frame render the card at its flow position. */
  transform: translate3d(0, var(--fdy, 0px), 0) scale(var(--fs, 1));
}
/* Registered so the fallbacks ARE the static composition and so a bad or
   absent runtime can only ever render the resting state. Neither inherits:
   both are read by exactly one element. */
@property --fdy { syntax: '<length>'; inherits: false; initial-value: 0px; }
@property --fs  { syntax: '<number>'; inherits: false; initial-value: 1; }
.foot__grid { display: grid; gap: clamp(28px, 2.8vw, 40px); grid-template-columns: 1fr; }
/* the mark OR the word, not both — the brand was stated twice in one lockup */
.foot__brand { display: flex; align-items: center; }
.foot__brand img { width: 40px; height: auto; }
.foot__h { margin-top: 28px; max-width: 22ch; text-wrap: balance; }
.foot__form { display: flex; flex-wrap: wrap; gap: 10px; margin-top: 24px; max-width: 480px; }
.foot__form input {
  flex: 1 1 220px; height: 42px; padding: 0 16px;
  border-radius: var(--r-chip); border: 0;
  background: rgba(255,255,255,.07); color: var(--deep-fg);
  box-shadow: inset 0 0 0 1px var(--deep-line);
  font-size: 16px;   /* < 16 makes iOS Safari zoom the page on focus */
}
.foot__form input::placeholder { color: var(--deep-fg3); }
.foot__note { margin-top: 14px; line-height: 1.3; }
/* TWO headed groups, because there are only two left to run. The third was
   deleted rather than thinned, and a three-track grid holding two groups is a
   hole, not a column. */
.foot__cols { display: grid; gap: 28px; grid-template-columns: repeat(2, minmax(0,1fr)); }
.foot__cols h3 { font-size: 14px; font-weight: 500; color: var(--deep-fg3); margin-bottom: 14px; }
.foot__cols li { color: var(--deep-fg2); font-size: 16px; line-height: 1.2; }
.foot__cols li + li { margin-top: 10px; }
.foot__cols a { display: inline-block; padding-block: 5px; margin-block: -5px; color: var(--deep-fg2); font-size: 16px; transition: color var(--dur-fast) var(--ease-enter); }
.foot__cols a:hover { color: var(--deep-fg); }
/* three items spread across the full width, as the reference's bottom bar */
.foot__base {
  margin-top: clamp(32px, 4.2vw, 60px); padding-top: 22px;
  border-top: 1px solid var(--deep-line);
  display: flex; flex-wrap: wrap; gap: 14px 32px; justify-content: space-between;
  color: var(--deep-fg3); font-size: 14px;
}
.foot__base a { transition: color var(--dur-fast) var(--ease-enter); }
.foot__base a:hover { color: var(--deep-fg); }
@media (min-width: 900px) {
  .foot__grid { grid-template-columns: minmax(0, 1fr) minmax(0, .84fr); gap: 60px; }
  .foot__cols { gap: 24px; }
}

/* ============================================================= 17. MOTION ===
   THE SECTION ENTRANCES HAVE NO RULE IN THIS FILE. Their trigger lines,
   offsets, clocks and curves are one table in page.js §3, played as WAAPI on
   the individual translate / rotate / scale properties plus opacity, so no
   declaration here hides anything and a page without that script shows
   every block at rest. What stays here is the global reduced-motion floor. */
@media (prefers-reduced-motion: reduce) {
  *, *::before, *::after {
    animation-duration: .001ms !important;
    animation-iteration-count: 1 !important;
    transition-duration: .001ms !important;
    scroll-behavior: auto !important;
  }
  .curtain { display: none !important; }
  /* the pill's entrance is motion and goes: without this its delay would hold
     it off-screen, since only the duration above is killed */
  .nav { animation: none !important; }
}

/* A viewport too short to pin anything gets the narrow composition, which is
   the same composition without the pin.

   THIS USED TO READ max-height: 704px, AND THAT WAS THE BUG THE READER ACTUALLY
   HIT. The reasoning was sound — a pinned column taller than the fold holds the
   story's "Join the waitlist" button below the fold for the whole 1789px of pin
   travel, so it must not pin — but the premise was that the column is
   immovably 705px tall. Now that .story__pin shrinks to the fold, the column
   fits, and unpinning at 704 was throwing away the sticky column and the entire
   stage index on ordinary laptop windows: 1440x761 leaves ~674px of
   viewport after browser chrome, which failed the old gate by 31px and rendered
   four permanently-lit cards with no pin.

   The floor is now set by CONTENT, not by the old fixed height: the column
   needs 421px, and min-height gives it 100svh-116, so it fits while
   100svh-116 >= 421, i.e. 537px. Below that there genuinely is no room and the
   narrow composition is correct. */
@media (max-height: 536px) {
  #story[data-live] .story__pin { position: static !important; }
}


/* THE PHONE'S SCALE RUNGS, all of them, in one place. Each is the largest
   scale whose shell (411 + 16 bezel = 427 design px) still fits inside
   100vw - 2 * --pad-x. Below 1000 the phone is the only thing in the column,
   so it is allowed to be larger than the desktop's 0.68. */
@media (max-width: 359px)                        { .pscr { --ps: .65; } }  /* 427*.65 = 278 <= 280 */
@media (min-width: 360px) and (max-width: 389px) { .pscr { --ps: .74; } }  /* 427*.74 = 316 <= 320 */
@media (min-width: 390px) and (max-width: 445px) { .pscr { --ps: .78; } }  /* 427*.78 = 333 <= 350 */
@media (min-width: 446px) and (max-width: 999px) { .pscr { --ps: .78; } }


/* #understanding — the light ground moves to ::after and crossfades. The
   section already establishes the stacking this needs: ::before sits at
   z-index 0 and `#understanding > *` at z-index 1, so a second pseudo at
   z-index 0 lands above the fill and below every child. */
#understanding::after {
  content: ""; position: absolute; inset: 0; z-index: -1; pointer-events: none;
  border-radius: inherit;
  background: var(--r2-canvas);
  opacity: 0;
}
html[data-theme="light"] #understanding::after { opacity: 1; }
html[data-theme="light"] #understanding { background-color: transparent; }


/* #information — the light ground is painted permanently on a z-index:-1 pseudo and
   the section's own opaque dark fill animates to transparent over it. Nothing
   here interpolates a gradient: the reveal is a colour dissolving. */
#information { position: relative; }
#information::before {
  content: ""; position: absolute; inset: 0; z-index: -1; pointer-events: none;
  border-radius: inherit;
  background: var(--r2-canvas);
  opacity: 0;
}
html[data-theme="light"] #information::before { opacity: 1; }
html[data-theme="light"] #information { background: transparent; }

#closing { position: relative; }
#closing::before {
  content: ""; position: absolute; inset: 0; z-index: -1; pointer-events: none;
  border-radius: inherit;
  background: var(--r2-canvas);
  opacity: 0;
}
html[data-theme="light"] #closing::before { opacity: 1; }
html[data-theme="light"] #closing { background-color: transparent; }

/* THE STORY'S GROUND (light theme) — Reference 5's mood in SEEK's hand: a
   warm charcoal base with a muted amber glow low on the left, bronze in the
   upper left and moss on the right, blended as four soft radial fields over
   one diagonal. CSS only — no photograph, no live blur — plus a 4.5% grain
   tile so the long gradients never band. It sits on the one fixed layer the
   page already has, so the colour stays still while the chapter scrolls
   through it. It only shows where no opaque
   panel is — the How SEEK works chapter and the panel corners around it.
   Opacity 0 in dark, so dark keeps its own plum field underneath. */
.atmos::after {
  content: ""; position: absolute; inset: 0; pointer-events: none;
  background:
    url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='180' height='180'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='.85' numOctaves='2' stitchTiles='stitch'/%3E%3CfeColorMatrix values='0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 .045 0'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E"),
    radial-gradient(58% 62% at 30% 76%, rgba(172,130,62,.58) 0%, rgba(172,130,62,.20) 45%, rgba(172,130,62,0) 72%),
    radial-gradient(52% 50% at 8% 14%, rgba(104,80,40,.62) 0%, rgba(104,80,40,0) 70%),
    radial-gradient(56% 64% at 90% 26%, rgba(56,70,50,.66) 0%, rgba(56,70,50,0) 72%),
    radial-gradient(70% 55% at 76% 104%, rgba(34,40,30,.80) 0%, rgba(34,40,30,0) 70%),
    linear-gradient(158deg, #2C2717 0%, #262619 46%, #1C1E18 100%);
  opacity: 0;
}
html[data-theme="light"] .atmos { --glow-1: 146,96,148; --glow-2: 255,142,122;
  --glow-3: 31,211,174; --glow-4: 10,8,18; --atmos-base: #281F38;
  --atmos-1: .52; --atmos-2: .18; --atmos-3: .07; --atmos-4: .66; }
html[data-theme="light"] .atmos::after { opacity: 1; }

/* ═══════════════ THE THEME TRANSITION ═══════════════════════════════════════
   Gated on a class the toggle adds and removes, for the three reasons the brief
   names: the first paint must never animate from the wrong theme, ordinary
   state changes (accordions, the scroll-driven lanes and scenes) must not
   be slowed, and nothing may be left transitioning once the switch is over.

   The property list is explicit — no `transition: all`, and not one layout
   property — so the page cannot move while it changes theme.

   opacity is absent from the universal list on purpose: page.js writes
   per-frame values from the scroll on several layers, and a universal opacity
   transition would smear any of them for half a second. The layers that
   genuinely need opacity are named individually below. */
html.theme-anim,
html.theme-anim *,
html.theme-anim *::before,
html.theme-anim *::after {
  /* THREE ENTRIES WERE REMOVED AS DEAD WEIGHT, counted on the built page:
       text-decoration-color  1203 subjects report a change and NOT ONE of them
                              has any text-decoration at all — the value tracks
                              currentColor, so it "changes" on every element
                              whose ink changes and paints nowhere.
       outline-color          same story; an outline is only drawn on :focus-
                              visible, which is not a theme surface and must not
                              animate under a keyboard user mid-switch.
       backdrop-filter        0 subjects change it in either direction.
     border-*-color STAYS even though only 31 of its 1203 changes land on
     something that actually draws a border: those 31 are the hairlines inside
     the app mockups and the FAQ rows, they are visible, and there is no
     selector for "has a border" that would not go stale. */
  transition-property: color, background-color, border-color, border-top-color,
    border-right-color, border-bottom-color, border-left-color, box-shadow, fill, stroke;
  transition-duration: var(--theme-dur);
  transition-timing-function: var(--theme-ease);
  transition-delay: 0s;
}
/* the crossfading grounds, by name */
html.theme-anim #understanding::before,
html.theme-anim #understanding::after,
html.theme-anim #orb::after,
html.theme-anim #information::before,
html.theme-anim #closing::before,
html.theme-anim .atmos::after,
html.theme-anim .nav__theme .ico-sun,
html.theme-anim .nav__theme .ico-moon {
  transition: opacity var(--theme-dur) var(--theme-ease),
              transform var(--theme-dur) var(--theme-ease),
              color var(--theme-dur) var(--theme-ease),
              background-color var(--theme-dur) var(--theme-ease),
              box-shadow var(--theme-dur) var(--theme-ease);
}
/* media is not a theme surface and must not be touched mid-switch */
html.theme-anim video, html.theme-anim img,
html.theme-anim picture, html.theme-anim source { transition: none !important; }


/* ── ONE VISUAL SYSTEM: THE SNAPSHOT DISSOLVE ──────────────────────────────
   Everything above this comment is the FALLBACK. It runs only where the View
   Transitions API is missing, because page.js adds .theme-anim only in that
   case. Where the API is present the class is never added, not one of those
   property transitions runs, and the entire switch is a single compositor
   crossfade between two complete renders of the page.

   That is the whole point. Interpolating a thousand colours independently is
   what produced the staged look: gradients cannot interpolate at all (a
   background-image is a DISCRETE property in every engine), so they had to be
   crossfaded as separate opacity layers while everything around them lerped
   its own channels, and no amount of shared duration makes a dozen different
   kinds of change read as one movement. A root snapshot has no such seam —
   there is exactly ONE animated value on the page, and it is an opacity.

   NO reveal, NO clip-path, NO mask, NO wipe, NO scale, NO slide. The two
   snapshots are the same box at the same place by definition, and group() is
   pinned to `animation: none` so no browser default can invent motion.

   mix-blend-mode: plus-lighter IS DECLARED, NOT INHERITED, and it is the most
   important line here. Chrome sets it in its own sheet, but that value did not
   survive naming a custom animation — measured, both ways, at eleven points:
   without this line the midpoint sat at 0.79 of the way to the destination
   instead of 0.50, and the first tenth covered 0.21. That IS the complaint:
   the switch appeared to reach most of the new theme almost immediately.

   The cause is compositing, not timing. Without plus-lighter the two layers
   source-over each other, and two premultiplied layers at (1-t) and t sum to
   an alpha of only t + (1-t)^2 — 0.75 at the midpoint — so a quarter of the
   live page beneath, ALREADY in the destination theme, showed through the
   middle of every switch. plus-lighter ADDS them instead: the composite is
   exactly old*(1-t) + new*t, fully opaque at every instant, and the measured
   progression became 0.100 0.199 0.301 0.399 0.502 0.599 0.702 0.799 0.902.
   isolation on the image-pair is what keeps that addition off the page. */
::view-transition-group(root)      { animation: none; }
::view-transition-image-pair(root) { isolation: isolate; }
::view-transition-old(root),
::view-transition-new(root) {
  mix-blend-mode: plus-lighter;   /* MEASURED, not inherited — see above */
  animation-duration: var(--theme-dur);
  animation-timing-function: var(--theme-blend);
  animation-fill-mode: both;
  transform: none;            /* geometry is identical; nothing may move */
}
::view-transition-old(root) { animation-name: seek-dissolve-out; }
::view-transition-new(root) { animation-name: seek-dissolve-in; }
@keyframes seek-dissolve-out { from { opacity: 1 } to { opacity: 0 } }
@keyframes seek-dissolve-in  { from { opacity: 0 } to { opacity: 1 } }

/* NOTHING ELSE MAY MOVE WHILE THE DISSOLVE RUNS, and this is the list of what
   otherwise would — counted live during a real switch, not guessed. A theme
   change starts exactly two transitions outside the crossfade:

     the toggle's own sun/moon swap   270ms   opacity + transform, 2 icons
     the four How SEEK Works chips    420ms   color (S.4 spring)

   Both are quick, so both would finish about a third of the way through the
   700ms dissolve and read as a layer arriving early — which is the exact
   complaint this pass exists to fix.

   They are SUPPRESSED rather than re-clocked, because ::view-transition-new is
   LIVE, not a still: an instant change inside it is invisible while it is
   fading up from zero, and the crossfade is already carrying the motion. Giving
   them a second 700ms animation would just animate the same change twice.

   Named, not universal. A blanket `html.theme-vt * { transition: none }` would
   sit at (0,1,1) and silence every (0,1,0) state clock — button hover, the
   skip link, the stepper — for two thirds of a second, freezing motion that
   has nothing to do with the theme. The check that keeps this list honest is a
   measurement, not a comment: during a switch, document.getAnimations() must
   contain the two view-transition entries and nothing else. */
html.theme-vt .nav__theme .ico-sun,
html.theme-vt .nav__theme .ico-moon { transition: none; }
/* The chips need a scalpel, not a switch. Their transition lives on
   `#story .story__chip` (1,1,0), out of reach of a (0,2,0) override, and it
   also carries the height/font-size of the stage hand-over — motion that
   must not freeze. All three properties share one duration and one curve,
   so dropping colour from the property list re-pairs nothing. */
html.theme-vt #story .story__chip { transition-property: height, font-size; }

@media (prefers-reduced-motion: reduce) {
  :root { --theme-dur: 1ms; }
  ::view-transition-group(root),
  ::view-transition-old(root),
  ::view-transition-new(root) { animation: none !important; }
}


/* ══════════════════════════════════════════════════════════════════════════
   14. THE MOBILE PASS — iPhone / Safari 26
   ══════════════════════════════════════════════════════════════════════════
   Everything below was designed and measured against a real iPhone over a
   diagnostic cycle (candidates A→L). It is the promoted, approved state:
   K's visual work plus L's Safari theme-switch behaviour. It is LAST in the
   sheet on purpose — several rules here deliberately override earlier ones,
   and the order is what lets them.

   Every rule is scoped to (max-width: 999px) except the two color-scheme
   lines, which are inert unless data-theme and data-chrome disagree — a state
   that only exists inside a theme switch. Desktop is provably untouched.

   ACCEPTED LIMIT, do not try to fix again: a tiny Safari-native settle at the
   end of a theme switch. WebKit clears the temporary root-contents background
   at View Transition teardown and Liquid Glass settles the toolbar material on
   its own clock; neither is reachable from page CSS. */
/* ── L: color-scheme scheduled with theme-color, not with data-theme ──────
   Active ONLY while data-theme and data-chrome disagree, i.e. only during the
   350ms first half of a switch. Specificity (0,2,1) beats page.css's
   html[data-theme="light"] (0,1,1) and :root (0,1,0), and this block is
   emitted after the stylesheet link, so it wins on order as well. Nothing
   here can affect a resting page: at rest the two attributes always agree. */
html[data-theme="light"][data-chrome="dark"] { color-scheme: dark; }
html[data-theme="dark"][data-chrome="light"] { color-scheme: light; }

/* ── L: the safe-area strips stop stepping ────────────────────────────────
   MEASURED, from the 60fps recording: for the whole 700ms the two strips
   iOS keeps outside the View Transition's snapshot — the 62pt above the
   viewport and the 98pt under the floating toolbar — stop showing page
   content and show a FLAT fill instead. Horizontal detail inside them
   collapses from a spread of 14-152 luma to 0-5, and it comes back in ONE
   frame at +733ms. That is the appear/disappear, and it is not theme-color:
   the fill is the document's base background, i.e. body's own
   "background: var(--page)", which flips the instant data-theme flips inside
   the update callback — 700ms before the page it borders gets there.

   So the strips are given the same clock as the dissolve. body's background
   is 99%+ covered by the two snapshots while they run — forcing it to full
   magenta mid-dissolve moves the viewport by at most 2 of 255 — so this is
   invisible on the page and lands almost entirely in the strips. It is not a
   new movement either: K already changes this value mid-dissolve, as a step.
   This only makes the same change gradual.

   Scoped to html.theme-vt, which page.js adds for exactly the transition
   window, so nothing at rest can be touched; and to mobile, because the
   strips are an iOS Safari artefact and desktop must stay provably untouched. */
@media (max-width: 999px) {
  html.theme-vt body { transition: background-color var(--theme-dur) linear; }
}

/* ---- THE STORY'S GROUND ON PHONES IS THE SAME FIXED LAYER AS DESKTOP'S ----
   The reference's ground is one fixed viewport layer on both breakpoints: the
   rounded panel edges uncover it and it never scrolls. Phones used to paint a
   ~2800px field on #story::before/::after that travelled with the section at
   1.0x; that is gone and .atmos (§5, fixed, z-index 0, one composited layer)
   shows at every width. 100lvh so the field still reaches the bottom of the
   screen after iOS retracts its toolbar (svh would leave the last ~100px bare).
   will-change keeps it on its own layer in both engines, so scrolling over it
   only recomposites and never repaints its gradients.

   THE SECTION JOINS. The atmosphere sits below every section. #story drops to
   z-index 0 (desktop does the same at >= 1000) so its context sits below the
   z 1-2 panels either side of it, whose rounded corners uncover the field;
   every other section is an opaque positioned panel above it. */
@media (max-width: 999px) {
  .atmos { height: 100lvh; will-change: transform; }
  #story { z-index: 0; }
}
/* SAFARI'S BARS. iOS Safari 26 paints its status area and toolbar from body's
   background-color (live), unless a position:fixed box of roughly viewport
   size (0.9-1.05x on both axes) with z-index >= 0 sits at the edge — then it
   paints that box's background-color and KEEPS it for the rest of the visit.
   .atmos is fixed and 100lvh: 1.06x the viewport with the toolbar up (too
   large, ignored), 1.01x once a finger scroll collapses the toolbar — and
   from then on both bars were its #281F38, in both themes, whatever the page
   did. Reproduced in Simulator Safari with a real touch scroll; WebKit's rule
   is LocalFrameView::fixedContainerEdges (negative z-index is rejected).
   z-index:-1 takes it out of that test and changes nothing it paints: every
   section above it is a positioned panel (z 0-2), and the root canvas it now
   sits on is exactly what it already covered. body is again the only source,
   so the approved theme-color / data-chrome / canvas-pin machine drives the
   bars as designed. */
.atmos { z-index: -1; }

/* HOW SEEK WORKS UNDER SAFARI'S BARS — the chapter paints its own field (page.js
   §4c). Safari's bars copy only the scrolling page, never fixed layers, so the
   fixed .atmos alone left them the bare canvas. On touch devices #story gets a
   clipped box at the bottom of its own stacking context holding a copy of
   .atmos, moved by a compositor scrub so it stays exactly on the fixed field:
   same pixels on the page, and real glow in the bars' copy. The strips show
   the page just outside the viewport, so the copy is mirrored 160px past
   its top and bottom edges. */
#story.has-ground { isolation: isolate; }
#story.has-ground > .wrap { position: relative; z-index: 1; }
.story__ground {
  position: absolute; inset: 0; z-index: 0;
  overflow: clip; pointer-events: none;
}
.story__field {
  position: absolute; left: 0; top: 0; width: 100vw; height: 0;
  will-change: transform;
}
.story__field .atmos {
  position: absolute; left: 0; top: 0; right: auto; bottom: auto;
  z-index: 0; will-change: auto;
}
/* the strips beyond the viewport's edges: the field mirrored about its own
   top and bottom rows, 160px deep (more than either obscured inset) */
.story__ext { position: absolute; left: 0; width: 100vw; height: 160px; overflow: clip; }
.story__ext--top { top: -160px; }
.story__ext--top > .atmos { top: 160px; transform-origin: 50% 0; transform: scaleY(-1); }
.story__ext--bot { top: 100svh; }
.story__ext--bot > .atmos { top: auto; bottom: 160px; transform-origin: 50% 100%; transform: scaleY(-1); }
@media (max-width: 999px) { .story__ext--bot { top: 100lvh; } }

/* ---- the skip link, unchanged ---- */
@media (max-width: 999px) {
  .skip { position: absolute; clip-path: inset(50%); }
  .skip:focus { clip-path: inset(50%); transform: translateY(-90px); }
  .skip:focus-visible { clip-path: none; transform: none; }
  @supports not selector(:focus-visible) { .skip:focus { clip-path: none; transform: none; } }
}



@media (min-width: 1000px) {
  #story { z-index: 0; }
}






/* ╔═══════════════════════════════════════════════════════════════════════════
   ║  THE REFERENCE-2 SYSTEM — dx-reference2-system-final
   ║
   ║  One layer, written once, after the machinery it depends on. It is the
   ║  approved specification translated into SEEK: Reference 2's proportions,
   ║  material and pacing, SEEK's own fonts, imagery, product UI and copy.
   ║
   ║  PAGE RHYTHM (light):  hero (dark / video)
   ║                        → LIGHT 1   Why SEEK + SEEK is designed differently
   ║                        → DARK      How SEEK works — ambient photograph
   ║                        → DARK      See it in action — product
   ║                        → PHOTO     How two lives actually fit
   ║                        → LIGHT 2   FAQ + closing, one surface
   ║                        → dark inset footer
   ║  Exactly two light chapters. Dark keeps its approved palette; only
   ║  structure (type tiers, spacing, the Why SEEK and Designed-differently
   ║  compositions) is shared between the themes, so a toggle never reflows.
   ║
   ║  SCROLL-LINKED, AND NOTHING ELSE: (1) the Why SEEK field + the phone's own
   ║  screen, (2) the How SEEK works sticky column, (3) the photo chapter's
   ║  zoom, (4) the FAQ panel lift, (5) the footer card. Everything else is a
   ║  one-shot entrance or normal flow.
   ╚═══════════════════════════════════════════════════════════════════════════ */

:root {
  --r2-canvas: #EAEAE8;
  /* geometry — one container, one margin, one spacing ladder */
  --r2-container: 1320px;
  --r2-margin: clamp(20px, 6.35vw, 96px);
  --r2-sec: clamp(96px, 11.1vw, 160px);
  --s-1: 4px;  --s-2: 8px;  --s-3: 12px; --s-4: 16px; --s-5: 24px;
  --s-6: 40px; --s-7: 60px; --s-8: 88px; --s-9: 160px;
  /* type tiers — SEEK's Newsreader / Instrument Sans, Reference 2's ratios */
  --t-d1: clamp(56px, 35.2px + 5.333vw, 112px);  /* closing                  */
  --t-d2: clamp(46px, 30.6px + 3.95vw, 84px);    /* hero, designed differently */
  --t-h:  clamp(34px, 30.3px + .95vw, 44px);     /* chapter headline         */
  --t-l:  clamp(24px, 21px + .77vw, 32px);       /* the two-tone lead        */
  --t-t:  20px;                                  /* cards                    */
  --t-b:  16px;                                  /* UI, body                 */
  --t-s:  14px;                                  /* chips, labels            */
}

/* ── TYPE TIERS ─────────────────────────────────────────────────────────── */
.t-hero      { font-size: var(--t-d2); }
.t-closing   { font-size: var(--t-d1); }
.t-story     { font-size: var(--t-h); line-height: 1; }
.t-statement { font-size: var(--t-l); font-weight: 400; line-height: 1.2; }
.t-card {
  font-family: var(--f-ui); font-weight: 500;
  font-size: var(--t-t); line-height: 1.25; letter-spacing: -.015em;
}
#moment .t-section { font-size: clamp(42px, 30.3px + 2.6vw, 68px); letter-spacing: -.035em; }

/* ── ONE CONTAINER ──────────────────────────────────────────────────────── */
.wrap {
  max-width: calc(var(--r2-container) + 2 * var(--r2-margin));
  padding-inline: var(--r2-margin);
}

/* ── CHIPS: one form, fill step, no hairline in light ───────────────────── */
html[data-theme="light"] .chip { color: var(--fg-2); }

/* ── HERO: the dark substrate under the poster; the media is never animated ─ */
.hero__media { position: absolute; inset: 0; }
html[data-theme="light"] #hero { background: #141413; }
.hero__cta-t { color: rgba(255,255,255,.92); }


/* ═══════════════════════════ WHY SEEK ═══════════════════════════════════════
   The preferred three-zone composition (dx-seek-premium-final), made clearer
   for someone who has never heard of SEEK.

   TEXT ABOVE — Reference 2's text group: the chip in its own narrow column,
   then ONE editorial block ~560px wide: the two-line claim in the serif, the
   plain-language explanation under it in the body sans, secondary tone.

   BELOW — one frame, three zones, read left to right without arrows:
     LEFT        many people: full-colour profile cards in three lanes that
                 move up with the scroll — "Typical dating apps / You sort
                 through hundreds of profiles." They fade out before the phone.
     CENTRE-RT   one person: SEEK's real full profile, the whole device, the
                 largest and sharpest object; its "Why I introduced you" card
                 carries a quiet teal cue and lifts once when the phone settles.
     FAR RIGHT   the reason: "With SEEK / No endless feed. / A reason behind
                 every introduction." — bottom-aligned with the field's label,
                 so the two captions share one line across the frame.
   No Orb, no sticky, no arrows, no diagram. */
#understanding { padding-bottom: calc(var(--r2-sec) * .6); overflow: clip; }
/* THE HEAD — ONE EDITORIAL COMPOSITION on Reference 2's skeleton: a quiet
   eyebrow in its own 181px column, then ONE text column (560) that the
   headline and the secondary statement share, the whole block centred.
     headline   the SEEK serif at 36 on 1.18, a touch of air (-.015em), set as
                the two deliberate lines the markup already breaks
     statement  Instrument Sans 400 at 23 on 1.45 in a soft charcoal (74% of
                the ink, not a washed-out grey), balanced over three lines
     rhythm     28px headline -> statement: one clear step, not a paragraph gap
     eyebrow    13px, lightly tracked, a pill that sits on the headline's first
                line — a label, not a button
   Below 900 it stacks: eyebrow, 24, headline, 20-28, statement. */
.ws__head {
  width: min(741px, 100%); margin-inline: auto;
  display: grid; grid-template-columns: minmax(0, 1fr); row-gap: var(--s-5);
  align-items: start;
}
.ws__head .chip {
  justify-self: start; align-self: start;
  padding: 6px 11px; border-radius: 999px;
  font-size: 13px; line-height: 1; letter-spacing: .02em;
  color: var(--fg-2);
}
.ws__copy { max-width: 560px; }
.ws__lead {
  margin: 0; color: var(--fg);
  font-family: var(--f-display); font-weight: 400; font-optical-sizing: auto;
  font-size: clamp(28px, 22px + .95vw, 36px); line-height: 1.18; letter-spacing: -.015em;
  text-wrap: balance;
}
.ws__more {
  margin: clamp(20px, 16px + .8vw, 28px) 0 0;
  color: color-mix(in srgb, var(--fg) 74%, transparent);
  font-family: var(--f-ui); font-weight: 400;
  font-size: clamp(19px, 16px + .5vw, 23px); line-height: 1.45; letter-spacing: -.01em;
  text-wrap: balance;
}
@media (min-width: 900px) {
  /* the eyebrow column (181) + the shared text column (560), centred as one */
  .ws__head { grid-template-columns: 181px minmax(0, 560px); justify-content: center; }
  /* centred on the headline's first line (42.5 line box, 25 pill) */
  .ws__head .chip { margin-top: 9px; }
  .ws__lead { white-space: nowrap; }
}

.ws {
  /* dark — the approved dark stage: the plinth's lavender whisper */
  --ws-bg:
    radial-gradient(60% 80% at 68% 30%, rgba(146,96,148,.16) 0%, rgba(146,96,148,0) 70%),
    linear-gradient(180deg, rgba(169,155,255,.07) 0%, rgba(169,155,255,.025) 100%);
  --ws-edge: inset 0 1px 0 rgba(247,242,246,.07), inset 0 0 0 1px rgba(169,155,255,.10);
  --ws-card: #221C2C;
  --ws-field-op: .94;
  --ws-accent: #1FD3AE;
  --ws-tag-bg: rgba(16,15,24,.58);
  --ws-tag-fg: rgba(247,242,246,.72);
  --ws-tag-strong: #F7F2F6;
  --ws-phone-shadow: 0 44px 60px -36px rgba(0,0,0,.70);
}
html[data-theme="light"] .ws {
  /* an editorial ground, not a container: a step lighter than the canvas,
     no border, no box shadow */
  --ws-bg: linear-gradient(180deg, rgba(255,255,255,.52) 0%, rgba(255,255,255,.30) 100%);
  --ws-edge: none;
  --ws-card: #D8D8D5;
  --ws-field-op: .95;
  --ws-accent: var(--l-teal-text);   /* 6.3:1 — .ws__say-k is 10.5px caps; --l-teal is 3.7:1 there */
  --ws-tag-bg: rgba(255,255,255,.78);
  --ws-tag-fg: rgba(17,17,17,.66);
  --ws-tag-strong: #111111;
  --ws-phone-shadow: 0 34px 40px -30px rgba(0,0,0,.34), 0 80px 120px -80px rgba(38,34,20,.30);
}

.ws__stage {
  --ws-ps: .8;                               /* the phone's scale; JS fits it */
  --ws-cx: .64;                              /* the phone's centre, x         */
  --ws-phw: calc(427px * var(--ws-ps));      /* the whole device, bezel incl. */
  --ws-pad: clamp(24px, 2.8vw, 40px);
  --ws-sg: clamp(28px, 3.2vw, 52px);         /* phone → payoff                */
  position: relative;
  width: calc(100% - 24px); max-width: 1680px;
  margin: var(--s-7) auto 0;
}
.ws__frame {
  position: relative;
  height: clamp(540px, 78svh, 740px);
  border-radius: var(--r-frame);
  background: var(--ws-bg);
  box-shadow: var(--ws-edge);
  isolation: isolate;
}

/* LEFT — the many. Three lanes of full-colour cards that end just under the
   phone's edge and fade out before it. Every lane is longer than the frame
   plus its whole travel, so nothing wraps or resets. */
.ws__field {
  position: absolute; top: 0; bottom: 0; left: 0;
  right: calc((1 - var(--ws-cx)) * 100% + var(--ws-phw) / 2 - 56px);
  display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 16px;
  padding-left: var(--ws-pad);
  border-radius: var(--r-frame) 0 0 var(--r-frame);
  overflow: hidden;
  opacity: var(--ws-field-op);
  -webkit-mask-image:
    linear-gradient(90deg, #000 0%, #000 64%, transparent 100%),
    linear-gradient(180deg, transparent 0%, #000 8%, #000 92%, transparent 100%);
  -webkit-mask-composite: source-in;
  mask-image:
    linear-gradient(90deg, #000 0%, #000 64%, transparent 100%),
    linear-gradient(180deg, transparent 0%, #000 8%, #000 92%, transparent 100%);
  mask-composite: intersect;
}
.ws__col {
  display: grid; gap: 16px; align-content: start;
  transform: translate3d(0, var(--o, 0px), 0);
  will-change: transform;
}
.ws__card {
  position: relative; margin: 0;
  aspect-ratio: 4 / 5;
  border-radius: 18px; overflow: hidden;
  background: var(--ws-card);
}
.ws__card img { display: block; width: 100%; height: 100%; object-fit: cover; object-position: 50% 24%; }
.ws__card::after {
  content: ""; position: absolute; inset: 0; pointer-events: none;
  background: linear-gradient(180deg, rgba(0,0,0,0) 55%, rgba(0,0,0,.5) 100%);
}
.ws__nm {
  position: absolute; left: 14px; bottom: 14px; z-index: 1;
  font-family: var(--f-ui); font-weight: 600; font-size: 14px; line-height: 1;
  letter-spacing: -.01em; color: #FFF;
}
.ws__act { position: absolute; right: 12px; bottom: 10px; z-index: 1; display: flex; gap: 6px; }
.ws__act i {
  width: 24px; height: 24px; border-radius: 50%;
  background: rgba(255,255,255,.22);
  display: grid; place-items: center;
}
.ws__act i::before {
  content: ""; width: 10px; height: 10px; background: #FFF;
  -webkit-mask: var(--ws-g) center / contain no-repeat; mask: var(--ws-g) center / contain no-repeat;
}
.ws__act i:first-child { --ws-g: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 10 10'%3E%3Cpath d='M1.5 1.5l7 7M8.5 1.5l-7 7' stroke='%23000' stroke-width='1.8' stroke-linecap='round'/%3E%3C/svg%3E"); }
.ws__act i:last-child  { --ws-g: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 10 10'%3E%3Cpath d='M5 9S.8 6.4.8 3.4C.8 2 1.9 1 3.1 1c.8 0 1.5.4 1.9 1 .4-.6 1.1-1 1.9-1C8.1 1 9.2 2 9.2 3.4 9.2 6.4 5 9 5 9z'/%3E%3C/svg%3E"); }

/* CENTRE-RIGHT — the one. The whole device, a little taller than the frame,
   on one soft contact shadow. It never moves; its screen does. */
.ws__one {
  position: absolute; z-index: 2;
  left: calc(var(--ws-cx) * 100%); top: 50%;
  transform: translate(-50%, -50%);
}
.ws .pscr { --ps: var(--ws-ps); flex: 0 0 auto; }
/* LIGHT THEME. The white inset ring and the hard 1px black outline were the
   old flat frame's stand-ins for an edge; the titanium band is that edge now,
   and both were painting straight over it. What a device on a pale ground
   needs instead is the near half of the same three-layer drop — a tight
   contact shadow that sits it ON the page, a short body shadow — over this
   section's own wide ambient, which stays as authored. */
html[data-theme="light"] .ws .pscr {
  box-shadow:
    0 2px 4px rgba(26,22,34,.28),
    0 14px 26px -8px rgba(26,22,34,.20),
    var(--ws-phone-shadow);
}
/* THE KEY MOMENT — SEEK's reason, inside the real profile UI: a teal label,
   a faint teal tint and edge, the reason set a size larger with more air.
   JS holds it at the app's plain lavender until the phone has settled, then
   lets it take the teal once (1.4s) — no loop, no glow. No JS: teal. */
.ws .pscr__why {
  padding: 16px 18px 18px;
  background: var(--ap-canvas);
  border-color: color-mix(in srgb, var(--ap-teal) 44%, transparent);
  box-shadow: inset 0 0 0 200px color-mix(in srgb, var(--ap-teal) 8%, transparent);
  transition: border-color 1.4s cubic-bezier(.22,.61,.36,1), box-shadow 1.4s cubic-bezier(.22,.61,.36,1);
}
.ws .pscr__whylabel { color: var(--ap-teal-text); transition: color 1.4s cubic-bezier(.22,.61,.36,1); }
.ws .pscr__whybody { margin-top: 10px; font-size: 16.5px; line-height: 1.42; color: var(--ap-ink); }
.ws[data-why="wait"] .pscr__why,
[data-why="wait"] .pscr__why {
  border-color: var(--ap-why-border);
  box-shadow: inset 0 0 0 200px color-mix(in srgb, var(--ap-teal) 0%, transparent);
}
[data-why="wait"] .pscr__whylabel { color: var(--ap-lav); }

/* the field's label — small, compact, on the frame's lower-left edge */
.ws__tag {
  position: absolute; z-index: 3;
  left: var(--ws-pad); bottom: var(--ws-pad);
  max-width: 290px;
  margin: 0; padding: 11px 15px 12px; border-radius: 14px;
  font-family: var(--f-ui); font-weight: 500; font-size: 14px; line-height: 1.3; letter-spacing: -.01em;
  color: var(--ws-tag-fg);
  background: var(--ws-tag-bg);
  -webkit-backdrop-filter: blur(14px); backdrop-filter: blur(14px);
}
.ws__tag-k {
  display: block; margin-bottom: 5px;
  font-weight: 700; font-size: 10.5px; line-height: 1; letter-spacing: .14em; text-transform: uppercase;
  color: var(--ws-tag-strong);
}

/* FAR RIGHT — the payoff, in the open space right of the phone, bottom-
   aligned with the field's label. The second statement carries the weight. */
.ws__say {
  position: absolute; z-index: 3;
  left: calc(var(--ws-cx) * 100% + var(--ws-phw) / 2 + var(--ws-sg));
  right: var(--ws-pad); bottom: var(--ws-pad);
  margin: 0;
  font-family: var(--f-display); font-weight: 400; font-optical-sizing: auto;
}
.ws__say-k {
  display: block; margin-bottom: 16px;
  font-family: var(--f-ui); font-weight: 700; font-size: 10.5px; line-height: 1; letter-spacing: .14em; text-transform: uppercase;
  color: var(--ws-accent);
}
.ws__say-a {
  display: block;
  font-size: clamp(18px, 1.25vw, 21px); line-height: 1.25; letter-spacing: -.01em;
  color: var(--fg-2);
}
.ws__say-b {
  display: block; margin-top: .45em; white-space: nowrap;
  font-size: clamp(22px, 1.8vw, 28px); line-height: 1.14; letter-spacing: -.02em;
  color: var(--fg);
}

/* ---- the full profile's lower half — the app's own sections, in dp ------- */
.pscr__body { padding-bottom: 40px; }
.pscr__label {
  margin: 0 0 9px;
  font-weight: 800; font-size: 10.4px; line-height: 1; letter-spacing: 1.664px;
  color: var(--ap-muted);
}
.pscr__take { display: grid; gap: 12px; }
.pscr__tl { margin: 0; font-weight: 700; font-size: 13.2px; line-height: 1.3; color: var(--ap-muted); }
.pscr__tb { margin: 3px 0 0; font-weight: 600; font-size: 13.2px; line-height: 21.4px; color: var(--ap-ink); }
.pscr__tb--2 { color: var(--ap-ink2); }
.pscr__ask { margin: 14px 0 0; font-weight: 600; font-size: 13.2px; line-height: 21.4px; color: var(--ap-muted); }
.pscr__ask b { font-weight: 600; color: var(--ap-ink2); }
.pscr__privacy { margin: 30px 0 0; font-weight: 600; font-size: 11.8px; line-height: 1.5; color: var(--ap-muted); }
.pscr__actions { display: grid; grid-template-columns: 1fr 1.35fr; gap: 10px; margin-top: 16px; }
.pscr__btn {
  height: 52px; border-radius: 26px;
  display: grid; place-items: center;
  font-weight: 800; font-size: 15px; line-height: 1;
  border: 1.5px solid var(--ap-chip-border); color: var(--ap-ink2);
}
.pscr__btn--go { border-color: transparent; background: var(--ap-teal); color: var(--ap-on-teal); }
/* the screen's own scroll is written by JS as one transform with the scale */
.ws .pscr__frame { will-change: transform; }

/* narrower desktops: two lanes, the phone a touch nearer the centre */
@media (max-width: 1199px) {
  .ws__stage { --ws-cx: .59; }
  .ws__field { grid-template-columns: repeat(2, minmax(0, 1fr)); }
  .ws__field .ws__col:nth-child(3) { display: none; }
}

/* PHONES — the story stacks instead of squeezing: the field's label, a short
   band of profiles moving up, SEEK's full profile rising out of the band's
   faded foot, then the payoff. No sticky; native scroll. */
@media (max-width: 899px) {
  .ws__stage {
    --ws-cx: .5; --ws-pad: 12px;
    width: calc(100% - 16px);
    margin-top: var(--s-6);
    display: flex; flex-direction: column;
  }
  .ws__tag {
    order: 0; position: static; max-width: none;
    margin: 0 0 14px; padding: 0 8px;
    border-radius: 0; background: none; -webkit-backdrop-filter: none; backdrop-filter: none;
    font-size: 15px; color: var(--fg-2);
  }
  .ws__tag-k { color: var(--fg); }
  .ws__frame { order: 1; height: auto; padding-bottom: 24px; border-radius: 20px; }
  .ws__field {
    position: relative; top: auto; bottom: auto; left: auto; right: auto;
    height: clamp(240px, 34svh, 300px);
    grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 8px;
    padding-inline: 8px;
    border-radius: 20px 20px 0 0;
    -webkit-mask-image: linear-gradient(180deg, transparent 0%, #000 14%, #000 58%, transparent 100%);
    mask-image: linear-gradient(180deg, transparent 0%, #000 14%, #000 58%, transparent 100%);
  }
  .ws__field .ws__col:nth-child(3) { display: grid; }
  .ws__col { gap: 8px; }
  .ws__card { border-radius: 12px; }
  .ws__nm, .ws__act { display: none; }
  .ws__one {
    position: relative; left: auto; top: auto; transform: none;
    width: max-content; margin: -84px auto 0;
  }
  .ws__say { order: 2; position: static; margin-top: 24px; padding: 0 8px; }
  .ws__say-b { white-space: normal; }
}


/* ═══════════════════════ SEEK IS DESIGNED DIFFERENTLY ═══════════════════════
   Its own section, on the SAME light surface as Why SEEK: in light the two
   grounds are one colour, so the rounded join between them disappears; in
   dark the approved deep-2 chapter stays exactly where it was. Reference 2's
   proportions: a lead-tier setup, one statement at the D2 tier, and three
   flat fill-step cards whose words sit at the bottom. */
#designed {
  position: relative; z-index: 1; isolation: isolate;
  margin-top: calc(var(--r-panel) * -1);
  padding-block: calc(var(--r-panel) + var(--r2-sec) * .4) var(--r2-sec);
  border-radius: 0 0 var(--r-panel) var(--r-panel);
  background: var(--deep-2);
  scroll-margin-top: 96px;
  overflow: clip;
}
#designed::after {
  content: ""; position: absolute; inset: 0; z-index: -1; pointer-events: none;
  border-radius: inherit;
  background: var(--r2-canvas);
  opacity: 0;
}
html[data-theme="light"] #designed { background-color: transparent; }
html[data-theme="light"] #designed::after { opacity: 1; }
html.theme-anim #designed::after { transition: opacity var(--theme-dur) var(--theme-ease); }

.wd {
  /* dark — SEEK's own dark palette, as approved */
  --wd-setup: rgba(247,242,246,.62);
  --wd-card: rgba(169,155,255,.055);
  --wd-card-edge: inset 0 0 0 1px rgba(169,155,255,.12);
  --wd-copy: rgba(247,242,246,.74);
  --wd-dim: rgba(247,242,246,.56);
  --wd-ink: #F7F2F6;
  --wd-accent: #1FD3AE;
  width: min(var(--r2-container), 100% - 2 * var(--r2-margin));
  margin-inline: auto;
  color: var(--wd-ink);
}
html[data-theme="light"] .wd {
  --wd-setup: var(--l-ink-2);
  --wd-card: var(--l-fill-a);
  --wd-card-edge: none;
  --wd-copy: var(--l-ink-2);
  --wd-dim: var(--l-ink-2);   /* the five layers are body copy; ink-3 is labels only (:342) */
  --wd-ink: var(--l-ink);
  --wd-accent: var(--l-teal);
}
.wd__setup {
  margin: 0; max-width: 14.5em;
  font-family: var(--f-display); font-weight: 400; font-optical-sizing: auto;
  font-size: var(--t-l); line-height: 1.1; letter-spacing: -.02em;
  color: var(--wd-setup);
  text-wrap: balance;
}
/* ONE HEADLINE IN TWO VOICES — the reference's pair, measured live on
   fixaplan.com (1512): a compact two-line lead-in (32 / 1.1, 424 wide) sitting
   directly on a statement three times its size (96), the statement's cap
   height landing ONE lead-in line below the lead-in's last baseline, so the
   big line reads as the next line of the same sentence. Same left edge; the
   mask gives back .16em of the margin. */
.wd__h {
  margin: calc(var(--t-l) * .66) 0 0;
  font-family: var(--f-display); font-weight: 300; font-optical-sizing: auto;
  font-size: clamp(44px, 25px + 4.7vw, 96px); line-height: 1; letter-spacing: -.035em;
  color: var(--wd-ink);
}
.wd__h .mask { padding-block: .16em; margin-block: -.16em; }
.wd__seek { letter-spacing: -.01em; }
.wd__h em { font-style: italic; font-weight: 400; letter-spacing: -.02em; padding-right: .04em; }
@media (min-width: 1200px) { .wd__h { white-space: nowrap; } }
.wd__cards {
  list-style: none; margin: var(--s-8) 0 0; padding: 0;
  display: grid; grid-template-columns: repeat(3, minmax(0, 1fr));
  gap: var(--s-3);
}
.wd__card {
  position: relative;
  display: flex; flex-direction: column;
  min-height: 352px;
  padding: var(--s-6);
  border-radius: var(--r-card);
  background: var(--wd-card);
  box-shadow: var(--wd-card-edge);
}
/* The mark's bottom margin is a FIXED rung, not `auto`. With auto, each card
   bottom-anchored its own text block, so the middle card's extra .wd__dims row
   lifted its heading 34px above its neighbours' and the three titles sat on
   three different lines at every desktop width. The rung is the one the
   <=899px branch below already uses, so both branches now read the same. */
.wd__mark { width: 40px; height: 40px; margin-bottom: var(--s-7); overflow: visible; }
.wd__ln { fill: none; stroke: var(--wd-ink); stroke-width: 1.4; stroke-linecap: round; stroke-linejoin: round; }
.wd__ac { fill: var(--wd-accent); }
.wd__t {
  margin: var(--s-6) 0 0;
  font-family: var(--f-ui); font-weight: 500;
  font-size: var(--t-t); line-height: 1.25; letter-spacing: -.015em;
  color: var(--wd-ink);
}
.wd__p {
  margin: 10px 0 0; max-width: 24em;
  font-family: var(--f-ui); font-weight: 400;
  font-size: 17px; line-height: 1.45; letter-spacing: -.01em;
  color: var(--wd-copy);
  text-wrap: pretty;
}
.wd__dims {
  margin: var(--s-4) 0 0;
  display: flex; flex-wrap: wrap; column-gap: 14px; row-gap: 4px;
  font-family: var(--f-ui); font-weight: 500;
  font-size: 13px; line-height: 1.5; letter-spacing: 0;
  color: var(--wd-dim);
}
.wd__dims span { display: inline-flex; align-items: center; gap: 7px; white-space: nowrap; }
.wd__dims span::before {
  content: ""; width: 4px; height: 4px; border-radius: 50%; background: var(--wd-accent);
}
@media (max-width: 899px) {
  .wd__cards { grid-template-columns: 1fr; margin-top: var(--s-6); }
  .wd__card { min-height: 0; padding: 28px; }
  .wd__mark { margin-bottom: var(--s-7); }
  .wd__t { margin-top: 0; }
  .wd__p { font-size: 16px; }
}

/* the one entrance (setup, the statement through its mask, the three cards)
   is page.js §3's table; nothing here hides or clocks it */


/* ══════════════════ HOW SEEK WORKS — THE WARM DARK CHAPTER ══════════════════
   In light, this chapter is the page's colour moment: the olive / amber /
   moss ground (.atmos::after, the fixed layer, at every width), crisp
   white type, cards as a deep 30% plate with a 7% hairline, and the app's
   dark screens. Dark theme keeps its own plum
   atmosphere, untouched. */
html[data-theme="light"] #story {
  color: #F5F5F3;
  --fg: #F5F5F3; --fg-2: rgba(245,245,243,.72); --fg-3: rgba(245,245,243,.56);
  --tint: rgba(255,255,255,.08); --line: transparent;
}
html[data-theme="light"] #story .story__card { --card-bg: rgba(16,15,10,.30); --card-line: rgba(255,255,255,.07); }
html[data-theme="light"] #story .scr { box-shadow: inset 0 0 0 1px rgba(255,255,255,.08); }
/* THE HOLD IS CAPPED. Four cards at 32px padding, a 290px screen and a 12px
   gap put the sticky column's travel at ~1.41k px — under the 1.45k budget,
   down from 1.8k — in both themes, so the toggle never changes the track. */
@media (min-width: 1000px) {
  #story .story__cards { gap: var(--s-3); }
  #story .story__card { padding: 32px; min-height: 0; }
  #story .scr { --sw: 290; }
}


/* ═══════════════════ SEE IT IN ACTION — THE DARK PRODUCT CHAPTER ═══════════
   In light, the whole chapter is neutral near-black (#171716), with the demo
   on a #1C1C1B inner panel — the same charcoal the footer card uses. */
html[data-theme="light"] #orb,
html[data-theme="light"] .orb__card,
html[data-theme="light"] #footer {
  --deep:      #171716;
  --deep-2:    #1C1C1B;
  --deep-fg:   #F5F5F3;
  --deep-fg2:  #C8C8C4;
  --deep-fg3:  #9A9A95;
  --deep-tint: rgba(255,255,255,.05);
  --deep-line: rgba(255,255,255,.10);
}
html[data-theme="light"] .orb__card { box-shadow: inset 0 0 0 1px var(--deep-line); }  /* the stage is dark in both themes; --deep-line above is declared for exactly this */
html[data-theme="light"] .orb__row     { background: rgba(255,255,255,.05); }
html[data-theme="light"] .orb__ico     { background: rgba(255,255,255,.09); }
html[data-theme="light"] .dev__pager,
html[data-theme="light"] .orb__step    { background: rgba(255,255,255,.05); }
html[data-theme="light"] .dev__pager i { background: rgba(255,255,255,.24); }
/* .phone gets NO light-theme shadow override: the stage is dark in both themes,
   so the device is lit the same way in both, and the old rule here replaced the
   three-layer drop with a flat one and ringed the band in white. */
/* balance holds the tinted clause on one line: the break falls after "you", so
   "one question at a time." is one whole tinted line instead of starting with a
   stranded "one" at the end of line 1 (both 1440 and 390). */
#orb .orb__h { text-wrap: balance; }
.orb__h2 { color: var(--fg-2); }

/* the How SEEK Works screens take the app's dark roles here and are handed the
   light ones back in that section's own block below. #orb is NOT listed: its
   presentation (ground, card, rows, pager, bezels) stays the charcoal stage
   above in light, and only the software inside the phones follows the theme. */
html[data-theme="light"] #story {
  --ap-canvas: #100F18;
  --ap-surface: #141020;
  --ap-ink: #F7F2F6;
  --ap-ink2: #C9C1D6;
  --ap-muted: #9A92AE;
  --ap-chip-fill: rgba(169,155,255,.09);
  --ap-chip-border: rgba(169,155,255,.22);
  --ap-rule: rgba(247,242,246,.16);
  --ap-teal: #1FD3AE;
  --ap-teal-text: #1FD3AE;
  --ap-teal-fill: rgba(31,211,174,.14);
  --ap-teal-border: rgba(31,211,174,.42);
  --ap-teal-glow: rgba(31,211,174,.5);
  --ap-on-teal: #06231C;
  --ap-verify-fill: rgba(31,211,174,.14);
  --ap-verify-border: rgba(31,211,174,.4);
  --ap-verify-text: #4FE0C4;
  --ap-lav: #A99BFF;
  --ap-surface2: #1A1520;
  --ap-floating: rgba(26,22,38,.86);
  --ap-veil: rgba(5,4,10,.74);
  --ap-coral: #FF8E7A;
  --ap-on-coral: #2B120B;
  --ap-why-fill: rgba(169,155,255,.12);
  --ap-why-border: rgba(169,155,255,.3);
  --ap-atmos:
    radial-gradient(78% 42% at 14% 0%,   rgba(132,88,140,.72) 0%, rgba(132,88,140,0) 62%),
    radial-gradient(60% 30% at 94% 10%,  rgba(169,155,255,.20) 0%, rgba(169,155,255,0) 64%),
    radial-gradient(80% 34% at 50% 104%, rgba(255,142,122,.16) 0%, rgba(255,142,122,0) 66%),
    #100F18;
}


/* ════ SEE IT IN ACTION — LIGHT SCREENS ON THE DARK STAGE (dx-product-demo-dark-stage-light-ui)
   Layer A, the presentation, is untouched (the charcoal block above).
   Layer B, the app UI inside the phones, inherits the app's real light roles
   from html[data-theme="light"] (seekTokens.ts / discoverDesign.ts). Two
   screen-level facts from the app source complete it: */
/* the reflection screen sits on the app's light atmosphere
   (InterviewReflectionScreen -> atmosphereFor('light'), seekTokens.ts:471-479) */
html[data-theme="light"] #orb .ui--reflect {
  background-image:
    radial-gradient(78% 40% at 14% 0%, rgba(124,86,130,.22) 0%, rgba(124,86,130,0) 62%),
    radial-gradient(60% 30% at 94% 10%, rgba(95,91,150,.14) 0%, rgba(95,91,150,0) 64%),
    radial-gradient(46% 24% at 99% 40%, rgba(10,142,119,.085) 0%, rgba(10,142,119,0) 62%),
    radial-gradient(80% 34% at 50% 104%, rgba(180,84,60,.15) 0%, rgba(180,84,60,0) 66%);
}
/* LISTENING is p.teal, not the text variant (VoiceAnswerPanel.tsx:195) */
html[data-theme="light"] #orb .ui__state { color: var(--ap-teal); }


/* ═══════════════════════ FAQ + CLOSING — LIGHT 2, ONE SURFACE ══════════════
   #information, its skirt, #closing and the footer's frame all resolve to the
   same canvas, so there is nothing left to draw a seam with: no second panel,
   no rounded shoulder, no divider. ~170px of canvas, then the statement, then
   the dark card inset 12px inside the same surface. */
.faq { gap: var(--s-3); }
#closing { padding-top: clamp(96px, 11.8vw, 170px); }
/* mobile floor only — see the nav-clearance note in the mobile demo block.
   The vw term takes over at ~915px, so 1024 and up are unchanged. */
@media (max-width: 759.98px) { #closing { padding-top: 108px; } }

/* ═══════ THE ENDING — TOUCH TARGETS, SAFE AREA, AND THE LAST FRAME ═════════
   MOBILE ONLY. Measured at 390x844 the whole tail of the page was under the
   iOS 44px minimum: the email field and its button 42 tall, every footer link
   a 29px box on a 29px stride (zero dead space between two adjacent targets,
   so a thumb aimed at "How it works" can only land on a neighbour), and "Back
   to top" — the sole navigation control at the end of the page — a bare 72x17
   text line with no padding at all. The footer also ended on a flat 12px
   gutter while the document declares viewport-fit=cover, so on a phone with a
   home indicator the base bar sat under it.
   AND THE LAST FRAME. At maxScroll — the page's terminal rest, where every
   reader stops and where the waitlist lives — the floating nav pill landed ON
   the closing statement: measured 22px of vertical overlap across 190px of
   "Less searching.", i.e. the page's largest type cut mid-word by the nav in
   the final frame. Because the nav is fixed and the statement is 3 lines on a
   phone, NO partial shift helps — moving the tail by less than the statement's
   own height just swaps which line is covered, and two of the intermediate
   values put TWO lines behind the pill instead of one. Every number below is a
   defect worth fixing on its own terms, and the touch targets are what the tail
   is made of now: the 32px the ending gives back (see #footer and .foot__base)
   comes out of spacing, not out of them, and it deliberately brings the
   statement's last line back INTO the final frame rather than above it — the
   reference's ending, where the pill and the last line share the top band. */
@media (max-width: 759.98px) {
  .foot__form input,
  .foot__form .btn { height: 48px; }
  /* 43px stride for EVERY row, link or not: the Availability column's first
     two items are plain <li> with no anchor, so pushing the rhythm onto the
     anchor alone left those two at a 19px stride. The margin carries the
     rhythm; the anchor's padding only grows the hit box back over it, so two
     adjacent targets touch exactly and never overlap. */
  .foot__cols li + li { margin-top: 24px; }
  .foot__cols a { padding-block: 12px; margin-block: -12px; }
  /* 16, not 32: the other half of the 32px of tail the ending gives back (see
     the #footer padding note below). The 14px hit-box padding on the links
     keeps the bar's own rhythm, so only the gap above the rule closes. */
  .foot__base { margin-top: 16px; align-items: center; }
  .foot__base a { display: inline-block; padding-block: 14px; margin-block: -14px; }
  /* NO EMPTY BAND AND NO CHAPTER BREAK. The footer is its natural height (no
     min-height) and #closing has no bottom padding at any width (its base
     shorthand), so the card's layout top is the statement's box bottom — the
     reference's document gap of 0.0px — and the card's release (page.js §4b)
     ends this padding before the document end, landing flush as the page runs
     out. The padding itself clears the home indicator.

     12 IS THE REFERENCE'S TAIL. SEEK's card plus a 28px gutter ran 52px longer
     than the reference's (622 + 12), while the statement above it is 76px
     taller, so at the document end the statement had climbed off the top of
     the screen entirely — the reference keeps its own last line in frame. The
     base drops to 12 and the bar above it loses 16 (.foot__base below): 32px
     of tail, taken from spacing only — no footer content, type or card padding
     moves. env() is still the floor, so the home indicator is cleared by its
     own inset rather than by 16px on top of it. */
  #footer {
    padding-bottom: max(12px, env(safe-area-inset-bottom));
  }

  /* ── TYPOGRAPHY: THREE ONE-WORD LAST LINES ───────────────────────────────
     Measured per word with Range boxes at 375/390/430: .info__lede broke as
     "You stay in control of what you" / "share." — a 54px line under a 288px
     one — and .foot__note ended on "opens.". text-wrap: balance is the whole
     fix; nothing about the size, leading or measure changes. */
  .info__lede,
  .info__sub,
  #information .faq__btn,
  .foot__note { text-wrap: balance; }

  /* ── CONTRAST: TWO PAIRS BELOW AA ────────────────────────────────────────
     Both sampled the honest way — the element's own glyphs made transparent and
     the pixels of its exact box read back. The story chapter's olive ground is
     the brightest ground any dim ink sits on anywhere on the page.
       · "SEEK is currently in private beta." in the story CTA: --fg-2 over
         rgb(101,85,47) = 4.35:1 mean in LIGHT, worse at its brightest.
       · the "How SEEK works" eyebrow: --fg-3 on the chip = 3.97:1 in dark.
     --fg-3 is documented "labels only, never body copy", and the first of these
     IS body copy. */
  html[data-theme="light"] #story .story__cta p { color: var(--fg); }
  #story .story__eyebrow { color: var(--fg-2); }
  html[data-theme="light"] #story .story__eyebrow { color: var(--fg); }

  /* #designed's own opening. 62.4px is the tightest chapter start on the page
     and the only one where ink rests INSIDE the nav band: at the section-top
     rest the setup line's glyph box measured y63..114 against a band that ends
     at 64. 88 puts it with the rest of the page (#moment 104, #closing 108,
     everyone else 100) and 25px clear of the pill's drop shadow. */
  #designed { padding-top: 88px; }

  /* ── #designed CARD RHYTHM ───────────────────────────────────────────────
     --s-7 (60px) between the 40px mark and the card title is a DESKTOP rung: it
     holds three-line titles apart in a >=352px-tall card. On a phone the card is
     261-296px tall, so that one gap was 21-23% of the whole card — measured
     identical at all five widths. --s-6 keeps the mark clearly separate and is
     on the scale. */
  .wd__mark { margin-bottom: var(--s-6); }
}


/* ═══════════════════════════ REDUCED MOTION ════════════════════════════════
   The runtime writes none of the five scroll-linked effects under reduced
   motion; the CSS side only has to make sure nothing is left mid-state. */
@media (prefers-reduced-motion: reduce) {
  .ws__col { transform: translate3d(0, var(--o, 0px), 0) !important; }
}


/* ═══════════════════ LIGHT THEME — LAYERED CHAPTERS ═════════════════════════
   The light page was one flat sheet. Three quiet moves give it depth without
   glass: (1) each light chapter opens with a soft top light, a wide white
   bloom that fades into the canvas; (2) "SEEK is designed differently" is
   its own lifted panel, a step lighter, with rounded top corners rising over
   Why SEEK and a long faint shadow — a clear chapter break inside the light
   run; (3) the Why SEEK frame is a luminous translucent plate (tokens above).
   All of it is light-only; dark keeps its own grounds. */
#understanding::after,
#information::before {
  background:
    radial-gradient(90% 560px at 50% 0%, rgba(255,255,255,.62) 0%, rgba(255,255,255,0) 100%),
    var(--r2-canvas);
}
/* LIGHT KEEPS THE DARK GEOMETRY AT THIS SEAM. The override here used to give
   #designed all four corners and raise it to z-index 3, so it sat ON #understanding
   instead of sliding under it and its two rounded top corners opened a notch at
   each page edge. In dark the notch is invisible (3 units); in light the olive
   atmosphere showed through as an ~8x30px khaki wedge on the left and a moss one
   on the right, a full viewport above the chapter they belong to. Inheriting the
   base rule closes both, and retires the only section shadow in the light theme. */
#designed::after {
  background: linear-gradient(180deg, #F3F3F1 0%, #EEEEEB 100%);
}




/* ═══════════ PRODUCT UI FIDELITY — THE APP'S REAL LIGHT ROLES ═══════════
   Verbatim from matching-app-mobile (read-only): src/theme/seekTokens.ts light
   roles and src/components/app/discoverDesign.ts (re-checked 11 Sep 2026, app
   repo unchanged at e660476). Only product mockups read --ap-*, so nothing else
   on the page moves. Dark theme is untouched: none of this matches
   html[data-theme="dark"]. */
html[data-theme="light"] {
  --ap-canvas: #F4EFEA;                     /* seekTokens.ts:163 canvas          */
  --ap-atmos: #F4EFEA;                      /* interview: flat canvas, no atmos  */
  --ap-surface: rgba(252,250,255,.84);      /* :165 surface                      */
  --ap-surface2: rgba(254,252,255,.94);     /* :167                              */
  --ap-floating: #FFFCF8;                   /* :171 floating                     */
  --ap-ink: #241E2B;                        /* :176 ink                          */
  --ap-ink2: #453C52;                       /* :177 inkSecondary                 */
  --ap-muted: #625870;                      /* :178 inkMuted                     */
  --ap-rule: rgba(92,78,132,.2);            /* :215 border (progress track)      */
  --ap-veil: rgba(52,42,68,.46);            /* :249 scrim                        */
  --ap-teal-fill: rgba(10,142,119,.15);     /* :202 aiFill                       */
  --ap-teal-border: rgba(10,142,119,.42);   /* :204 aiBorder                     */
  --ap-chip-fill: rgba(252,250,255,.84);    /* discoverDesign.ts:437             */
  --ap-chip-border: rgba(92,78,132,.2);     /* :438                              */
  --ap-verify-fill: rgba(10,142,119,.15);   /* :433                              */
  --ap-verify-border: rgba(10,142,119,.4);  /* :434                              */
  --ap-verify-text: #0A8E77;                /* :435                              */
  --ap-why-fill: rgba(92,78,132,.1);        /* :429                              */
  --ap-why-border: rgba(92,78,132,.28);     /* :430                              */
}
/* the why label is discoverDesign's #5C4E84 (:431), not the lavender token */
html[data-theme="light"] .pscr__whylabel,
html[data-theme="light"] [data-why="wait"] .pscr__whylabel { color: #5C4E84; }
html[data-theme="light"] .ws__stage:not([data-why="wait"]) .pscr__whylabel { color: var(--ap-teal-text); }  /* the user-approved one-time teal lift (Why SEEK) */
/* Why SEEK: the reason card waits in the app's own plum fill (discoverDesign.ts:429),
   carried on the inset layer that already transitions into the approved teal lift */
html[data-theme="light"] [data-why="wait"] .pscr__why { box-shadow: inset 0 0 0 200px var(--ap-why-fill); }
/* the prompt question is inkSecondary (FullProfileScreen.tsx:409) */
html[data-theme="light"] .pscr__promptq { color: var(--ap-ink2); }


/* ═══════════ HOW SEEK WORKS — REFERENCE 1 (dx-how-seek-reference1) ═══════════
   ONE block for the section; it replaces the four stacked passes before it
   (fixaplan-scale, theme-aware, fixaplan-material and its corrections).
   Three layers, in both themes:
     1. the chapter ground   — the amber -> olive -> dark-green colour field
     2. the presentation plate — the reference's material, measured live on
        fixaplan.com (Chrome 1512x762 + headless 1512x796 / 1440x900): a flat
        black film at 20%, a 1px white hairline at 10%, radius 24, NO backdrop
        blur, no shadow. 80% of the ground passes straight through, so the plate
        reads warm over the amber and green over the green — real see-through,
        nothing painted in.
     3. the SEEK app screen — the app's own opaque canvas (#F4EFEA in light),
        the one solid object on the plate.
   Geometry = the reference's at the same moment (panel top 78 at 1512x796):
   panel 556x580 r24 pad 40; 40 icon / 20 / 24px title / 40 / graphic / 40 /
   16px description; pills 36 (42 active) r8 gap 8; headline 46 on a 43.7 pitch.
   Markup, copy, the pin and page.js §4 are untouched. */
#story .t-story .mask { white-space: nowrap; }
html[data-theme="light"] #story {
  /* the app's light roles reach the screens (the shared #story/#orb block
     forces dark ones; #orb keeps them) */
  --ap-canvas: inherit; --ap-surface: inherit; --ap-surface2: inherit; --ap-floating: inherit;
  --ap-ink: inherit; --ap-ink2: inherit; --ap-muted: inherit;
  --ap-chip-fill: inherit; --ap-chip-border: inherit; --ap-rule: inherit; --ap-veil: inherit;
  --ap-teal: inherit; --ap-teal-text: inherit; --ap-teal-fill: inherit; --ap-teal-border: inherit;
  --ap-teal-glow: inherit; --ap-on-teal: inherit;
  --ap-verify-fill: inherit; --ap-verify-border: inherit; --ap-verify-text: inherit;
  --ap-lav: inherit; --ap-coral: inherit; --ap-on-coral: inherit;
  --ap-why-fill: inherit; --ap-why-border: inherit; --ap-atmos: inherit;
}

/* ---- desktop geometry (both themes) ---- */
@media (min-width: 1000px) {
  #story { --hs-left: 464px; --hs-ph: 580px; }
  #story .story__grid { grid-template-columns: var(--hs-left) minmax(0, 1fr); column-gap: 150px; }
  /* the eyebrow sits on the panel's top line (78) so the headline lands where
     the reference's does; the floor stays at 100 + 80vh, where the CTA ends */
  #story[data-live] .story__pin { top: 78px; }
  #story .story__pin { min-height: calc(80svh + 22px); }
  /* optical weight: at the site's display 300 the serif carried half the
     reference headline's ink (pixel-measured); 400 of the same Newsreader
     brings the left column back to equal authority with the panel */
  #story .t-story { font-size: 46px; line-height: .95; letter-spacing: -.025em; font-weight: 400; }  /* -.025em holds the 400 to the reference's 454px measure */
  /* each .mask carries .14em of clip room above and below and gives it back
     with a negative margin, but two negative margins COLLAPSE to one, so the
     lines sat on a 50px pitch, not 43.7. The second line takes back both. */
  #story .t-story .mask + .mask { margin-top: -.28em; }
  /* 34, not the reference's 40: the serif has no descenders and sits high in
     its line box, so 34 gives the same baseline-to-label gap the eye reads */
  #story .story__index { gap: 8px; margin-top: 34px; }
  #story .story__chip { height: 36px; padding: 0 12px; font-size: 16px; border-radius: 8px; }
  #story .story__chip[data-on="1"] { height: 42px; font-size: 18px; font-weight: 500; }
  #story .story__cta { gap: 12px; }
  #story .story__cta p { font-size: 16px; line-height: 1.2; }

  #story .story__cards { gap: 40px; }
  #story .story__card { min-height: var(--hs-ph); padding: 40px; gap: 40px; justify-content: flex-start; }
  /* icon above title, as in the reference: graphic starts 169 below the top */
  /* SEEK's marks are drawn small inside their 40-unit box (17-24 units of
     ink); at 52 they carry the reference mark's optical size, and the 8 gap
     keeps the head block at 89 (52 + 8 + 29), so nothing below moves */
  #story .card__head { flex-direction: column; align-items: flex-start; gap: 8px; }
  #story .card__ico svg { width: 52px; height: 52px; }
  #story .story__card .t-card { font-size: 24px; line-height: 1.2; letter-spacing: -.02em; }
  #story .story__card > .t-body { font-size: 16px; line-height: 1.2; max-width: 30em; }
  /* 272 x 290: the reference graphic's footprint (300 x 274) at the app's own
     375:400 aspect — 49% of the panel's width, 50% of its height */
  #story .scr { --sw: 272; }
}
@media (min-width: 1440px) {
  #story .story__grid { grid-template-columns: var(--hs-left) 556px; justify-content: space-between; column-gap: 0; }
}
@media (min-width: 1000px) and (max-width: 1199px) {
  #story { --hs-left: 400px; }
  #story .story__grid { column-gap: 64px; }
  #story .t-story { font-size: 40px; }
  #story .scr { --sw: 256; }
}

/* ---- phones / tablets: the existing stack (headline -> steps -> CTA -> panels);
   the steps are a quiet equal-weight list, there is no pin to drive a state ---- */
@media (max-width: 999px) {
  #story .story__index { display: flex; gap: 8px; margin-top: 24px; }
  #story .story__chip,
  #story .story__chip[data-on="1"] { height: 34px; padding: 0 12px; font-size: 15px; border-radius: 8px; color: var(--fg-2); }
  #story .story__cta { margin-top: 28px; gap: 12px; }
  #story .story__card { padding: 24px; gap: 24px; }
  #story .story__card .t-card { font-size: 22px; line-height: 1.2; }
  #story .story__card > .t-body { font-size: 16px; line-height: 1.35; }
}
@media (min-width: 600px) and (max-width: 999px) { #story .scr { --sw: 360; } }

/* ---- light: the plate, the labels, the screen ---- */
html[data-theme="light"] #story .story__card {
  --card-bg: rgba(0,0,0,.2);                 /* the reference's film, verbatim */
  --card-line: rgba(255,255,255,.1);         /* its 1px hairline               */
  --fg: #F5F5F3; --fg-2: rgba(245,245,243,.64); --fg-3: rgba(245,245,243,.56);
  color: #F5F5F3;
  background: var(--card-bg);
  box-shadow: inset 0 0 0 1px var(--card-line);
}
html[data-theme="light"] #story .story__chip { background: rgba(0,0,0,.28); }
html[data-theme="light"] #story .story__chip[data-on="0"] { color: rgba(245,245,243,.70); }  /* >= 4.5:1 on the glow */
/* phones: all four steps at one weight in both themes (outranks the line above,
   which is for the pinned desktop state) */
@media (max-width: 999px) { html #story .story__index .story__chip[data-on] { color: var(--fg-2); } }
/* the screen: opaque app canvas, the app's own hairline, one soft contact shadow */
html[data-theme="light"] #story .scr {
  box-shadow: inset 0 0 0 1px var(--ap-rule), 0 22px 44px -28px rgba(0,0,0,.5);
}
/* LISTENING is p.teal, not the text variant (VoiceAnswerPanel.tsx:195) */
html[data-theme="light"] #story .ui__state { color: var(--ap-teal); }
/* 03/04 are the reflection stack, which the app paints on its light atmosphere
   (seekTokens.ts:471-479); 01/02 (the interview) stay flat */
html[data-theme="light"] #story .ui--know,
html[data-theme="light"] #story .ui--fit {
  background-image:
    radial-gradient(78% 40% at 14% 0%, rgba(124,86,130,.22) 0%, rgba(124,86,130,0) 62%),
    radial-gradient(60% 30% at 94% 10%, rgba(95,91,150,.14) 0%, rgba(95,91,150,0) 64%),
    radial-gradient(46% 24% at 99% 40%, rgba(10,142,119,.085) 0%, rgba(10,142,119,0) 62%),
    radial-gradient(80% 34% at 50% 104%, rgba(180,84,60,.15) 0%, rgba(180,84,60,0) 66%);
}
html[data-theme="light"] #story .ui__why span { color: #5C4E84; }

/* ---- the chapter ground (light), matched to the reference by pixel samples:
   amber glow left-of-centre (~14% / 56%) reaching the left edge, a warm lift at
   the top centre, the right third deep olive-green, a greener floor; 4.5% grain
   so the long ramps never band ---- */
.atmos::after {
  background:
    url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='180' height='180'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='.85' numOctaves='2' stitchTiles='stitch'/%3E%3CfeColorMatrix values='0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 .045 0'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E"),
    radial-gradient(50% 62% at 14% 60%, rgba(150,118,54,.80) 0%, rgba(146,114,53,.66) 26%, rgba(136,106,50,.36) 50%, rgba(128,100,48,.10) 70%, rgba(128,100,48,0) 86%),
    radial-gradient(30% 26% at 36% 80%, rgba(150,118,54,.40) 0%, rgba(150,118,54,0) 80%),   /* the glow's low-right spill */
    radial-gradient(26% 20% at 24% 100%, rgba(128,100,48,.25) 0%, rgba(128,100,48,0) 80%),  /* its warm hump on the floor */
    radial-gradient(42% 32% at 52% 4%, rgba(122,95,52,.70) 0%, rgba(122,95,52,0) 72%),
    radial-gradient(18% 24% at 58% 20%, rgba(122,95,52,.35) 0%, rgba(122,95,52,0) 75%),     /* warmth behind the panel's top-left */
    radial-gradient(40% 36% at 26% 12%, rgba(112,88,48,.34) 0%, rgba(112,88,48,0) 78%),
    radial-gradient(48% 58% at 97% 10%, rgba(46,54,40,.50) 0%, rgba(46,54,40,0) 72%),
    radial-gradient(22% 30% at 100% 55%, rgba(20,30,24,.45) 0%, rgba(20,30,24,0) 80%),      /* the right edge stays green, not olive */
    radial-gradient(75% 55% at 80% 104%, rgba(38,58,38,.62) 0%, rgba(38,58,38,0) 70%),
    linear-gradient(90deg, rgba(10,12,8,0) 46%, rgba(10,12,8,.30) 100%),
    linear-gradient(158deg, #393020 0%, #2A2A1C 48%, #1C2019 100%);
}


/* ═══════════════════════════════════════════ 18. THE WAITLIST SHEET ════════
   THE CHOREOGRAPHY IS THE REFERENCE'S (MOTION-SPEC §5, fixa-interact §1),
   ms from [data-open] / [data-closing]:

     open   scrim    opacity 0 -> 1                         0 +400  E2
            stack    translateY(100% + 30px) scale(.80) -> none   0 +400  E2
            close X  rises 62px out from behind the sheet's top edge,
                     circle -> resting radius             300 +550  EBTN (7.9% overshoot)
            X icon   opacity 0 -> 1                       500 +400  E2
     close  X icon   hidden on the first frame
            close X  sinks back behind the sheet           0 +550  EBTN
            stack    back to translateY(100% + 30px) scale(.80)   130 +400  E2
            scrim    holds, then 1 -> 0                   500 +400  E2 — settle at its end

   SEEK'S DESIGN, NOT THE REFERENCE'S. Only the motion is taken. The sheet is
   SEEK's own waitlist lockup — the --deep card, --r-card, the .foot__form
   field and the .btn--light submit that the footer already ships — and the
   scrim keeps SEEK's plum hue at the reference's .20 dim and 4px blur. */
:root {
  --ease-sheet: var(--e2);
  --ease-sheet-btn: var(--ebtn);   /* overshoots */
}

/* THE SCROLL LOCK IS A CLASS WITH !important, AND THAT IS NOT DECORATION.
   MEASURED on this candidate at 1440x900: the runtime's inline overflow:hidden
   was applied at t+12ms and WIPED at t+78ms — Lenis owns documentElement's
   inline overflow (it writes "clip" when it stops and removeProperty()s it when
   it starts) and this page builds it with autoToggle:true, so it rewrites that
   property on its own schedule for as long as the page is open. An inline lock
   is therefore guaranteed to be erased mid-modal, and the page behind the sheet
   becomes scrollable again. It only ever appeared on desktop, because Lenis is
   not constructed at all on a touch-primary device (section 0b).
   A normal inline declaration loses to an author !important, so this class wins
   for exactly as long as it is set and hands the property straight back to
   Lenis — at whatever value Lenis currently wants — the moment it is removed. */
html.wlm-lock { overflow: hidden !important; }

.wlm {
  /* the UA sheet gives <dialog> width/height: fit-content and margin: auto;
     all of that is replaced so the element is the viewport and the sheet is
     positioned inside it by flow. 100dvh, not 100svh or 100%: this is the ONE
     place on the page that wants the viewport as it is RIGHT NOW, and while
     the sheet is open the scroll is locked, so Safari cannot resize the bar
     underneath it and dvh cannot move. */
  position: fixed; inset: 0; margin: 0; padding: 0; border: 0;
  width: 100%; max-width: none; height: 100dvh; max-height: none;
  background: none; overflow: visible;
  display: flex; align-items: flex-end; justify-content: center;
}
.wlm:not([open]) { display: none; }
/* the UA's own backdrop is left transparent — the scrim below is a real
   element, so its blur, its dim and its timing are all ours and none of it
   depends on ::backdrop being animatable in a given engine. */
.wlm::backdrop { background: none; }

/* THE TINT AND THE BLUR ARE CONSTANT; only the layer's opacity moves, so the
   dim and the perceived blur ramp together, as the reference's do. */
.wlm__scrim {
  position: fixed; inset: 0;
  background: rgba(12, 9, 20, .20);
  -webkit-backdrop-filter: blur(4px);
          backdrop-filter: blur(4px);
  opacity: 0;
}
.wlm[data-open] .wlm__scrim { opacity: 1; transition: opacity .4s var(--ease-sheet); }

/* SAFARI MUST NOT MISTAKE THE SHEET FOR A SOURCE OF BAR COLOUR (5 Oct 2026).
   iOS Safari colours its status-bar strip and toolbar from a hit test at the
   middle of each screen edge (WebKit LocalFrameView::fixedContainerEdges). A
   fixed box found there that is a "dimming layer" - translucent background,
   no children - or 0.9-1.05x the viewport with z-index >= 0 replaces the
   blurred copy of the page with a flat colour, in ONE frame. MEASURED on the
   owner's iPhone and in the iOS 27 Simulator: the scrim (top edge) and the
   dialog box (bottom edge) qualified the moment the fade-in had run, and both
   bars snapped. So neither may qualify:
   - the scrim gets an empty child, so it is no longer a dimming layer, and
     z-index -1, so a viewport-sized box is ignored; it still paints first in
     the dialog, under the stack, exactly where it did;
   - the dialog starts 12vh above the screen, 1.12x tall, "too large" to count;
     it has no background and the stack stays on the same bottom edge.
   With nothing at either edge Safari keeps the page copy throughout. On screen
   the sheet, the dim and the blur are pixel-identical; desktop included. */
.wlm { top: -12vh; height: calc(100dvh + 12vh); }
.wlm__scrim { z-index: -1; }
.wlm__scrim::before { content: ""; }


.wlm__stack {
  position: relative;
  width: 100%; max-width: 520px;
  padding: 0 16px max(28px, calc(env(safe-area-inset-bottom) + 12px));
  /* the close control rides inside the stack, so the card and the control
     arrive as ONE object; 100% is the card plus its bottom strip, so +30px
     starts the card fully below the fold */
  transform: translate3d(0, calc(100% + 30px), 0) scale(.80);
  will-change: transform;   /* exactly one element, only while it is mounted */
  /* THE GUTTERS ARE BACKDROP, AND MUST BEHAVE LIKE IT. This box is capped at
     520px and carries 16px of side padding and a bottom strip, so on a wide
     phone — and on every desktop — a tap in those bands hit the STACK, not the
     scrim, and the sheet did not close even though the reader had plainly
     tapped outside it. pointer-events is hit-testing only: it moves nothing,
     and the press-and-release identity test in the runtime still works because
     the events now land on the scrim, which is what they look like they hit. */
  pointer-events: none;
}
.wlm__x, .wlm__sheet { pointer-events: auto; }
.wlm[data-open] .wlm__stack { transform: translate3d(0, 0, 0) scale(1); transition: transform .4s var(--ease-sheet); }

/* THE CLOSE CONTROL LIVES BEHIND THE CARD. Its resting place is unchanged —
   44x44, centred, 16px clear of the card's top edge — but it is out of flow
   (so the stack's 100% is the card) and at z-index -1 inside the stack's own
   stacking context, i.e. painted under the card. Resting translate 62px puts
   it wholly behind the card's top edge; it rises out, circle to
   rounded-square, and sinks back on the way out. */
.wlm__x {
  position: absolute; z-index: -1; top: -60px; left: 50%; margin-left: -22px;
  display: flex; align-items: center; justify-content: center;
  width: 44px; height: 44px;
  border: 0; border-radius: 50%; cursor: pointer;
  background: #FFF; color: #16111F;
  translate: 0 62px;
}
.wlm__x svg { opacity: 0; }
.wlm[data-open] .wlm__x {
  translate: none; border-radius: var(--r-pill);
  transition: translate .55s var(--ease-sheet-btn) .3s, border-radius .55s var(--ease-sheet-btn) .3s;
}
.wlm[data-open] .wlm__x svg { opacity: 1; transition: opacity .4s var(--ease-sheet) .5s; }

.wlm__sheet {
  background: var(--deep); color: var(--deep-fg);
  --fg: var(--deep-fg); --fg-2: var(--deep-fg2); --fg-3: var(--deep-fg3);
  --tint: var(--deep-tint); --line: var(--deep-line);
  border-radius: var(--r-card);
  padding: clamp(24px, 3.3vw, 32px);
  /* a sheet taller than the phone still has to be operable; contain stops the
     overscroll chaining to the locked document underneath */
  max-height: calc(100dvh - 120px);
  overflow-y: auto; overscroll-behavior: contain;
  -webkit-overflow-scrolling: touch;
}
.wlm__h { margin-top: 20px; }
.wlm .foot__form { max-width: none; }

/* CLOSING IS STAGED, NOT REVERSED. [data-open] comes off and [data-closing]
   goes on in the same style pass, so these are the transitions the exit runs
   (a transition takes its timing from the after-change style): the icon goes
   at once, the control sinks, the card follows 130ms later and the scrim
   holds until 500ms. The runtime settles on the scrim's transitionend. */
.wlm[data-closing] .wlm__x { transition: translate .55s var(--ease-sheet-btn), border-radius .55s var(--ease-sheet-btn); }
.wlm[data-closing] .wlm__stack { transition: transform .4s var(--ease-sheet) .13s; }
.wlm[data-closing] .wlm__scrim { transition: opacity .4s var(--ease-sheet) .5s; }

@media (prefers-reduced-motion: reduce) {
  .wlm[data-open] .wlm__scrim, .wlm[data-open] .wlm__stack, .wlm[data-open] .wlm__x, .wlm[data-open] .wlm__x svg,
  .wlm[data-closing] .wlm__scrim, .wlm[data-closing] .wlm__stack, .wlm[data-closing] .wlm__x {
    transition: none;
  }
  /* no travel, no scale, no rise: the surface and its control are simply there */
  .wlm__stack { transform: none; }
  .wlm__x { translate: none; border-radius: var(--r-pill); }
  .wlm__x svg { opacity: 1; }
}

/* ═══════════════ THE SHEET, REDESIGNED (5 Oct 2026) ═══════════════════════════
   Cleaner, calmer, more legible: a brand row (mark + name, "Private beta"), one
   Newsreader line that says what this is, a sentence on early access, a
   LABELLED field, a full-width button, one quiet reassurance. The motion, the
   close control's rise and every timing above are unchanged.

   IT FOLLOWS THE THEME. Light: a white card - the page's own chrome white - with
   the page's ink at its three strengths. Dark: the --deep card it always was,
   with a hairline so it reads against the dimmed page. Everything inside is
   drawn from the per-card tokens below, so the two themes cannot drift apart. */
.wlm__sheet {
  --wlm-ink: var(--deep-fg); --wlm-ink2: var(--deep-fg2); --wlm-ink3: var(--deep-fg3);
  --wlm-field: rgba(255,255,255,.06); --wlm-field-line: rgba(255,255,255,.10);
  --wlm-cta: #FFFFFF; --wlm-cta-ink: #16111F; --wlm-tag: rgba(255,255,255,.07);
  border-radius: 28px;
  padding: 26px 24px 24px;
  box-shadow: inset 0 0 0 1px rgba(255,255,255,.07);
}
html[data-theme="light"] .wlm__sheet {
  --wlm-ink: #111111; --wlm-ink2: rgba(17,17,17,.64); --wlm-ink3: rgba(17,17,17,.56);
  --wlm-field: #F2F2F0; --wlm-field-line: transparent;
  --wlm-cta: #111111; --wlm-cta-ink: #FFFFFF; --wlm-tag: rgba(17,17,17,.055);
  background: #FFFFFF; color: #111111;
  /* a white card on a lightly dimmed light page needs an edge of its own */
  box-shadow: 0 24px 60px -28px rgba(17,17,17,.45), 0 0 0 1px rgba(17,17,17,.04);
}
@media (min-width: 600px) { .wlm__sheet { padding: 32px 32px 28px; } }

.wlm__top { display: flex; align-items: center; justify-content: space-between; gap: 12px; }
.wlm__brand {
  display: inline-flex; align-items: center; gap: 8px;
  font-family: var(--f-ui); font-size: 16px; font-weight: 600; letter-spacing: -.01em;
  color: var(--wlm-ink);
}
.wlm__brand img { width: 30px; height: auto; }
.wlm__tag {
  display: inline-flex; align-items: center; gap: 6px;
  padding: 5px 10px 5px 9px; border-radius: 999px; background: var(--wlm-tag);
  font-family: var(--f-ui); font-size: 12px; font-weight: 500; letter-spacing: .01em;
  color: var(--wlm-ink2); white-space: nowrap;
}
.wlm__tag::before { content: ""; width: 6px; height: 6px; border-radius: 50%; background: var(--accent); }

.wlm__h {
  margin: 28px 0 0;
  font-family: var(--f-display); font-weight: 400;
  font-size: 32px; line-height: 1.08; letter-spacing: -.025em;
  color: var(--wlm-ink); text-wrap: balance;
}
.wlm__sub {
  margin: 12px 0 0; max-width: 40ch;
  font-family: var(--f-ui); font-size: 16px; line-height: 1.45; letter-spacing: -.01em;
  color: var(--wlm-ink2); text-wrap: pretty;
}
@media (min-width: 600px) { .wlm__h { font-size: 36px; } }

.wlm__form { display: flex; flex-direction: column; margin-top: 26px; }
.wlm__label {
  margin: 0 0 8px 2px;
  font-family: var(--f-ui); font-size: 13px; font-weight: 500; letter-spacing: -.005em;
  color: var(--wlm-ink2);
}
.wlm__input {
  height: 54px; padding: 0 18px; border: 0; border-radius: 14px;
  background: var(--wlm-field); color: var(--wlm-ink);
  box-shadow: inset 0 0 0 1px var(--wlm-field-line);
  font-family: var(--f-ui); font-size: 16px; letter-spacing: -.01em;   /* < 16px makes iOS Safari zoom on focus */
  -webkit-appearance: none; appearance: none;
  transition: box-shadow var(--dur-fast) var(--ease-enter);
}
.wlm__input::placeholder { color: var(--wlm-ink3); }
.wlm__input:focus { outline: none; box-shadow: inset 0 0 0 1.5px var(--accent); }
.wlm__input[aria-invalid="true"] { box-shadow: inset 0 0 0 1.5px #E5735F; }
.wlm__cta {
  margin-top: 12px; height: 54px; border: 0; border-radius: 14px; cursor: pointer;
  background: var(--wlm-cta); color: var(--wlm-cta-ink);
  font-family: var(--f-ui); font-size: 16px; font-weight: 600; letter-spacing: -.01em;
  transition: transform var(--dur-fast) var(--ease-enter), opacity var(--dur-fast) var(--ease-enter);
}
.wlm__cta:hover { transform: translateY(-1px); }
.wlm__cta:disabled { opacity: .6; cursor: default; transform: none; }
.wlm__note {
  margin: 14px 0 0; text-align: center;
  font-family: var(--f-ui); font-size: 13px; line-height: 1.4; letter-spacing: -.005em;
  color: var(--wlm-ink3); text-wrap: pretty;
}

/* the close control keeps its rise; it settles to the card's softer corner */
.wlm[data-open] .wlm__x { border-radius: 14px; }
html[data-theme="light"] .wlm__x { box-shadow: 0 8px 22px -10px rgba(17,17,17,.4); }

/* ═══════════════ FINAL PERFORMANCE PASS (24 Sep) — TOUCH ONLY ════════════════
   Why SEEK keeps its full read-down: the profile still scrolls through the
   phone on one compositor scrub (page.js §3b, unchanged). What changes is the
   work the compositor is asked to do around it on a phone. Desktop (hover +
   fine pointer) matches none of this.

   THE REASON CARD's ONE-TIME TEAL LIFT used to repaint the WHOLE profile
      layer — the animated .pscr__frame, a 31MB backing store — on every frame
      of its 1.4s colour transition (border, inset tint, label: paint
      properties), and it runs exactly while the reader is scrolling through the
      section (LayerTree, Simulator Safari 27: 93 paints on that layer in one
      visit). Given its own small layer, the lift repaints only the card; the
      card is a sublayer of the scrubbed layer, so it rides the same compositor
      animation and cannot drift from it. Same pixels, same 1.4s, same curve.

   THE FIXED ATMOSPHERE, ONLY WHERE IT CAN BE SEEN. The smooth candidate hid
   it on touch outright and stretched #story's copy under #orb's corners;
   re-tested from mobile-polish, a rubber-band past the footer then showed the
   flat canvas instead of the field, and Safari's bars (which copy the in-flow
   page, never a fixed layer) picked up the stretched copy at #orb's corners.
   Here nothing about the copy changes: the fixed layer is only put to rest
   (html.atmos-rest, from page.js §4c) while neither #orb's corners nor the
   end of the page is near the fold. */
@media (hover: none) and (pointer: coarse) {
  .ws .pscr__why { will-change: transform; }
  html.atmos-rest body > .atmos { visibility: hidden; }
}

/* ═══════════ FINAL PERFORMANCE PASS — SEE IT IN ACTION: THE STATE CHANGE ═════
   Was: the outgoing composition CUT in one frame and the incoming one faded in
   (700ms on phones, 550ms wider) on an in-out curve that barely moves for its
   first frames — so a tap read as an instant swap; the phone caption swapped
   between two frames and the pager set off at once, ahead of the phone.
   Now ONE event of ~600ms, the same at every width (the idea is the smooth
   candidate's, re-timed to the brief):
     leaving   opacity 1 -> 0 over 280ms, a 10px drift the way the reader is
               going, scale 1 -> .99, standard curve                0 -> 300ms
     arriving  from 12px on the other side and .99, opacity 0 -> 1, a soft
               ease-out, 120ms after the leaver starts               120 -> 600ms
     caption   out 160ms, in 360ms from 200ms (phones)                0 -> 560ms
     pager     500ms from 100ms, so the pill lands with the phone    100 -> 600ms
   The two stages overlap from 120 to 280ms, so there is never a blank frame.
   Transform + opacity only, on the stage layers the state already swaps; idle
   stages keep visibility:hidden (no backing store). [data-leave] is set by
   page.js for the leaver's 320ms and nothing else; [data-orb-dir] says which
   way the reader stepped. The first paint is untouched. */
.orb__card[data-orb-live] .stage {                       /* idle = entry pose */
  translate: 12px 0; scale: .99;
}
.orb__card[data-orb-live][data-orb-dir="prev"] .stage { translate: -12px 0; }
.orb__card[data-orb-live] .stage[data-leave] {           /* on its way out */
  visibility: visible; opacity: 0; z-index: 0;
  translate: -10px 0; scale: .99;
  transition: opacity 280ms cubic-bezier(.4, 0, .2, 1),
              translate 300ms cubic-bezier(.4, 0, .2, 1),
              scale 300ms cubic-bezier(.4, 0, .2, 1);
}
.orb__card[data-orb-live][data-orb-dir="prev"] .stage[data-leave] { translate: 10px 0; }
.orb__card[data-orb-live][data-orb-state="1"] .stage--1,
.orb__card[data-orb-live][data-orb-state="2"] .stage--2,
.orb__card[data-orb-live][data-orb-state="3"] .stage--3 {
  visibility: visible; opacity: 1; translate: none; scale: none; z-index: 1;
  transition: opacity 480ms cubic-bezier(.22, .61, .36, 1) 120ms,
              translate 480ms cubic-bezier(.22, .61, .36, 1) 120ms,
              scale 480ms cubic-bezier(.22, .61, .36, 1) 120ms;
}
.orb__card[data-orb-live] .dev__pager i,
.orb__card[data-orb-live][data-orb-state="1"] .dev__pager i:nth-child(1),
.orb__card[data-orb-live][data-orb-state="2"] .dev__pager i:nth-child(2),
.orb__card[data-orb-live][data-orb-state="3"] .dev__pager i:nth-child(3) {
  transition: width 500ms var(--rows-nobounce) 100ms,
              background-color 400ms var(--e2) 100ms;
}
@media (max-width: 759.98px) {
  .orb__card[data-orb-live] .orb__row {
    opacity: 0;
    transition: opacity 160ms cubic-bezier(.4, 0, 1, 1), visibility 0s linear 160ms;
  }
  .orb__card[data-orb-live][data-orb-state="1"] .orb__row:nth-child(1),
  .orb__card[data-orb-live][data-orb-state="2"] .orb__row:nth-child(2),
  .orb__card[data-orb-live][data-orb-state="3"] .orb__row:nth-child(3) {
    opacity: 1;
    transition: opacity 360ms cubic-bezier(.22, .61, .36, 1) 200ms, visibility 0s;
  }
}
/* reduced motion: the page floor makes every duration .001ms; the delays go too */
@media (prefers-reduced-motion: reduce) {
  .orb__card[data-orb-live] .stage,
  .orb__card[data-orb-live] .orb__row,
  .orb__card[data-orb-live] .dev__pager i { transition-delay: 0s !important; }
}

/* ════════════ FINAL MOTION PASS (25 Sep) — WHY SEEK RE-ENTRY, TOUCH ONLY ══════════
   HOW the graphic appears, never WHERE it is: the lanes and the profile inside
   the phone stay on their scroll scrub (page.js §3b), untouched. This is the
   frame as one object — opacity and a 12px settle, transform/opacity only, so
   it runs on the compositor. page.js writes [data-away] only once the frame is
   wholly off screen by 15% of a viewport, and removes it once the frame is
   within 30% of a viewport of the fold again: at a normal pace the settle has
   finished before the frame is on screen; after a fast fling back it is the
   visible bridge — the late repaint of the profile layer happens under the
   fade instead of popping in. Transitions retarget from the current value, so
   a reversal mid-way never jumps or queues. Desktop (hover + fine pointer) and
   reduced motion never get the attribute. */
@media (hover: none) and (pointer: coarse) {
  /* opacity on the standard in-out curve (a front-loaded ease-out reads as a
     pop: measured 50% opacity 30ms in), the settle on a decelerating one */
  .ws__frame {
    transition:
      opacity 260ms cubic-bezier(.4,0,.2,1),
      translate 300ms cubic-bezier(.22,.61,.36,1);
  }
  .ws__frame[data-away] {
    opacity: 0; translate: 0 12px;
    transition:
      opacity 160ms cubic-bezier(.4,0,1,1),
      translate 160ms cubic-bezier(.4,0,1,1);
  }
}
