/* Static-export fixes. Loaded AFTER gs-site-lab.css, which stays byte-identical
   to Lovable's file — keeping corrections here means the export can be replaced
   wholesale without losing them, and the diff stays readable.

   ---------------------------------------------------------------------------
   1. Body background does not follow the theme
   ---------------------------------------------------------------------------
   gs-site-lab.css:65 sets `body { background-color: var(--color-background) }`,
   and --background is the Tailwind/shadcn base token: oklch(97.2% .003 258),
   a near-white (line 1035). Its dark counterpart on line 1036 is gated on the
   shadcn `.dark` class, which React used to set and the static pages do not —
   they theme with [data-theme] instead. So in dark mode every page keeps a
   light body behind the .gsh wrapper, visible as a pale band around the header
   and down the page edges.

   .gsh itself is correct (#07080D dark / #F7F3EE light); it is only the body
   underneath that never switched. The ground values are written literally here
   because body sits outside .gsh and cannot read --gs-ground. */

html:not([data-theme="dark"]) body {
  background-color: #F7F3EE;
}
html[data-theme="dark"] body {
  background-color: #07080D;
}

/* Belt and braces for the first paint: the inline bootstrap sets
   data-theme before the stylesheet lands, but a viewer with no stored
   preference gets no attribute at all, so the light rule above has to be the
   one that applies by default — which it does, via :not(). */

/* ---------------------------------------------------------------------------
   2. About hero tab wordmark uses themed tokens instead of lockup colours
   ---------------------------------------------------------------------------
   The small "growthsanity" inside the About hero's browser-tab graphic takes
   --wm-g/--wm-i from var(--gs-green)/var(--gs-indigo). In dark mode those
   resolve through oklch() to #4BC680 and #8498EF — 22 and 17 levels off the
   real wordmark colours, so the mark in the tab graphic does not match the
   lockup in the header two inches above it.

   Measured in this export, dark mode:
     header .wm  #35BD6B   (correct)
     .ab-wm      #4BC680   (wrong)

   The wordmark is brand artwork, not themed UI: it takes the lockup's two
   colours and nothing else. Do not re-tokenise these.

   This was introduced in the About rewrite and corrected in the source CSS we
   hold, but that correction has not reached Lovable's project yet — so it is
   patched here as well until the next export carries it. */

.gsh .ab-tab {
  --wm-g: #173D24;
  --wm-i: #485289;
}
.gsh[data-theme="dark"] .ab-tab {
  --wm-g: #35BD6B;
  --wm-i: #7D8BDE;
}

/* ---------------------------------------------------------------------------
   3. Closing section: split the two bands, and make the status strip readable
   ---------------------------------------------------------------------------
   DESIGN CHANGE (Kyle, 29 Oct 2026). Supersedes the single-card treatment.

   The bench and the Acquret note were merged into one card with a hairline
   between them, to stop the page bottom reading as banner after banner. Seen
   rendered, the hairline does the opposite: two full-width bands sharing one
   container read as one thing awkwardly divided rather than two related
   things. They are now separate cards with real space between.

   The hierarchy that fixed the original problem stays: the bench keeps the
   full-size headline, the Acquret note stays demoted at 23px. Splitting the
   container does not make them two competing banners again, because they were
   never the same weight.

   The status strip set a full sentence in uppercase mono with wide tracking,
   which is slow to read and against the type system — JetBrains Mono is for
   kickers, IDs and timestamps, never for reading. Only the label stays mono
   now; the sentence and links move to the UI and heading faces in sentence
   case.
   ========================================================================= */

/* --- split the card into two --- */
.gsh[data-lab] .lab-close-card {
  border: 0;
  background: none;
  border-radius: 0;
  overflow: visible;
  display: grid;
  gap: 18px;
}

.gsh[data-lab] .lab-close-band {
  border: 1px solid var(--gs-rule);
  border-radius: 20px;
  background: var(--card-b);
  overflow: clip;
}
/* REMOVED 19 Nov. This pinned /about's closing band to --card-a (the stark
   white) while every other page took --card-b from the rule above, so the
   card read as a different component on the one page that is about the
   company. /about now inherits --card-b like the homepage and /shelf.

   Removing it was not enough, and the leftover is worth recording. The white
   that survived was the CARD behind the band, not the band: the split-card
   rule at the top of this section sets `.lab-close-card { background: none }`
   at (0,3,0) — the bands carry the colour and the card is transparent — but
   `gs-site-lab.css:4162` pins `.gsh[data-lab].ab .lab-close-card` to
   `--card-a` at (0,4,0) and wins. It showed only where the band's 20px radius
   curves away from the card's corner, which is why it read as four small
   white wedges rather than a white card.

   The locked file cannot be edited, so this ties that selector at (0,4,0) and
   wins on load order. `none`, not `--card-b`: the card is transparent
   everywhere else and the bands do the painting. */
.gsh[data-lab].ab .lab-close-card { background: none; }
/* the hairline that used to divide them is gone with the shared container */
.gsh[data-lab] .lab-close-band + .lab-close-band {
  border-top: 1px solid var(--gs-rule);
}

/* --- status strip: its own quiet row, outside both cards --- */
.gsh[data-lab] .lab-close-strip {
  display: grid;
  grid-template-columns: auto minmax(0, 1fr) auto;
  align-items: center;
  gap: 8px 22px;
  margin-top: 4px;
  padding: 14px 4px 0;
  border-top: 1px solid var(--gs-rule);
  background: none;
}
.gsh[data-lab].ab .lab-close-strip {
  background: none;
}

/* label stays mono — it is a label */
.gsh[data-lab] .lab-close-status {
  font-family: var(--gs-mono);
  font-size: 10px;
  letter-spacing: 0.12em;
  text-transform: uppercase;
  white-space: nowrap;
}

/* sentence and links become readable type */
.gsh[data-lab] .lab-close-ask {
  font: 400 13.5px/1.5 var(--gs-ui);
  letter-spacing: 0;
  text-transform: none;
  color: var(--gs-ink-2);
}
.gsh[data-lab] .lab-close-links {
  display: inline-flex;
  gap: 18px;
  margin-left: 0;
  white-space: nowrap;
}
.gsh[data-lab] .lab-close-links a {
  font: 700 13px/1.3 var(--gs-head);
  letter-spacing: 0;
  text-transform: none;
}

@media (max-width: 900px) {
  .gsh[data-lab] .lab-close-strip {
    grid-template-columns: minmax(0, 1fr);
    gap: 10px;
  }
  .gsh[data-lab] .lab-close-links {
    white-space: normal;
    flex-wrap: wrap;
  }
}

/* ---------------------------------------------------------------------------
   4. Bench pending note: same readability fix as the status strip
   ---------------------------------------------------------------------------
   DESIGN CHANGE (Kyle, 29 Oct 2026). Follows §3.

   "Email signup opens at launch — your pick isn't submitted yet." was still set
   in uppercase JetBrains Mono with wide tracking, like the strip sentence was
   before §3. Two sentences in the same section, one fixed and one not, reads as
   an oversight. Mono stays for kickers, IDs and timestamps; this is a sentence,
   so it takes the UI face in sentence case.

   The markup keeps its `mono` class (it comes from the exported component), so
   the override is by class here rather than a markup change — same reason as
   §3: no drift between the export and the Lovable source.
   ========================================================================= */

.gsh[data-lab] .lab-close-pending {
  font-family: var(--gs-ui);
  font-size: 12.5px;
  line-height: 1.5;
  letter-spacing: 0;
  text-transform: none;
  color: var(--gs-ink-3);
}

/* ---------------------------------------------------------------------------
   5. CTA typeface: one face for every pill CTA on the site
   ---------------------------------------------------------------------------
   DESIGN FIX (Kyle, 29 Oct 2026). The locked type system gives Helvetica Neue
   bold to headings AND all pill CTAs; Space Grotesk is UI, body and metrics.

   Pages whose body sits inside .gsh already follow it — gs-lab.css's
   `.gsh .btn` wins there. The legacy page bodies do not, because they predate
   the rule and site.css still carries the old system:

     site.css   `.btn`                        font-family: var(--gs-ui)   600
     design.css `.rd main .btn, .rd main .btn-dl`  font-family: var(--gs-ui)   600

   Ten CTAs were affected:
     /demand-gen-roi-calculator  "See my snapshot"
     /linkedin-post-auditor      "Audit my post"
     /tools                      Browse the tools, Visit acquret.com,
                                 Open the calculator, Audit a post,
                                 Book a consultation
     /404                        Browse the tools, Demand Gen ROI Calculator,
                                 LinkedIn Post Auditor

   Specificity is matched deliberately, not inflated: this file loads after both
   stylesheets, so an equal-specificity rule wins on order. No !important.
   `--gs-head` is declared at :root in site.css, so it resolves outside .gsh.

   Size is left alone. A 15px page CTA next to a 14px header CTA is hierarchy,
   not inconsistency — only the typeface was wrong.

   NOT changed, because they are not CTAs: the step pills (1 Inputs / 2 Snapshot
   …), the /shelf jump chips, and the Pipeline value / Closed revenue toggles.
   Those are labels and tabs, and JetBrains Mono / Space Grotesk is correct for
   them. Turning seven jump chips into Helvetica bold would read as seven
   competing CTAs.
   ========================================================================= */

.btn,
.btn-dl,
.nf-btn {
  font-family: var(--gs-head);
  font-weight: 700;
  letter-spacing: -0.01em;
}

/* matches design.css's own selector so it wins on load order */
.rd main .btn,
.rd main .btn-dl {
  font-family: var(--gs-head);
  font-weight: 700;
  letter-spacing: -0.01em;
}

/* 404 only: its own <style> block sits AFTER this file in the head, so an
   equal-specificity rule loses on order. `a.nf-btn` / `button.nf-btn` is one
   notch higher (0,1,1 vs 0,1,0) — the minimum needed to win, rather than
   !important. The inline rule is:
       .nf-btn { font-family: var(--gs-ui); font-size: 13.5px; font-weight: 600 }
   Its 13.5px size is kept; only the face and weight change. */
a.nf-btn,
button.nf-btn {
  font-family: var(--gs-head);
  font-weight: 700;
  letter-spacing: -0.01em;
}

/* ---------------------------------------------------------------------------
   6. Footer: one treatment site-wide
   ---------------------------------------------------------------------------
   DESIGN FIX (Kyle, 29 Oct 2026). The footer must look the same everywhere.

   `.lft` ships its own palette in gs-lab.css — a white panel in light, #0B0E1C
   in dark — so it lifts off the page ground. The fold-cadence block in
   gs-site-lab.css (line 4138) then overrides `--lft-bg` to `var(--fold-a)` on
   .gsh[data-lab] pages, which flattens the footer into the body:

     homepage, About, Shelf, Playlists, Films   #F7F3EE on #F7F3EE  (flat)
     tools, games, both tool pages              #FFFFFF on #F7F3EE  (lifted)

   The lifted version is the standard. This restores the native values on the
   pages the cadence had flattened; the pages that already looked right are
   unchanged, because these are the values they were already using.

   `.mq-shelf` is deliberately NOT included. The marquee sits above the footer
   and belongs to the page ground — the linen strip between content and footer
   is what gives the white panel its edge. */

.gsh[data-lab] .lft {
  --lft-bg: #FFFFFF;
  background: var(--lft-bg);
}
.gsh[data-lab][data-theme="dark"] .lft {
  --lft-bg: #0B0E1C;
  background: var(--lft-bg);
}

/* ---------------------------------------------------------------------------
   7. Quick Q (/quick-q)
   ---------------------------------------------------------------------------
   New page, 30 Oct. Built on the /shelf shell, so header, footer and tokens
   come free; only the page's own blocks are here.

   THE POINT OF THE CARD IS THE TYPE. Four faces, four jobs, in reading order:

     mechanical top    JetBrains Mono   dispatch no., time, persona pill
     the incoming ping Helvetica Neue   bold, 19px — hits at a glance
     the decode        Libre Caslon     italic lead-in, the human voice...
                       Space Grotesk    ...then the explanation
     the reply         Space Grotesk    calm, readable
     buttons           Helvetica Neue   bold, per §4
     watermark         JetBrains Mono   survives a screenshot

   That sequence is the retro-modern feel: cold machine header, loud incoming
   message, warm editorial turn, pragmatic answer.

   Caslon is used here as an editorial lead-in, not as a headline accent. It is
   the one place on the page the brand's voice speaks rather than reports.
   ========================================================================= */

.gsh[data-lab] .qq-sec { padding-block: 56px; background: var(--fold-a); }

/* --- page head ---
   Corrected 6 Nov (§44). This rule previously read:

     text-align: left; align-items: start; gap: 10px; margin: 0 0 26px;

   which was a half-done override and the cause of the ragged heads Lovable
   flagged. `.qq-head` / `.dt-head` are also `.sec-head`, and the locked sheet
   sets BOTH axes there (line 1242): `text-align: center; justify-items:
   center`. Flipping `text-align` to left while leaving `justify-items:
   center` standing meant every child shrink-wrapped, centred itself on the
   720px track, then ran its text left from wherever it landed — three
   different left edges at 1280 (kicker 343, headline 70, subhead 125) all
   sharing one centre at 430. `align-items` was the wrong property: it is the
   block axis. The inline axis is `justify-items`.

   The locked sheet already has the correct way to left-align a head — the
   `.sec-head.left` modifier at line 1469 sets `text-align: left` AND
   `justify-items: start` AND `margin-left: 0` together. Had left been wanted,
   that class in the HTML was the route, not this.

   Centred is also the house pattern: /shelf, /games and /tools all run their
   heads centred. These two were the only outliers, and only in this one
   property. So the fix is to stop declaring it and let line 1242 govern.
   The gap and bottom margin are kept — those were deliberate spacing — but
   the margin regains its `auto` inline sides so the 720px block itself
   centres in the column instead of sitting hard left. */
.gsh[data-lab] :is(.qq-head, .dt-head) { gap: 10px; margin: 0 auto 26px; }
.gsh[data-lab] :is(.qq-kick, .dt-kick) {
  color: var(--gs-green-active);
  font: 500 11px/1 var(--gs-mono); letter-spacing: .12em; text-transform: uppercase;
}
.gsh[data-lab][data-theme="dark"] :is(.qq-kick, .dt-kick) { color: var(--gs-green); }
.gsh[data-lab] :is(.qq-kick, .dt-kick) b { font-weight: 700; }
.gsh[data-lab] :is(.qq-h1, .dt-h1) {
  margin: 0; color: var(--gs-ink);
  font: 700 clamp(34px, 5vw, 56px)/1.05 var(--gs-head); letter-spacing: -.035em;
}
/* `balance` rather than `pretty`: the sub is two lines at desktop width and
   the complaint was a one-word orphan on line two, which is a line-LENGTH
   problem. (Not the same call as the /about headline in §42, where the ask
   was one sentence per line and balance actively fought it.) */
.gsh[data-lab] :is(.qq-sub, .dt-sub) { margin: 0; max-width: 56ch; font: 400 17px/1.6 var(--gs-ui); color: var(--gs-ink-2); text-wrap: balance; }

/* --- filter tabs --- */
.gsh[data-lab] .qq-tabs { display: flex; flex-wrap: wrap; gap: 8px; margin: 0 0 22px; }
.gsh[data-lab] .qq-tab { gap: 6px;
  padding: 8px 15px; border: 1px solid var(--gs-rule); border-radius: 999px;
  background: var(--card-a); color: var(--gs-ink-2);
  font: 700 13px/1 var(--gs-head); letter-spacing: -.01em; cursor: pointer;
  transition: border-color .2s ease, color .2s ease, background .2s ease;
}
.gsh[data-lab] .qq-tab i { font: 400 11px/1 var(--gs-ui); font-style: normal; }
.gsh[data-lab] .qq-tab:hover { border-color: var(--gs-rule-2); }
.gsh[data-lab] .qq-tab.is-on { background: var(--gs-ink); border-color: var(--gs-ink); color: var(--gs-ground); }
.gsh[data-lab] .qq-tab.is-on i { color: var(--gs-ground); opacity: .7; }
.gsh[data-lab] .qq-tab:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }

/* --- card --- */
.gsh[data-lab] .qq-grid {
  /* Two columns, not auto-fit. With four cards, auto-fit computed three
     369px tracks at desktop width and stranded card #004 alone on row 2
     between two empty slots. A fixed 2x2 gives all four personas equal
     weight and widens each card to ~490px, which the telex layout needs —
     at 369px the cards ran 677px tall and read as mobile strips. */
  display: grid; gap: 16px;
  grid-template-columns: repeat(2, 1fr);
  max-width: 1040px; margin-inline: auto;
  align-items: start;
}
@media (width <= 768px) {
  .gsh[data-lab] .qq-grid { grid-template-columns: 1fr; }
}
.gsh[data-lab] .qq-card {
  display: grid; gap: 14px; padding: clamp(18px, 2.4vw, 22px);
  border: 1px solid var(--gs-rule); border-left: 3px solid var(--pc, var(--gs-rule));
  border-radius: 20px; background: var(--card-a);
  box-shadow: inset 0 1px 0 rgba(255,255,255,.6);
}
.gsh[data-lab][data-theme="dark"] .qq-card { box-shadow: inset 0 1px 0 rgba(255,255,255,.05); }
.gsh[data-lab] .qq-card[hidden] { display: none; }

.gsh[data-lab] .p-felix { --pc: #1E432A; }
.gsh[data-lab] .p-chase { --pc: #1E3456; }
.gsh[data-lab] .p-sasha { --pc: #7F3025; }
.gsh[data-lab] .p-rohan { --pc: #6B4412; }
.gsh[data-lab][data-theme="dark"] .p-felix { --pc: #4ADE80; }
.gsh[data-lab][data-theme="dark"] .p-chase { --pc: #60A5FA; }
.gsh[data-lab][data-theme="dark"] .p-sasha { --pc: #FB7185; }
.gsh[data-lab][data-theme="dark"] .p-rohan { --pc: #FBBF24; }

/* mechanical top: dispatch number + persona pill */


/* brackets as content, so the text stays clean for copy and screen readers */


/* micro-labels — 11px / .08em per the type spec */


/* the ping: the loudest thing on the card */


/* the decode: Caslon turn, then Space Grotesk */

.gsh[data-lab] .qq-lead {
  font: italic 400 15.5px/1.5 var(--gs-serif); letter-spacing: -.01em;
  color: var(--gs-ink); margin-right: .15em;
}

/* the reply */
.gsh[data-lab] .qq-reply {
  padding: 14px 15px; border-radius: 14px;
  background: var(--fold-a); border: 1px solid var(--gs-rule);
}
.gsh[data-lab] .qq-reply .qq-text { color: var(--gs-ink); }


.gsh[data-lab] .qq-btn {
  position: relative; padding: 8px 16px;
  border: 1px solid var(--gs-ink);
  background: var(--gs-ink); color: var(--gs-ground); letter-spacing: -.01em; cursor: pointer;
  transition: opacity .2s ease, border-color .2s ease, color .2s ease;
}
.gsh[data-lab] .qq-btn:hover { opacity: .88; }


.gsh[data-lab] .qq-btn:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }
/* confirmation overlays the label, so the button never changes width */
.gsh[data-lab] .qq-done {
  position: absolute; inset: 0; display: grid; place-items: center;
  border-radius: inherit; background: inherit; color: inherit;
  opacity: 0; transition: opacity .18s ease;
}
.gsh[data-lab] .qq-btn.is-done .qq-done { opacity: 1; }

/* watermark — quiet on screen, legible in a screenshot */


.gsh[data-lab] .qq-note { margin-top: 24px; max-width: 70ch; font-family: var(--gs-ui); }

@media (prefers-reduced-motion: reduce) {
  .gsh[data-lab] :is(.qq-tab, .qq-btn, .qq-done) { transition: none; }
}
@media (max-width: 560px) {
  .gsh[data-lab] .qq-sec { padding-block: 36px; }
  .gsh[data-lab] .qq-tab { padding: 7px 12px; font-size: 12.5px; }
  .gsh[data-lab] .qq-tab i { display: none; }
  
  
}

/* ============================================================
   §8  /downtime — the Playlists & films hub

   The shelf card "Playlists & films" used to point at /shelf#listen,
   which only scrolled back into the same page and landed on a podcast
   list. This page is its real destination.

   It reuses .pl-grid / .pl-card wholesale from /playlists and /films,
   so the only new rules are the two section headings that separate the
   collections. Nothing here restyles a card.
   ============================================================ */

/* The page head reuses Quick Q's rules verbatim via :is() above — same
   stylesheet chain, same skeleton, so the two shelf destinations cannot
   drift apart when either is polished. Only the group headings below are
   specific to this page. */
.gsh[data-lab] .dt-grp { margin-top: 44px; }
.gsh[data-lab] .dt-grp:first-of-type { margin-top: 32px; }

.gsh[data-lab] .dt-grp-h {
  display: flex; align-items: baseline; flex-wrap: wrap; gap: 0 12px;
  margin: 0 0 18px;
  font: 700 23px/1.25 var(--gs-head);
  letter-spacing: -.02em; color: var(--gs-ink);
}

/* The number is an index into the page, not a sequence to follow, so it
   stays quiet: mono, small, and never the same weight as the heading.

   The box it used to carry — 4px 7px padding, 1px border, 6px radius — came
   off on 5 Nov under invariant 2: a pill wrapper on a kicker. It was also
   `display: block` inside a wrapping flex row, so it took a whole line and
   left "01 //" stranded above the title on phones. */
.gsh[data-lab] .dt-grp-n {
  display: inline;
  font: 500 11px/1 var(--gs-mono);
  letter-spacing: .1em; text-transform: uppercase;
  color: var(--gs-ink-3);
  padding: 0; border: 0; border-radius: 0; background: none;
}

/* Full-width so the sentence sits under the heading rather than running
   on after it, which is what the baseline flex row would otherwise do. */
.gsh[data-lab] .dt-grp-h small {
  flex: 1 0 100%;
  margin-top: 6px;
  font: 400 14px/1.6 var(--gs-ui);
  letter-spacing: 0; color: var(--gs-ink-2);
}

/* Each collection carries its own "All …" link, so the two sit closer to
   their grid than the page-level "Back to the shelf" does. */


@media (max-width: 560px) {
  .gsh[data-lab] .dt-grp { margin-top: 34px; }
  .gsh[data-lab] .dt-grp-h { font-size: 20px; }
}

/* ---------------------------------------------------------------------------
   9. Pillar kicker on pages whose body is not inside a .gsh wrapper
   ---------------------------------------------------------------------------
   §8 styles `.dt-kick` as `.gsh[data-lab] :is(.qq-kick, .dt-kick)`, which is
   right for the exported pages. /tools is not built that way: its two .gsh
   wrappers are the header chrome and the footer, and the hero sits directly on
   <body>. So the scoped rule never matched and the kicker rendered as plain
   15px Space Grotesk — body copy, where the system wants a mono eyebrow.

   This is the same treatment at a lower specificity, so the .gsh pages keep
   their own rule and nothing else changes. Values copied from
   gs-site-lab.css's `.gsh .mono` rather than re-invented. */
.dt-kick {
  display: inline-block;
  font-family: var(--gs-mono, "JetBrains Mono", ui-monospace, monospace);
  font-size: 11px; font-weight: 500; line-height: 1.4;
  letter-spacing: 0.13em; text-transform: uppercase;
  color: var(--gs-ink-3, #6B7389);
}
.dt-kick b { font-weight: 500; color: var(--gs-green-active, var(--gs-green)); }


/* ---------------------------------------------------------------------------
   10. body still asked for Inter on the five exported pages
   ---------------------------------------------------------------------------
   gs-site-lab.css:10 sets `--font-body: "Inter", ...` and line 65 applies it to
   `body`. Lovable called this a dead fallback string; it is not — it is the
   body font on /, /about, /shelf, /playlists and /quick-q.

   The visible blast radius is small, because almost everything sits inside a
   .gsh wrapper that sets its own face. But Inter is out of the type system and
   is never downloaded, so anything that does inherit from body rendered Inter
   for viewers who happen to have it installed locally and Helvetica Neue — a
   heading face — for everyone else. Same divergence we fixed in site.css.

   Set here rather than in gs-site-lab.css so that file stays byte-identical to
   Lovable's export. --font-ui is their own token, defined in the same file. */
body { font-family: var(--font-ui, var(--gs-ui, "Space Grotesk")), "Helvetica Neue", Helvetica, Arial, sans-serif; }

/* ---------------------------------------------------------------------------
   11. Curation fold: per-card CTA, and unlinked items
   ---------------------------------------------------------------------------
   The rack's cards gained a uniform "Explore all" footer and, in two places,
   an item with no destination (Moneyball on the homepage, Kyle's rotating slot
   on /playlists). Everything else here is Lovable's existing chassis —
   .lab-sh-card, .lab-sh-i, .pl-card and the tones are untouched. */

/* Footer CTA. Pinned to the bottom so all four read as one row of cards even
   when one list runs a line longer than the others. */
.gsh .lab-sh-card { display: flex; flex-direction: column; }
.gsh .lab-sh-card ul { flex: 1 1 auto; }
.gsh .lab-sh-all {
  display: inline-flex; align-items: center; align-self: flex-start;
  margin-top: 14px; padding-top: 12px;
  border-top: 1px solid var(--gs-rule);
  width: 100%;
  font-size: 11px; letter-spacing: 0.13em; text-transform: uppercase;
  color: var(--gs-green-active); text-decoration: none;
}
.gsh[data-theme="dark"] .lab-sh-all { color: var(--gs-green); }
.gsh .lab-sh-all:hover { text-decoration: underline; }

/* An item with no link must not look clickable: no arrow, no hover, and the
   cursor stays default. */
.gsh .lab-sh-i.is-static { cursor: default; }
.gsh .lab-sh-i.is-static .gs-arr { display: none; }
.gsh .lab-sh-i.is-static:hover { background: none; }

/* The rotating slot on /playlists, same idea on the bigger chassis. */


/* .pl-why needs nothing here. gs-site-lab.css already styles it — Space
   Grotesk 13.5px with a hairline above — and that is right: DESIGN-LANGUAGE.md
   §2 reserves Caslon for a punchline inside a heading, never for a line of
   body copy. An earlier draft of this file set it in Caslon italic and was
   out-specified by Lovable's own rule, which is the correct outcome. */

/* ---------------------------------------------------------------------------
   12. /tools — Lovable's redesigned body, two rules it needs
   ---------------------------------------------------------------------------
   The new markup reuses `.card tool` from the legacy chain, and site.css sets
   `.tool h3 { font-family: var(--gs-accent); font-style: italic }` — Libre
   Caslon italic, left from the old tool cards. DESIGN-LANGUAGE.md §2 reserves
   Caslon for a punchline inside a heading, never a card title, and Lovable's
   own mock shows these in bold sans. Put back on the heading face.

   The heading is also no longer a direct child of .wrap — it sits inside
   `.sec-head tk-head` now — so the rule that sized it no longer reaches it.
   `.tk-h1` had no CSS at all, which left the page headline at 15px body copy. */
#tools .tool h3,
#tools .card h3 {
  font-family: var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  font-style: normal; font-weight: 700;
  letter-spacing: -0.02em; line-height: 1.25;
  margin-top: 0;
}
.tk-hero .tk-h1 {
  font-family: var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  font-weight: 700; letter-spacing: -0.03em; line-height: 1.06;
  font-size: clamp(34px, 5.4vw, 64px);
}

/* ---------------------------------------------------------------------------
   13. gs-flip has to reach the legacy pages too
   ---------------------------------------------------------------------------
   gs-site-lab.css scopes it as `html.gs-flip .gsh, html.gs-flip .gsh *`, which
   is right for the exported pages — everything there is inside a .gsh wrapper.
   On /tools, /privacy-and-terms, /404 and the games, only the header chrome
   and the footer are wrapped; the body sits directly on the page. So the
   suppression never reached the thing actually changing colour, and the
   grounds and cards eased at different rates — the flicker gs-flip exists to
   prevent.

   Unscoped, for one frame only, during a deliberate theme change. */
html.gs-flip,
html.gs-flip * { transition: none !important; }

/* ---------------------------------------------------------------------------
   14. (retired 4 Nov) /tools proof grid phone breakpoint
   ---------------------------------------------------------------------------
   `.tk-proof` was the four uppercase mono bullets. They are micro-cards now
   (§25), which carry their own breakpoints, so this rule matched nothing and
   was removed rather than left to mislead. Kept as a note because the reason
   still applies to the cards: 11px uppercase at 0.12em tracking does not fit
   a 147px column — "Backed by 20+ years" ran to four lines there. */

/* ---------------------------------------------------------------------------
   15. /tools hero: keep the typography, drop the decoration
   ---------------------------------------------------------------------------
   site.css scopes the whole tool-page hero on `.hero:has(.ctas)` — the 60px
   headline, the Caslon accent styling and the entrance animation all hang off
   it, which is why `.ctas` had to be restored. But the same selector also
   paints an indigo radial glow (::before) and a star field (::after), and
   those were written when tools-home.html WAS the homepage.

   The real homepage does not have them: measured on `/`, the hero's ::before
   content is `none` and `--gs-stars` is unset. So /tools was the only page on
   the site with a glow, which is what made dark mode look wrong there.

   Typography stays, decoration goes. */
.tk-hero::before,
.tk-hero::after { content: none !important; }

/* ---------------------------------------------------------------------------
   16. /tools headline leading
   ---------------------------------------------------------------------------
   Lovable specified `line-height: 1.06`. At the 60px desktop size that leaves
   3.6px of leading, and this headline is not plain text: the Caslon accents
   carry a drawn rule at `bottom: .04em`, so the underline beneath "funnel"
   lands in the cap height of the line below it. "stage" has a descender in the
   same place. The two lines read as one block.

   1.12 gives 7px of leading — still tight, still the same typographic
   intention, but the rule and the descenders clear the next line. A little
   more air under the headline for the same reason. */
.tk-hero .tk-h1 {
  line-height: 1.08 !important;
  margin-bottom: 1.25rem !important;
}
/* §16b: the inline clamp has to beat the `.hero:has(.ctas) h1` rule from
   site.css, which caps the headline at 64px. Lovable's layout runs it wider
   and larger — a statement across the measure rather than a two-line heading.
   Colour, weight and the accent treatment are untouched: those come from the
   homepage system via §17. */
.tk-hero .tk-h1 { font-size: clamp(32px, 4.5vw, 58px) !important; }

/* §16c (4 Nov): 74px was a statement, but it cost the page its job. Measured
   on a 1440x800 laptop: header 74 + 92px hero pad + a 166px headline + a 112px
   lede put the tool cards at y=600, with 200px of a 480px card showing. The
   cards are what the page is for, so the headline gives ground to them —
   58px max, tighter leading, less air under each block. Cards now start at
   y=471. Nothing else about the headline changed: the Caslon accents, the
   drawn rule and the colour all still come from the homepage system. */
.tk-hero .lede { margin-bottom: 1.75rem !important; }

/* ---------------------------------------------------------------------------
   17. /tools light mode was running a cool palette
   ---------------------------------------------------------------------------
   The two stylesheet chains define the same token NAMES with different VALUES,
   and only in light. Measured:

     token          homepage (gs-site-lab.css)   /tools (site.css)
     --gs-surface   #F2ECE4  warm beige          #FAFBFC  cool near-white
     --gs-raised    #FDFCF9  warm off-white      #FFFFFF  pure white
     --gs-rule      #E8E1D8  warm linen          #E4E7EC  cool grey

   In dark the two agree exactly (#0E1018 / #1A1D2B / #ffffff17), which is why
   only light looked wrong. The cards sat in cool grey on a warm linen ground
   while every other page sat in warm on warm.

   Kyle's call: the homepage values are correct, so this page takes them. The
   homepage's own picks cards measure #FFFFFF on a #E8E1D8 hairline — that is
   what these mirror. Scoped to this page so no other legacy page shifts
   underneath us. */
.gsh[data-lab]:not([data-theme="dark"]) .tk-hero,
.gsh[data-lab]:not([data-theme="dark"]) #tools,
html:not([data-theme="dark"]) .tk-hero,
html:not([data-theme="dark"]) #tools {
  --gs-surface: #F2ECE4;
  --gs-raised:  #FDFCF9;
  --gs-rule:    #E8E1D8;
}
/* The tool cards use the homepage's card recipe verbatim, measured off
   `.lab-sh-card` on `/` in each mode. Not approximated — background, hairline
   and both shadow layers are exactly what the picks rack uses, which is what
   makes a card on this site read as raised rather than merely tinted.

     light   #FFFFFF on #E8E1D8, inset white highlight + a 2px drop
     dark    #141825 on rgba(255,255,255,.09), inset white highlight only

   The ground under them is #F7F3EE / #07080D, the same folds the homepage
   rack sits on, so the contrast is identical too.

   Radius stays at Lovable's 16px rather than the homepage's 8px: the layout
   of this page is theirs, only the surface treatment is ours. */
html:not([data-theme="dark"]) #tools .card {
  background: #FFFFFF !important;
  border: 1px solid #E8E1D8;
  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, .7),
              0 1px 2px 0 rgba(26, 29, 43, .04);
}
html[data-theme="dark"] #tools .card {
  background: #141825 !important;
  border: 1px solid rgba(255, 255, 255, .09);
  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, .05);
}

/* ---------------------------------------------------------------------------
   18. /tools hero sat too close under the header
   ---------------------------------------------------------------------------
   site.css: `.hero:has(.ctas) { margin-top: -4.6rem; padding-top: 7rem }`.
   That -73.6px pull exists because tools-home.html used to BE the homepage,
   under a header of a different height — the comment in site.css says it
   "slid the headline under the sticky header" on the pages that inherited it
   unscoped.

   Restoring `.ctas` to get the headline typography back (§12) brought the pull
   with it. Measured against the real homepage:

     /          hero top  +6px,  padding-top 74px  ->  content at 80px
     /tools     hero top -14px,  padding-top 74px  ->  content at 60px

   Correction to an earlier note: the homepage hero has a negative pull too —
   `margin-top: -72px` with `padding-top: 103.5px`. The pull is not the
   difference; the net is. Measured at 1440px with the header bottom at 74px:

     /          first hero content at 166px  ->  92px clear of the header
     /tools     first hero content at 140px  ->  66px clear

   So this neutralises the -4.6rem and sets the padding to land on the same
   92px. One number, taken off the homepage, not guessed. */
/* Tried and reverted: `min-height: calc(100svh - 150px)` with the content
   centred, copied from the homepage's `.hero-split`. That rule is why the
   homepage headline sits lower on a tall window — its content is centred in a
   930px box. But the homepage hero holds the animated desk; this one holds
   four lines. Same rule, far less content, so the leftover space is bigger:
   measured 242px of clearance against the homepage's 158px, with the tool
   cards pushed off the first screen. Matching the RULE gives a worse result
   than matching the NUMBER. */
.tk-hero { margin-top: 0 !important; padding-top: 56px !important; }
@media (width <= 560px) {
  .tk-hero { padding-top: 36px !important; }
}

/* ---------------------------------------------------------------------------
   19. The theme toggle's knob, defined once for every page
   ---------------------------------------------------------------------------
   The knob lives in two different stylesheets with two different scopes:

     gs-site-lab.css   `.tico::before`          (5 exported pages + 3 ported)
     gs-lab.css        `.gsh .tico::before`     (/games, both tool pages)
     neither           —                        (/play and the 3 game boards)

   On /games the `.gsh .tico.is-dark::before` rule matches the element, has no
   !important competing with it and sits under no media query, yet the computed
   transform stayed at translateX(0) — the knob never moved. Rather than keep
   chasing that through the cascade, the control is defined here, in the one
   stylesheet every page loads last, unscoped so it does not depend on a .gsh
   wrapper. Values are gs-site-lab.css's, unchanged.

   This makes the toggle identical on all 15 pages in both themes, which is the
   point: it is the only control in the universal header. */
.tico { position: relative; }
.tico::before {
  content: ""; position: absolute; top: 2px; left: 2px;
  width: 26px; height: 26px; border-radius: 999px;
  background: var(--gs-ink);
}
.tico.is-dark::before { transform: translateX(28px) !important; }
.tico:not(.is-dark)::before { transform: translateX(0) !important; }

/* The knob only animates once the page has settled. `data-theme` is set
   pre-paint but `is-dark` is set by JS after the DOM is parsed, so a dark-mode
   load used to show the knob slide across the track a beat after everything
   else. `gs-knob-ready` is added by section 6 of gs-static-fix.js, one frame
   later; until then the transition is off everywhere, !important because the
   lab stylesheets declare their own on `.tico::before` and `.gsh .tico::before`. */
html:not(.gs-knob-ready) .tico::before { transition: none !important; }
html.gs-knob-ready .tico::before {
  transition: transform .25s var(--gs-ease, cubic-bezier(.2,.7,.3,1));
}

/* ---------------------------------------------------------------------------
   20. Canonical surface tokens on the legacy chain
   ---------------------------------------------------------------------------
   `site.css` defines these at :root with cool-grey light values; the lab chain
   defines them on `.gsh` with the warm ones. Same names, different colours,
   light mode only — which is why /tools read cool-grey for a week and why any
   `var(--gs-surface)` written against the homepage resolves differently on
   seven pages.

   Measured on `/` with transitions disabled, and approved by Lovable as the
   canonical set:

       --gs-surface   #F2ECE4   (site.css had #FAFBFC)
       --gs-raised    #FDFCF9   (site.css had #FFFFFF)
       --gs-rule      #E8E1D8   (site.css had #E4E7EC)

   Dark already agreed exactly — #0E1018 / #1A1D2B / rgba(255,255,255,.09) —
   so only light is redefined here. Set at :root so it reaches the legacy
   pages; the lab pages define these on `.gsh`, which is more specific and
   wins for everything inside it, so they are untouched. */
:root:not([data-theme="dark"]) {
  --gs-surface: #F2ECE4;
  --gs-raised:  #FDFCF9;
  --gs-rule:    #E8E1D8;
}
/* Dark is set here too, even though site.css already agrees, because /games
   loads neither site.css nor gs-site-lab.css — it had NO :root definition for
   these at all in dark, so `var(--gs-surface)` resolved to nothing there. */
:root[data-theme="dark"] {
  --gs-surface: #0E1018;
  --gs-raised:  #1A1D2B;
  --gs-rule:    rgba(255, 255, 255, .09);
}

/* ---------------------------------------------------------------------------
   21. /tools — token pins for content outside `.gsh`
   ---------------------------------------------------------------------------
   This page's two `gsh gsh-chrome` wrappers hold the header and the footer
   only; everything between them is unwrapped. Outside `.gsh` these tokens
   resolve from site.css's `:root` — the cool-grey set — and
   `--gs-green-active` is not defined there at all. A
   `color: var(--gs-green-active)` with no definition is invalid at
   computed-value time and silently inherits, which once shipped a green
   kicker in ink.

   §20 canonicalised surface/raised/rule at `:root`. These six are pinned here
   rather than widened to `:root`, because §20's three were measured and
   approved and these have not been checked against the ten ported pages that
   also read them outside `.gsh`. Values are the lab chain's, measured on
   `.gsh` on this page.

   The tool cards themselves no longer need these: they are the homepage
   `.tl-card` component now, inside their own `.gsh[data-lab]` wrapper (see
   §26). The micro-cards in §25 still sit outside it and do. */
#tools {
  --gs-green-active: #1B6B38;
  --gs-rule-2: #D9D0C4;
  --gs-red: #C8443A;
  --gs-ink-2: #5D5348;
  --gs-ink-3: #6E655B;
  --gs-indigo: #485289;
}
html[data-theme="dark"] #tools {
  --gs-green-active: oklch(78% .16 150);
  --gs-rule-2: rgba(255,255,255,.18);
  --gs-red: #E05D52;
  --gs-ink-2: #9AA2B8;
  --gs-ink-3: #8E97AB;
  --gs-indigo: oklch(70% .13 273);
}

/* ---------------------------------------------------------------------------
   22. /about — flip-card state, card-above-form on mobile, flush footer
   ---------------------------------------------------------------------------
   Lovable's 4 Nov spec. The flip itself is gs-runtime.js:221 — one of the
   three locked files — which toggles `.is-on` and `aria-pressed` and nothing
   else. The two labels that now change with that state are written by
   gs-static-fix.js §7; everything visual is here. */

/* Mobile: the live card sits above the fields it is built from, so a phone
   keyboard cannot cover the thing you are editing. `.idwrap` is a single
   column below 900px, which is what makes order meaningful here. */
/* `order` only sorts siblings, and `.idform` is a GRANDCHILD of `.idwrap` —
   the grid's two children are `.idcard` and an unnamed div that holds the
   form. The spec's `order: 5` on `.idform` was inert; it goes on the child
   that actually sits in the grid. Measured: the card was still rendering
   below the form with both orders applied. */
@media (width <= 900px) {
  .gsh[data-lab] #card .idwrap > .idcard { order: 4; }
  .gsh[data-lab] #card .idwrap > :not(.idcard) { order: 5; }
}

/* Footer brand shares the page's left edge on phones. `.lft-brand` otherwise
   carries calc(var(--hd-pad-l) + 1px), which is the desktop header inset. */
@media (width <= 720px) {
  .gsh .lft-brand { padding-left: 0; }
}

/* ---------------------------------------------------------------------------
   23. Card slot baselines, first-line dot anchoring, click focus
   ---------------------------------------------------------------------------
   Lovable's 4 Nov invariant spec. Measured before applying; three of their
   specifics did not match this build, so the numbers below are ours.

   WHAT WAS ACTUALLY MISALIGNED
   /tools: at 1280 and 1440 both descriptions run 2 lines and the tag rows
   already matched to the pixel (740/740, 750/750). The break is at 1024,
   where card 01 runs THREE lines and card 02 two — the opposite way round
   from the spec, which has card 01 at one line. The CTAs were never
   misaligned at any width: `.tk-go` carries `margin-top:auto`, so flex had
   already pinned them to the bottom (781/781, 790/790, 801/801).

   /about: real and exactly as described. Card 02 runs 3 lines against 2 for
   the rest at 1280 and 1440, pushing its body text a full line down
   (1492 vs 1465). At 1024 it inverts — 01/02/03 run 3 lines and 04 runs 2.
   A 3-line envelope covers every case.

   WHY MIN-HEIGHT AND NOT SUBGRID
   Sibling grid items size their rows independently, so only `subgrid` shares
   a track between cards — and it reaches one level, while these slots sit
   two levels down inside `.tk-in`. Flattening that wrapper to use it would
   cost more than it buys. Locked envelopes expressed in line units it is.

   The class is `.tk-tags`, not the spec's `.tk-chips`. */

/* --- /tools: lock the description envelope to its worst case, 3 lines ------
   14px text at 1.55 line-height. Both tag rows then start on the same Y at
   every width, at the cost of one line of air where the copy runs short. */
/* --- Interactive cards: keyboard ring only --------------------------------
   Invariant 3. `.ab-card` already carried a `:focus-visible` rule and no
   plain `:focus` rule, so the double ring could not be reproduced here.
   Lovable confirmed on 4 Nov that what they saw came from focus transfer
   inside their live-preview iframe, not from this build — so this rule is a
   production safeguard rather than a fix for a fault that existed on the
   static site. Keep it: it costs nothing and seals the same class of problem
   in a real browser. It cannot regress keyboard focus — `:focus-visible`
   still draws. */
#tools .tl-card:focus:not(:focus-visible) {
  outline: none;
}

/* ---------------------------------------------------------------------------
   24. Mobile shelf accordion
   ---------------------------------------------------------------------------
   Lovable's spec, with the directory lifted from the desktop `.shf-dd`
   dropdown rather than retyped, so the two never drift and the SVGs match.

   Three notes.

   The mobile panel already lives inside a `<details>` — the hamburger itself
   is the outer disclosure and `.lab-mnav-panel` is its content. This is a
   nested disclosure, which is legal and behaves correctly: a `<summary>`
   toggles only its own `<details>`.

   `.shf-m` was already in the stylesheet
   (`.gsh .shf-m .shf-cols { grid-template-columns: 1fr }`), so the body
   carries it and the two-column directory collapses to one for free.

   The existing `.gsh .shf-*` rules are NOT scoped to `.shf-dd`, so they reach
   into this panel. Most are wanted — same component, same look. One is not:
   `.shf-cols .shf-g + .shf-g` sets a `border-left`, which in a single column
   draws a stray vertical rule down the left edge. It is reset below. */


.mn-shelf-body {
  margin: 6px 0; padding: 6px;
  border-radius: 14px; border: 1px solid var(--gs-rule);
  background: color-mix(in srgb, var(--gs-ink) 3%, transparent);
}
/* The single-column collapse comes from .shf-m; this kills the column rule
   that only makes sense side by side, and restores it as a row divider. */
.mn-shelf-body .shf-cols .shf-g + .shf-g { border-left: 0; }
.mn-shelf-body .shf-g + .shf-g,
.mn-shelf-body .shf-cols + .shf-cols { border-top: 1px solid var(--gs-rule); }
.mn-shelf-body .shf-g { padding: 10px 8px; }

/* The directory's own `.shf-i` / `.shf-h` styling carries over from the
   dropdown, which is the point. Only the full-page link needs re-seating,
   since it loses its `.shf-allw` wrapper here. */
/* --- Step 1, approved 5 Nov: drawer polish --------------------------------

   1. Items lay out as a row with the arrow flush right.

   Implemented as a two-column grid rather than the spec's literal
   `display: flex; justify-content: space-between`. The item has THREE
   children — `<b>`, `<small>` and the arrow `<svg>` — so a flex row puts the
   description beside the title instead of under it. The grid keeps the text
   stacked in column one and pins the arrow in column two, vertically centred,
   which is the stated intent: arrows flush right, never orphaned onto a third
   line. The inherited `.gsh .shf-i { display: grid }` from the desktop
   dropdown is what stacked them in the first place. */
.mn-shelf-body .shf-i {
  display: grid;
  grid-template-columns: 1fr auto;
  align-items: center;
  column-gap: .75rem;
  row-gap: 2px;
}
.mn-shelf-body .shf-i > b,
.mn-shelf-body .shf-i > small { grid-column: 1; }
.mn-shelf-body .shf-i > .gs-arr {
  grid-column: 2;
  grid-row: 1 / span 2;      /* NOT `1 / -1`: with no explicit rows declared,
                                -1 resolves to the end of the EXPLICIT grid,
                                which is row 1 — the arrow then pinned to the
                                title line and sat 18px above centre on
                                two-line items. Measured. */
  align-self: center;
  margin-left: 0;            /* the lifted SVG carries an inline .55em left
                                margin for inline flow; the grid column owns
                                the spacing here. */
}

/* 2. Group heads are pure mono — no number box.
   `.gsh .shf-h b` gave `03` a filled, rounded, padded chip: a pill wrapper in
   a navigation drawer, which invariant 2 bans outright. The number stays, as
   type. */
.mn-shelf-body .shf-h b,
.gsh[data-theme="dark"] .mn-shelf-body .shf-h b {
  background: none;
  border-radius: 0;
  padding: 0;
  margin-right: .5ch;
  color: var(--gs-ink-3);
}

/* 3. Group 03 keeps its title and status on one line.
   "03 The spaces // Coming soon" ran to two lines at 375px. The head is a
   flex row, so the fix is to let it shrink rather than wrap: the status tag
   stops being a flex item that forces a break, and the tracking eases off
   just on phones. Measured below 420px only; the desktop dropdown that shares
   these classes is untouched. */
@media (width <= 420px) {
  .mn-shelf-body .shf-h {
    flex-wrap: nowrap;
    letter-spacing: .04em;
    gap: 5px;
  }
  .mn-shelf-body .shf-h .shf-cs { flex: none; letter-spacing: .04em; }
}

/* 4. "Explore the full shelf page" — legible on its gradient.

   Lovable reported dark-on-dark here and our first verification pass said it
   did not reproduce. **They were right and we were wrong.** The check read
   `backgroundColor`, which is transparent: the button is painted by a
   `linear-gradient` living in `backgroundImage`. Measured properly, both
   themes were wrong, in opposite directions:

     light:  ink  #1A1D2B  on indigo -> dark green  (both ends dark)
     dark:   near-white    on oklch .70/.74         (both ends light)

   The desktop `.shf-dd .shf-all` has always had this right — white on the
   dark gradient, near-black on the light one. This mirrors it. The gradient
   itself is untouched.

   The same trap cost time on `.tl-go` earlier: a transparent
   `backgroundColor` does not mean a transparent background. */
/* The button paints its gradient but its colour came out INHERITED ink: the
   upstream rule is `color: var(--gs-on-accent)`, and that token is not
   defined on the five exported pages — an undefined var makes the whole
   declaration invalid at computed-value time, so `color` silently falls back
   to the inherited value. Defining the token here fixes the upstream rule
   rather than fighting it, and the explicit colour covers the arrow too. */
.mn-shelf-body { --gs-on-accent: #FFFFFF; }
.gsh[data-theme="dark"] .mn-shelf-body { --gs-on-accent: #07080D; }
.mn-shelf-body a.shf-all,
.mn-shelf-body a.shf-all .gs-arr { color: var(--gs-on-accent); }

.mn-shelf-body .shf-allw { margin-top: 6px; }
.mn-shelf-body .shf-all {
  display: flex; align-items: center; justify-content: space-between;
  padding: 11px 12px; border-radius: 0 0 8px 8px;
}

/* The drawer has to scroll itself now that it can hold the whole directory.
   Measured on a 360x640 Android with the accordion open: the panel ran 916px
   tall and put "Get in touch" at y=975. It is `position: absolute` inside a
   `position: sticky` header, so it travels with the viewport — scrolling the
   page moved the CTA not one pixel. The primary CTA was unreachable on every
   phone size tested (640, 812 and 844 tall).

   Capping the panel and letting it scroll internally is the standard drawer
   behaviour and keeps About and the CTA reachable at any height. `svh` so a
   mobile browser's collapsing toolbar cannot clip it; `vh` first for anything
   that does not know `svh`. `overscroll-behavior` stops the page behind the
   drawer from scrolling when the list hits its end. */
.lab-mnav-panel {
  max-height: calc(100vh - 84px);
  max-height: calc(100svh - 84px);
  overflow-y: auto;
  overscroll-behavior: contain;
}

/* ---------------------------------------------------------------------------
   25. /tools — the Operator Baseline spec strip
   ---------------------------------------------------------------------------
   Lovable's 4-bay structure and copy, answering our question about whether
   the proof points still earned their place under the richer tool cards:
   keep them, make them authoritative.

   Their spec arrived with the structure and copy only — it ends after the
   markup block with no styling section. Rather than invent a different look,
   this carries the established card language across to the new class names:
   raised ground, hairline border, 12px radius against the tool cards' 16px,
   a green pip, a mono kicker and sentence-case copy. Swap in their treatment
   when the rest of the spec lands.

   `.tk-proof` in their spec no longer exists — it became micro-cards in §51
   and is now this. `.pip`, `.tk-specs` and `.tk-spec-*` were all free.

   Slot envelopes are locked on desktop per the §23 invariant, so the four
   bays align across the strip; fluid below 1024 where they pair or stack. */


@media (width >= 1024px) {
  
  
     /* worst case */
}

/* The Acquret callout follows immediately in the DOM and carries no top
   margin of its own, so the two blocks sat flush — 0px at every width. */


/* Sentence case, never all-caps — this is the copy, not a label. */


/* ---------------------------------------------------------------------------
   26. The tool-card band: full-bleed carousel on phones
   ---------------------------------------------------------------------------
   `.tl-grid` already carries the homepage carousel — `overflow-x: auto`,
   `scroll-snap-type: x mandatory`, and a -20px/+20px margin-padding pair that
   lets the first card sit flush to the screen edge. It was not reading that
   way on /tools because of one extra gutter: this page's outer
   `<section id="tools">` carries an inline `padding: 0 1rem`, so the `.wrap`
   inside it measured 343px where the homepage's measures the full 375.
   Cards came out 261px against the homepage's 288, boxed in rather than
   running off the edge.

   Cancelling that 1rem for the card band only — the micro-cards and the
   Acquret callout below still want it. At desktop this changes nothing
   visible: `.wrap` stays centred at its 1040px measure. */
/* Only in the carousel band. `.tl-sec .tl-grid` becomes a snap scroller at
   <=759px and returns to two columns from 720px, so the bleed belongs below
   719. Above it, cancelling the gutter would widen `.wrap` to 1072px and
   break the approved 1040 measure the hero and micro-cards share — measured,
   after the first attempt did exactly that. */
@media (width <= 719px) {
  #tools .tk-toolwrap { margin-inline: -1rem; }
}

/* The band sets its own rhythm; the section padding above it can go. */
#tools .tk-toolwrap .tl-sec { padding-top: 16px !important; }

/* `.gsh` is a PAGE SHELL, not a band. Alongside its tokens it carries
   `min-height: 100vh`, `background: var(--gs-ground)`, a font-family and a
   colour transition — fine on the five exported pages, where it wraps the
   whole document, and wrong here, where it wraps two cards in the middle of
   a page. Measured: the wrapper stood 1000px tall against 501px of content,
   painting ~499px of empty white between the cards and the micro-cards, and
   it tracked the viewport (600px tall in a 600px window).

   Only the shell behaviour is cancelled. The tokens, the colour and the font
   are the whole reason for wrapping, so they stay. */
#tools .tk-toolwrap,
.tk-mqwrap {
  min-height: 0;
  background: transparent;
}

/* The marquee needs the same wrapper for a different reason: every rule for
   it is `.gsh .marquee…`, and on this page it was sitting directly on <body>.
   It rendered as one unstyled run-on line — no overflow clip, no flex track,
   **no animation**. The earlier check confirmed the strip was present and
   above the footer and never asked whether it was styled, which is how a dead
   component passed verification. Presence is not rendering. */

/* The card band shares the page's left edge from tablet up. `.wrap` carries
   20px of side padding, which on the homepage is what insets the cards — but
   there the headline is centred in a wider 1180px wrap, so nothing lines up
   against them. Here the hero, the micro-cards and the Acquret callout all
   sit on the 1040 measure, and the cards were landing 20px inside it. Measured
   at 1440: cards at x=220 against everything else at x=200.

   Dropped from 720px up only — below that the same padding is what holds the
   first card 20px off the screen edge in the carousel, exactly as on the
   homepage. */
@media (width >= 720px) {
  #tools .tk-toolwrap .wrap { padding-inline: 0; }
}

/* ---------------------------------------------------------------------------
   27. Phase 2 Step 2 — /games
   ---------------------------------------------------------------------------
   Persona order, kickers, headline token and CTA. The markup changes are in
   `launch-build/games/index.html`; these carry the type and the treatment.

   The seat used to live in `.ib-chip`, a pill beside the number. Invariant 2
   bans pill wrappers on kickers, so the seat moved into the kicker line
   itself — `01 // Marketing // Felix Funnel` — and the chip element is gone
   from the markup entirely rather than hidden. `.ib-ts` stays: a duration is
   a value, not a label. */
/* The floating dot pip. `gs-pages.css:64` draws a 6px persona-coloured circle
   before every kicker via `.ib-id::before`. That is precisely what invariant
   2 bans, and removing `.ib-chip` left it as the last one on the page — it
   survived our first pass because we were looking for pill *wrappers* and
   this is a pseudo-element. The persona is named in the kicker text now, so
   nothing is lost but the dot. */


/* Phones: the four games become a horizontal snap carousel.

   Kyle's call, and it matches what the tool cards already do on `/tools` and
   the homepage — one pattern for card rows on a phone rather than two. A
   stack of four tall cards is a long scroll past three games to reach the
   fourth; a carousel puts them one swipe apart.

   `.gsh .gh-grid` is a one-column grid below 760px, so this switches it to
   `grid-auto-flow: column` with 86% columns and x-snap. The breakpoint is
   759px because that is where gs-pages.css already drops to one column —
   reusing it rather than inventing a second boundary.

   The grid lives inside the `.lw` window-frame mock, whose `.lw-body` carries
   16px of padding. The negative margin and matching padding let the first
   card sit flush to the frame's inner edge and the last one scroll fully into
   view, the same -pad/+pad pair the tool carousel uses against the viewport. */


/* Phones: the kicker gets the row to itself.

   `.ib-meta` is `flex-wrap: nowrap`, so at 375px the kicker was squeezed into
   204px of a 259px row by the `≈ 1 min` beside it, and wrapped mid-phrase —
   "01 // MARKETING // FELIX / FUNNEL", with the last word orphaned on its own
   line. Three of the four did it. Card 04 is 42 characters and cannot fit one
   line on a phone at any sensible size, so the answer is a deliberate break
   rather than a tighter squeeze: full width, eased tracking, and
   `text-wrap: balance` so the two lines come out even. The duration moves
   under it, right-aligned, instead of competing for the first line. */


/* In the carousel the cards sit side by side, so the peeking neighbour makes
   a mismatched kicker height visible in a way the old vertical stack never
   did. Cards 01 and 04 run two lines at phone widths while 02 and 03 run one.
   Floored to two lines below 560px, where the wrap actually happens — by
   759px the card is 585px wide and all four fit one line, so no floor there. */
@media (width <= 560px) {
  
  /* The persona line under the name wraps in some cards and not others —
     "10 dashboards before the 9 AM" against "12 form fills in the inbox" —
     which showed up at 390px as a one-line offset on the title below. Same
     floor, same reason. */
  
}

/* Slot envelope for the kicker, per the §23 invariant.

   "04 // Leadership & Finance // Rohan Runway" is the longest of the four and
   wraps to two lines from about 900px down, with Felix's joining it at 768.
   The CTA is bottom-anchored so it stayed aligned, but everything between the
   kicker and it — avatar, name, title, description — sat 10px lower in the
   wrapped cards than in their row partners. Measured kicker heights at 768:
   [21, 11, 11, 21].

   Locked across the two-column band only. Above 1000px all four run one line
   and the floor costs nothing; below 720 the cards are a single column and
   there is nothing to align against. */


/* The headline carried `em.g`, which is green. Invariant 3 reserves green for
   pips, confirmations and mono slashes — headlines are ink with Caslon
   italic. The class is now `.ink`; the italic and the face are untouched. */
.pg-h em.ink {
  color: inherit;   /* NOT var(--gs-ink): the dark headline is #E8EAF0, not
                       the ink token, so pinning the token made the accent 5
                       units brighter than its own headline. Inheriting is
                       what "same ink as the headline" actually means, and it
                       survives any future change to the heading colour. */
  font-family: var(--gs-serif, "Libre Caslon Text", Georgia, serif);
  font-style: italic;
}

/* "Play →" was a bare text link with a unicode arrow. Now a tactile operator
   pill with the site's thin-stroke SVG, which also settles invariant 7: no
   glyph arrow left in a CTA on this page. */


/* ---------------------------------------------------------------------------
   28. Phase 2 Step 3 — /playlists (Downtime Hub)
   ---------------------------------------------------------------------------
   Lovable's spec names a `.dl-` chassis to port. **It does not exist in this
   build** — the tiers are `.dt-grp` and the cards `.pl-card` / `.pl-grid`,
   and their `.dl-why` editorial pull-quote is our `.pl-why`, already live on
   the Films tier. So this is not a port: it is their treatment applied to the
   classes that are actually here. Same outcome, our names. */

/* --- 1. Typographic kickers: `01 // SOUNDTRACKS` -------------------------
   The number sat alone in `.dt-grp-n` with the title as a bare text node, so
   a `//` added inline would wrap away from its number on a narrow screen —
   exactly the orphan the spec warns about. The slash is its own element
   inside the number span, which makes `01 //` a single unbreakable unit and
   lets the slashes take the green without touching the title. */
.gsh .dt-grp-n {
  /* The chip itself is removed at source, where it was added — see the
     `.gsh[data-lab] .dt-grp-n` block above. This only keeps `01 //` from
     ever breaking between the number and its slashes. */
  white-space: nowrap;
  font-variant-numeric: tabular-nums;
}
/* The head is a wrapping flex row, so the title is a bare text node competing
   with the number for lines. Block display makes number and title one flow,
   and the title wraps under itself rather than the number standing alone. */
.gsh .dt-grp-h { display: block; }
.gsh .dt-slash {
  font-style: normal;
  color: var(--gs-green-active, #1B6B38);
  margin: 0 .5ch 0 .45ch;
  letter-spacing: .06em;
}
.gsh[data-theme="dark"] .dt-slash { color: var(--gs-green, #35BD6B); }
/* Below 720 the head stays a single flow — number, slashes and title together
   — so "01 //" can never sit on a line of its own above the title. */
@media (width <= 719px) {
  .gsh .dt-grp-h { text-wrap: balance; }
  .gsh .dt-grp-n { margin-right: .15ch; }
}

/* --- 2. A hairline accent per tier ---------------------------------------
   Tone tokens already exist per tier (`--tone-ink`): green on Soundtracks,
   blue on Podcasts, ochre on Films, coral on Teardowns. The hairline reads
   from the same token, so the accent cannot drift from the cards beneath it.

   NOTE: the spec asks for GREY on Films. Ours is ochre and has been since the
   page was built — left as is pending Lovable's call, since changing it
   silently would break the tier's relationship with its own cards. */
.gsh .dt-grp { --tier-accent: var(--gs-rule-2); }
.gsh .dt-grp:has(.pl-tone-green)  { --tier-accent: #1B6B38; }
.gsh .dt-grp:has(.pl-tone-blue)   { --tier-accent: #1E3456; }
.gsh .dt-grp:has(.pl-tone-ochre)  { --tier-accent: #6B4412; }
.gsh .dt-grp:has(.pl-tone-coral)  { --tier-accent: #7F3025; }
.gsh[data-theme="dark"] .dt-grp:has(.pl-tone-green) { --tier-accent: #4ADE80; }
.gsh[data-theme="dark"] .dt-grp:has(.pl-tone-blue)  { --tier-accent: #60A5FA; }
.gsh[data-theme="dark"] .dt-grp:has(.pl-tone-ochre) { --tier-accent: #FBBF24; }
.gsh[data-theme="dark"] .dt-grp:has(.pl-tone-coral) { --tier-accent: #FB7185; }
.gsh .dt-grp-h {
  border-top: 1px solid var(--tier-accent);
  padding-top: .85rem;
}

/* --- 3. Kyle's pick: dashed chassis, "Dropping next" ---------------------
   Still unlinked — a `<div>`, no href, no dummy URL, per the approved
   decision that the rotating slot carries no Open affordance. */


/* --- 4. Per-tier carousel on phones --------------------------------------
   Same pattern locked on /tools and /games: snap scroller below 720, one
   swipe per card, page itself never scrolls sideways. `.pl-grid` sits inside
   `.wrap`, which carries the page gutter, so the bleed is the usual
   -gutter/+gutter pair. */
@media (width <= 719px) {
  .gsh .pl-grid {
    display: grid;
    grid-auto-flow: column;
    grid-template-columns: none;
    grid-auto-columns: 82%;
    gap: 12px;
    overflow-x: auto;
    scroll-snap-type: x mandatory;
    overscroll-behavior-x: contain;
    scrollbar-width: none;
    margin-inline: -20px;
    padding: 0 20px 12px;
  }
  .gsh .pl-grid::-webkit-scrollbar { display: none; }
  
}

/* --- 5. Anchor clearance for /films --------------------------------------
   `/films` now 301s to `/playlists#films`, so the Films tier needs the same
   sticky-header clearance the other in-page anchors use. */
.gsh #films, .gsh #watch { scroll-margin-top: calc(var(--hd-h, 72px) + 1.5rem); }

/* ---------------------------------------------------------------------------
   29. Phase 2 Step 4 — /shelf directory
   ---------------------------------------------------------------------------
   The page ran a 7-bucket A–G taxonomy (toolkit / watch / listen / read /
   join / spaces / teardowns) with a lettered key chip on every heading and
   jump link. It is now five groups matching the header drawer, numbered and
   slashed like every other kicker on the site.

   Five, not four. Lovable's four groups had no home for the Read bucket
   (Lenny's Newsletter, SaaStr) or the Join bucket (Exit Five, RevGenius) —
   four curated external picks that exist nowhere else on the site, and the
   targets of two live "Explore all" links on the homepage. Kyle's call: keep
   them as `05 // Reading & rooms`. Those two homepage links now point at
   `#reading`.

   Watch and Listen were retired without loss: their items (Acquret, Y
   Combinator, Lenny's Podcast) are already on /playlists, so the Downtime
   group links to that page instead of duplicating its contents.

   `lab-sh-key` — the lettered chip — is gone from the markup entirely. */
.gsh .shp-n {
  white-space: nowrap;
  font-variant-numeric: tabular-nums;
  color: var(--gs-ink-3);
}
.gsh .shp-slash {
  font-style: normal;
  color: var(--gs-green-active, #1B6B38);
  margin: 0 .5ch 0 .45ch;
  letter-spacing: .06em;
}
.gsh[data-theme="dark"] .shp-slash { color: var(--gs-green, #35BD6B); }

/* Group headings: mono number and slashes, then the name in the heading face.
   No chip, no pip — the number is type, like every other kicker on the site. */
.gsh .shp-sec-head h2 .shp-n {
  font: 500 11px/1 var(--gs-mono);
  letter-spacing: .1em;
  text-transform: uppercase;
}

/* The jump bar was a row of lettered pills. Now typographic anchors: the
   number and slashes in green mono, the name in the UI face, nothing boxed. */
/* `gs-site-lab.css:3928` is `.gsh[data-lab] .shp-jump a` — a bordered, filled
   pill at specificity (0,3,1). That file is locked byte-identical, so this
   matches its specificity and wins on source order rather than being edited
   there. A plain `.gsh .shp-jump a` (0,2,1) lost and changed nothing. */
.gsh[data-lab] .shp-jump a {
  display: inline-flex;
  align-items: baseline;
  gap: .1rem;
  border: 0;
  background: none;
  border-radius: 0;
  padding: 6px 0;
  margin-right: 1.4rem;
}
.gsh[data-lab] .shp-jump a .shp-n {
  font: 500 10.5px/1 var(--gs-mono);
  letter-spacing: .1em;
  text-transform: uppercase;
}
.gsh[data-lab] .shp-jump a:hover .shp-slash,
.gsh[data-lab] .shp-jump a:focus-visible .shp-slash { opacity: .7; }

/* Anchor clearance under the sticky header, same as /playlists. */
.gsh #kit, .gsh #downtime, .gsh #spaces, .gsh #stealth, .gsh #reading {
  scroll-margin-top: calc(var(--hd-h, 72px) + 1.5rem);
}

/* ---------------------------------------------------------------------------
   30. Phase 2 Step 5 — /quick-q
   ---------------------------------------------------------------------------
   Two classes in the spec do not exist here. `.qq-who i` and
   `.qq-reply .qq-lab i` were named as the homes of the green pips; the actual
   pip was `.qq-pill > i.qq-dot`, a 6px persona-coloured circle inside a
   bracketed pill on the card's top bar. Both are now gone from the markup —
   the dot deleted, the pill unwrapped — and the name `.qq-who` adopted for
   the operator span so our markup matches their vocabulary from here.

   `.qq-sys` does not exist either; the system block is `.qq-block` with a
   `.qq-lead` lede, already Caslon italic. */

/* Top bar: DISPATCH #001 · 11:42 AM // FELIX FUNNEL · MARKETING */


/* The bracket ornaments went with the pill. */


/* Filter tabs: typographic, not bulky pills. The active state carries by
   weight and an underline rule rather than a filled block, so the row reads
   as navigation rather than four buttons. Specificity matches
   `.gsh[data-lab] .qq-tab` in the locked sheet — a lower one would lose
   silently, as the /shelf jump bar did. */
.gsh[data-lab] .qq-tab {
  border: 0; background: none; border-radius: 0;
  padding: 6px 0; margin-right: 1.3rem;
  display: inline-flex; align-items: baseline; gap: .45rem;
  font: 500 11px/1 var(--gs-mono);
  letter-spacing: .12em; text-transform: uppercase;
  color: var(--gs-ink-3);
  border-bottom: 1px solid transparent;
}
.gsh[data-lab] .qq-tab:hover { color: var(--gs-ink); border-bottom-color: var(--gs-rule-2); }
.gsh[data-lab] .qq-tab.is-on {
  background: none; color: var(--gs-ink);
  border-bottom-color: var(--gs-green-active, #1B6B38);
}
.gsh[data-lab] .qq-tab.is-on i { color: var(--gs-ink-2); opacity: 1; }
/* The seat sits in an <i> straight after the name. In a single mono uppercase
   run the two read as one string — "FELIX FUNNEL MARKETING" — so the seat
   takes the same `·` separator the card top bar uses, and stays a step
   quieter than the name. */
.gsh[data-lab] .qq-tab i {
  font: 500 11px/1 var(--gs-mono);
  letter-spacing: .12em; text-transform: uppercase;
  color: var(--gs-ink-3);
}
.gsh[data-lab] .qq-tab i::before {
  content: "\00b7"; margin-right: .45rem; opacity: .75;
}
.gsh[data-theme="dark"][data-lab] .qq-tab.is-on {
  border-bottom-color: var(--gs-green, #35BD6B);
}

/* Action CTAs: tactile pills, not hollow wireframes. */
.gsh[data-lab] .qq-btn {
  display: inline-flex; align-items: center; gap: .4rem;
  padding: 9px 16px; border-radius: 999px;
  border: 1px solid var(--gs-rule-2);
  background: var(--gs-raised);
  color: var(--gs-ink);
  font: 700 12.5px/1 var(--gs-head);
  transition: border-color .15s ease, color .15s ease, transform .15s ease;
}
.gsh[data-lab] .qq-btn:hover {
  opacity: 1;
  border-color: var(--gs-green-active, #1B6B38);
  color: var(--gs-green-active, #1B6B38);
  transform: translateY(-1px);
}
.gsh[data-lab] .qq-btn.is-done {
  border-color: var(--gs-green-active, #1B6B38);
  color: var(--gs-green-active, #1B6B38);
}
@media (prefers-reduced-motion: reduce) {
  .gsh[data-lab] .qq-btn { transition: none; }
  .gsh[data-lab] .qq-btn:hover { transform: none; }
}

/* Cards size to their content — no fixed heights, so nothing clips or
   truncates when one dispatch runs longer than its row partner. */
.gsh[data-lab] .qq-card { height: auto; }
.gsh[data-lab] .qq-text { overflow: visible; text-overflow: clip; }

/* ---------------------------------------------------------------------------
   31. Thin-stroke icons replace unicode glyphs (5 Nov)
   ---------------------------------------------------------------------------
   Locked with Lovable: U+2714 (heavy check) and U+2197 (NE arrow) carry a
   default emoji presentation and render as colour blocks on iOS; the rest
   vary by platform. `.gs-ic` matches `.gs-arr`'s weight so a tick beside an
   arrow reads as one family.

   Not converted, deliberately: the `×` in "GrowthSanity × Acquret" and the
   `→` in "Visit → lead CVR" are typography, not icons — an SVG there would
   break the phrase. And the directional labels inside the game zones are read
   by the engines via textContent, so they stay text under the
   engines-untouched invariant. */
.gs-ic {
  display: inline-block;
  flex: none;
  vertical-align: -0.08em;
  margin-left: 0.4em;
}
.gs-ic-left { margin-left: 0; margin-right: 0.4em; }

/* ---------------------------------------------------------------------------
   32. The shared Acquret dock
   ---------------------------------------------------------------------------
   One component on /tools, /quick-q, /games and /playlists, replacing the
   inline-styled callout that lived only on /tools. The inline rules had to
   come out first — they beat any stylesheet, so a shared class could never
   have applied while they were there.

   Each page passes its own `?ref=growthsanity-<page>-callout`, which is the
   only thing that differs between instances. */
.gs-dock {
  display: flex; flex-wrap: wrap; align-items: center;
  justify-content: space-between; gap: 1.5rem;
  max-width: 1040px; margin: 2rem auto 0;
  padding: 2rem 2.25rem;
  border-radius: 16px;
  background: var(--gs-surface);
}
.gs-dock-copy { min-width: 260px; max-width: 580px; }
.gs-dock-h {
  margin: 0 0 .4rem; color: var(--gs-ink);
  font: 700 19px/1.3 var(--gs-head);
}
.gs-dock-p {
  font: 400 14px/1.55 var(--gs-ui);
}
/* `.gsh a { color: inherit }` in the locked gs-site-lab.css is (0,1,1) and
   beat a bare `.gs-dock-cta` at (0,1,0), so the label took `--gs-ink` on the
   same `--gs-ink` ground. Measured contrast 1.0 — identical colour and
   background — on /quick-q, /games and /playlists, in BOTH themes. /tools
   escaped only because its dock sits outside `.gsh`.

   `.gsh .gs-dock-cta` is (0,2,0), which beats (0,1,1) on its own. No
   `!important` needed: raising specificity is the fix, and an `!important`
   here would be a hammer that hides the next collision instead of losing to
   it visibly. */
.gsh .gs-dock-cta,
.gs-dock-cta {
  display: inline-flex; align-items: center; white-space: nowrap;
  padding: .75rem 1.4rem; border-radius: 999px;
  background: var(--gs-ink); color: var(--gs-ground);
  font: 700 13px/1 var(--gs-head); letter-spacing: .02em;
  text-decoration: none;
  transition: transform .15s ease, opacity .15s ease;
}
.gsh .gs-dock-cta:hover,
.gs-dock-cta:hover { transform: translateY(-1px); opacity: .92; }
@media (prefers-reduced-motion: reduce) {
  .gs-dock-cta { transition: none; }
  .gs-dock-cta:hover { transform: none; }
}
@media (width <= 560px) {
  .gs-dock { padding: 1.5rem 1.25rem; }
  .gs-dock-cta { width: 100%; justify-content: center; }
}

/* Film years sit quieter than the title they qualify. */

/* The nomination card is a prompt, not a destination — dashed like Kyle's
   pick, so "unlinked standby" and "send us something" read as the same
   species rather than two different kinds of dead card. */


/* ---------------------------------------------------------------------------
   33. /quick-q — compact card anatomy
   ---------------------------------------------------------------------------
   Lovable asked for cards "~38% shorter". Measured first: at 375px a card is
   670px, and the body copy alone accounts for 314px of it — the subtext
   paragraph is 180px, the reply text 134px. Chrome (padding, gaps, labels,
   the watermark) is about 190px.

   So 38% — a 255px cut — cannot come from density. Even zeroing every pad
   and gap would not reach it. The remaining distance is copy length, which
   is Kyle's call, not a CSS one.

   What this does is take out the slack: tighter leading, a half-step smaller
   body, and every pad, gap and label margin trimmed. Measured result is
   reported honestly in the changelog rather than rounded up to the brief. */

/* Body copy: 1.6 leading to 1.45, 14px to 13.5. The two long paragraphs are
   where the height actually is. */
.gsh[data-lab] .qq-block p,
.gsh[data-lab] .qq-text {
  font-size: 13.5px;
  line-height: 1.45;
}


/* Chrome: every gap and pad down a step. */
.gsh[data-lab] .qq-card { padding: 14px; gap: 10px; }


.gsh[data-lab] .qq-reply { padding: 11px 13px; }

.gsh[data-lab] .qq-btn { padding: 7px 14px; }


/* ---------------------------------------------------------------------------
   34. Homepage hero — desk props hold Lovable's Felix @ 10 AM geometry
   ---------------------------------------------------------------------------
   Kyle: the cards should look messy but be messy in the SAME places, and
   Felix @ 10 AM on desktop is the reference to replicate. Lovable's design
   stands — only the alignment needed to become uniform.

   Measured all 4 personas x 4 time stops before changing anything. The drift
   has three causes, all emergent from the locked gs-site-lab.css meeting copy
   of different lengths:

   1. `.lab-flank` is a fixed-height flex column with
      `justify-content: space-between`. The first card pins to the top, the
      last pins to the BOTTOM, and the middle one floats in the remainder, so
      one extra wrapped line moves the two cards below it. Measured: `lab-dm`
      top ran 568 -> 608.1 while its bottom edge stayed at 682.5 in every
      state — bottom-pinned, confirmed. `lab-invite` 456.3 -> 479.9,
      `lab-ticket` 566.9 -> 581.8.
   2. Cards are shrink-to-fit, so width tracked copy length (158.4 -> 240.7).
      `lab-meet` sits in a `flex-end` column, so a narrower card pushes its
      LEFT edge right — 161.8 -> 216.0, a 54px jump, right edge constant.
   3. Height itself (74.3 -> 122.1) feeding back into (1).

   Fix: pin each prop to the geometry Lovable gives it at Felix @ 10 AM —
   its own width, its own horizontal anchor, its own top — so every persona
   and every time stop inherits that one arrangement. The props go
   `position: absolute` within the flank; `--x` / `--y` / `--r` are transforms
   and still paint on top untouched, so the deliberate mess is unchanged.

   Widths are per slot and NOT uniform, because Lovable's varied widths are
   the design: at >=1321px the reference measures meet 215, invite 220, dm
   236, quirk 192, sticky 190, ticket 236 against a 236px flank. They are
   written as percentages of the flank so they scale with it.

   SUPERSEDED 7 Nov, twice over, and left here because the reasoning below
   is still what the rules are arguing against. This originally ran two
   bands, mirroring the locked file's two compositions (236px flank from
   1321px, 210px flank at 1101-1320 with flank-r flipped to `flex-end`),
   and followed whichever edge the locked CSS anchored each card to. Both
   ideas are gone: the cards are uniform-width and left-anchored, and one
   236px band now covers 1101px up. Below 1101px the locked CSS still sets
   `.lab-flank { display: none }` — measured flank height 0 at 900px — so
   there are no floating cards to align and nothing is scoped there.

   Spacing note, stated honestly: holding Lovable's narrower widths means the
   wordiest states wrap one line further than they do at full width, so a few
   slots need slightly more room than Felix @ 10 AM happens to use. Tops are
   therefore Felix's spacing opened by 7-8px where a measured state would
   otherwise collide (invite 141 -> 148, dm 274 -> 282, ticket 263 -> 272).
   Without that, measured overlaps were 13.3px on meet/invite and 5.3px on
   sticky/ticket; a second measured pass opened dm and ticket in the lower
   band by a further 6px each to clear 3.5px and 0.9px residuals. Rotation is included in the margin: a rotated card's painted
   box is ~6-8px taller than its layout height.

   The flank's width is restated in both bands. It had been content-driven —
   set by the cards' own max-width — and taking them out of flow removes that
   basis, letting the `1fr` grid column collapse the flank to 95px.

   Specificity: the locked rules to beat are `.gsh .hero .lab-prop` (0,3,0),
   `.gsh .hero .lab-prop.lab-quirk` (0,4,0) and `.gsh .hero .lab-flank`
   (0,3,0). These are (0,4,0) and (0,6,0). Matching specificity is deliberate
   — loading later is not enough against the locked file. */

/* --- Desktop proper: Lovable's 236px flank composition (their line 3277).

   Boundary lowered 1321px -> 1200px on Lovable's ruling of 7 Nov, so the full
   236px composition reaches standard 1280 laptops. Their reason: the copy
   already fits at that width, and trimming 20 strings to suit one breakpoint
   solves it in the harder place.

   `margin-top: -24px` is carried into this block because the boundary moved.
   The locked file only applies their -24px correction from 1321px up; below
   it, it still sets `margin-top: -99px`, which yanks the props 80px over the
   H1. Extending the width band without the margin would have pulled that
   collision up into 1200-1320 along with it. At >=1321 this is a no-op — the
   locked file already sets the same value. --- */
@media (width >= 1101px) {
  .gsh[data-lab] .hero .lab-flank { position: relative; width: 236px; margin-top: -24px; }

  /* Widening the flank alone changed nothing: the locked file caps every prop
     at `max-width: 210px` from 1101-1320px (its own band), so cards sized as
     a percentage of a 236px flank were clipped straight back to 210 and the
     line counts did not move. The cap has to be lifted too.

     Only `max-width` is lifted. The same locked rule also sets
     `font-size: 12.5px`, but that is identical to the value already in force
     from 720px up — restating it would imply a difference that is not there.

     (0,4,0) against the locked (0,3,0). */
  .gsh[data-lab] .hero .lab-prop { max-width: 236px; }

  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-meet,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-invite,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-dm,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-quirk,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-sticky,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-ticket {
    position: absolute; bottom: auto; align-content: start;
  }

  /* UNIFIED GEOMETRY, 7 Nov, on Lovable's brief. Supersedes the per-slot
     widths and staggered tops above.

     Their diagnosis, confirmed against the file: the right flank stepped
     192 / 190 / 236px, so its outer edge ran as a staircase while the left
     edge ran ragged (meet pinned right at 91.10%, invite pinned left at
     93.22%, dm at 100%); and the two flanks sat on different baselines —
     left tier 2 at 148px against right at 134px, tier 3 at 282 against 272.

     All six cards now take the full flank width and share three baselines.
     Both outer edges go flush and the tiers line up across the hero.

     Full width is also what un-orphans "stopped." on the sticky card: the
     headline wrapped at 80.51% and does not at 100%.

     Note this REVERSES the reasoning in the header above, which held that
     Lovable's varied widths were the design and wrote them as percentages
     to preserve them. That was right about the intent and wrong about the
     result — the varied widths read as misalignment, not as deliberate
     mess. The mess that was wanted lives in `--r`, and `--r` is a transform
     applied on top of this box, so the organic tilt survives the grid being
     made uniform. Nothing here touches it.

     The side anchors are dropped rather than kept. At `width: 100%`,
     `left: 0` and `right: 0` describe the same box, so a mix of the two
     implied a distinction that no longer exists; `right: auto` clears the
     ones the superseded rules set.

     Tops: tier 2 moves to 142px, between the 148 and 134 it replaces, and
     tier 3 holds at 282. The header above opened these by 7-8px to clear
     measured collisions caused by narrow cards wrapping an extra line.
     Full width removes that wrapping, which is why the tighter baseline is
     affordable -- verified by measurement across all 16 states, not assumed.

     (0,5,0) against the locked `.gsh .hero .lab-prop.lab-quirk` (0,4,0). */
  .gsh[data-lab] .hero .lab-flank .lab-prop { left: 0; right: auto; width: 100%; }

  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-meet,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-quirk  { top: 0; }
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-invite,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-sticky { top: 142px; }
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-dm,
  .gsh[data-lab] .hero .lab-flank .lab-prop.lab-ticket { top: 282px; }
}

/* --- The 210px band is gone, 7 Nov, on Lovable's ruling.
   It used to re-run the whole composition at 1101-1199px on a 210px flank,
   with tops at 0 / 148 / 294 because 142 / 282 overlapped by 1.4px there.
   The band above now starts at 1101px instead of 1200px, so one 236px
   composition and one set of baselines cover every width the flanks appear
   at, and the `max-width: 236px` lift in that block carries down with it.

   Lovable's reason for taking this over keeping two bands: at 210px the
   Calendar headline ("Six meetings back to back") wrapped, and no set of
   tops fixes a width problem. 236px is what puts it back on one line.

   It fits. Measured before proposing it: at 1101px a 236px flank gives a
   978px stage inside the viewport with 61px of side margin and no overflow,
   because `.hero-art` is a fixed 440px and only the side margins absorb the
   change. Checked again at 1101 and 1199 after the merge.

   Two things the deleted block carried, and where they went:

   - `margin-top: -24px`. The locked file leaves `.lab-flank` on
     `margin-top: -99px` below 1321px, which yanks the props 80px over the
     H1; Lovable corrected it to -24px but only from 1321px up. The deleted
     block carried their value down to 1101. The surviving block sets the
     same -24px and now reaches 1101 itself, so nothing is lost.

   - `padding-bottom: 1.5rem`. Dropped. The props are absolutely positioned,
     so flank padding does not hold them; it only padded a box whose height
     the grid row already sets. Verified after removal that the deepest card
     still sits inside the hero at 1101 and 1199. --- */

/* ---------------------------------------------------------------------------
   35. Mobile nav drawer — align the panel to the header bar
   ---------------------------------------------------------------------------
   Kyle, from a phone: the shelf drawer "is still looking messy". Checked
   Lovable's specs first — their mobile drawer work is Phase 2 Step 1, four
   items, and all four are built and live (number chips gone, arrows pinned
   right and centred, group 03 on one line, CTA contrast matched to the
   desktop dropdown). Nothing in Steps 2-5 or Phase 3 touched this panel. So
   this is unspecced structure from the original export, not a missed note.

   Measured at 390x844: the panel is `position: absolute; right: 0` inside
   `.lab-mnav` — the 36px hamburger — and sized by `min-width: min(280px,
   100vw - 32px)`. So it anchors to the button, not to anything visible. It
   ran 81 -> 361 while the header bar `.hd-in` runs 20 -> 370: a 61px dead
   gutter on the left, stopping 9px short on the right. It aligns to nothing,
   which is why it reads as floating over the hero.

   Three symptoms, one cause — 280px is too narrow for the content:
     - 4 of the 6 directory items wrapped to two lines (56/73/73/56/71/71)
     - content ran 855px against a 758px cap, overflowing by 97px
     - so it scrolled internally and sliced text mid-line at the top edge

   Fix: match the header bar. `.wrap` gives a constant 20px inset at 320, 360,
   390 and 430, so the bar is always `100vw - 40px` wide. The panel's
   containing block is the hamburger, whose right edge sits 9px inside the
   bar's — `.hd-in` has `padding-right: 8px` plus its 1px border — so
   `right: -9px` lands the panel flush with the bar. Both numbers are read off
   the locked CSS, not eyeballed.

   Measured after, at 390x844: panel 20 -> 370, exactly the bar. **All six
   items single-line**, two-line wraps 4 -> 0. Overflow 97px -> 14px.

   `max-width: 420px` caps the tablet end. The hamburger still shows up to
   720px, where a full-bar panel would be 680px wide and absurd; capped, it
   stays a right-anchored dropdown (420 wide, flush right, zero overflow at
   both 600 and 720).

   Floored at 360px deliberately. At 320px the hamburger itself sits OUTSIDE
   the header bar — bar ends at 300, the button runs to 313 — a pre-existing
   overflow in the locked header that has nothing to do with this panel.
   Anchoring to the button there would push the panel 2px past the viewport,
   so below 360px everything is left exactly as it is today. Every current
   phone is 360 or wider.

   Specificity: the locked rules are `.gsh .lab-mnav-panel` at (0,2,0) on
   lines 1922, 3220 and 3493. This matches at (0,2,0) and loads later. */
@media (360px <= width <= 720px) {
  .gsh .lab-mnav-panel {
    width: calc(100vw - 40px);
    max-width: 420px;
    right: -9px;
  }
}

/* ---------------------------------------------------------------------------
   36. Homepage "Explore the tools" lands clean, like "Restore sanity" does
   ---------------------------------------------------------------------------
   Kyle, from desktop and phone: after clicking the second-fold CTA you can
   still see the previous fold peeking at the top, while the first-fold
   "Restore sanity" jump "shows much cleaner to the next fold".

   He is right, and it is a straight asymmetry in the locked gs-site-lab.css.
   The homepage has exactly two in-page anchors and they are built
   differently:

                        #desk (Restore sanity)   #tools (Explore the tools)
     scroll-margin-top   0  (line 3811)          100px desktop / 94px phone
     own padding-top     96px (--hd-h + 24)      58.5px desktop / 40px phone

   `#desk` is special-cased on line 3811: it lands at viewport y=0 and clears
   the header with its own padding, so its background fills the top of the
   screen and nothing can show above it. `#tools` falls through to the generic
   `.gsh[data-lab] [id]` rule and lands 100px down. The header is a FLOATING
   pill with transparent gaps around it, not a full-width bar, so the previous
   section shows straight through that strip.

   Measured after a real click, desktop 1280x900: #tools top at y=100, the
   previous section's bottom at y=100, and the "Explore the tools" button the
   user just pressed still painted at y=40-80 — right around the pill. On
   phone, 5 separate elements of the previous fold were still on screen. The
   button staying visible is the real defect: it reads as "the click did
   nothing".

   Fix: give #tools the treatment Lovable already wrote for #desk. Nothing
   invented — the same two declarations, the same `--hd-h + 24` clearance.

   `!important` is required and is not a shortcut here: the locked file sets
   this padding with `!important` at
   `.gsh[data-lab] section.sec.tl-sec` (0,4,1), which is exactly why Lovable's
   own #desk line carries `!important` too. Our selector is (1,2,0) — an id
   outranks any number of classes — so the cascade is won on specificity and
   the `!important` only matches theirs rather than hiding a collision.

   Verified: #tools now lands identically to #desk, on both.

                        #desk        #tools after
     lands at           y=0          y=0
     previous fold      0 elements   0 elements
     heading clears pill 22px        22px

   Phone 390x844: leaked elements 5 -> 0, heading clearance 22px, matching
   #desk's own 22px there. The cost, stated plainly: the tools fold gains top
   padding (58.5 -> 96 desktop, 40 -> 90 phone), so on an ordinary scroll
   there is ~40px more air above the heading, and after a jump the heading
   sits 22px under the pill instead of 66px. Both are the numbers #desk
   already uses — the fold Kyle picked out as the clean one. */
.gsh[data-lab] #tools {
  scroll-margin-top: 0 !important;
  padding-top: calc(var(--hd-h, 72px) + 24px) !important;
}

/* ---------------------------------------------------------------------------
   37. The desk fold gives its CTA the same room the hero gives its own
   ---------------------------------------------------------------------------
   Kyle, from an iPhone 17 Pro Max after §36 shipped: "this now has a huge gap
   on phone and the CTA is so close."

   Both halves of that are right, and measuring the fold he does NOT complain
   about is what explains it. At 440x956:

                              CTA -> fold edge    CTA -> next heading
     hero -> desk  (fine)           53px                143px
     desk -> tools (wrong)          20px                110px

   At 1280x900 the same comparison is starker: 115px vs 20px.

   So the problem is not the size of the gap. The transition Kyle is happy
   with has the LARGER total gap — 143 against 110 on phone, 211 against 116
   on desktop. What differs is where the space sits. The hero gives its CTA
   room to breathe and then changes fold. The desk fold crams its CTA 20px
   from the boundary and puts all the air on the far side, where §36's
   header clearance made it a conspicuous empty band in the tools fold's
   cream. Hence both halves of his sentence: crammed button, huge gap.

   §36 is not the cause — that 20px has always been there — but §36 doubled
   the air on the other side and so made the imbalance impossible to miss.
   Stated plainly because the fix for it is not to undo §36: reverting would
   bring back the button that stays on screen after its own click, which is
   the worse defect.

   Fix: give the desk fold the same bottom room the hero gives its CTA, so
   the two fold transitions read the same way. `clamp(53px, 9vw, 115px)`
   lands on the hero's measured values at both ends — 9vw is 115.2 at 1280,
   and the 53px floor takes over below ~590px where 9vw would undershoot.

   `!important` matches the locked `.gsh section.sec.dk-sec` (0,3,1), which
   sets this padding with `!important`. Ours is (1,2,0) — an id outranks any
   number of classes — so specificity wins it and the `!important` only
   matches theirs rather than hiding a collision.

   Note this changes resting layout only. The jump targets are unaffected:
   #desk's landing depends on its padding-TOP, and #tools' on its own. Both
   still land at y=0 with zero leakage. */
.gsh[data-lab] #desk {
  padding-bottom: clamp(53px, 9vw, 115px) !important;
}

/* ---------------------------------------------------------------------------
   38. /quick-q — the filter row stays one line on a phone
   ---------------------------------------------------------------------------
   Lovable, from their live-site audit: at 375-390px the category filters wrap
   and leave "Leadership" orphaned on a second row.

   Reproduced at 375x812 before changing anything: the five tabs sit on two
   rows, four at y=315 and Leadership alone at y=347.

   The cause is ours, not theirs, and not the `gap`. §30 gave `.qq-tab` a
   `margin-right: 1.3rem` to space the typographic tabs, and that margin sits
   ON TOP of the container's `gap: 8px` — so the real spacing is 28.8px, not
   8px. Measured: tabs total 269.3px and need 269.3 + 4 x 28.8 = 384.5px
   against a 335px bar. It was always going to wrap; 375px is simply where it
   first shows.

   Fix is Lovable's: one horizontal scrolling row. This is the same carousel
   pattern the site already uses for `.tl-grid`, `.gh-grid` and `.lab-sh-grid`
   on phones, including the `-pad / +pad` bleed pair so the row runs edge to
   edge while its first and last tabs still line up with the page gutter.

   `flex: 0 0 auto` is required: without it the tabs shrink to fit instead of
   scrolling, which trades a wrap for truncated labels.

   Scoped to 460px, not Lovable's 430. They reported the wrap at 375-390 and
   capped there, but measured it still wraps at 431. The row needs 384px of
   tabs AND the fifth tab's own trailing `margin-right: 20.8px`, which flex
   counts in the line width — so 405px against a bar of `100vw - 40px`, and
   the wrap only clears at a 445px viewport. 460 leaves headroom for font
   rendering that differs by device. Between 445 and 460 the rule is inert:
   `nowrap` with content that already fits lays out identically to the
   static row and shows no scrollbar. */
@media (width <= 460px) {
  .gsh[data-lab] .qq-tabs {
    flex-wrap: nowrap;
    overflow-x: auto;
    overscroll-behavior-x: contain;
    scrollbar-width: none;
    -webkit-overflow-scrolling: touch;
    margin-inline: calc(var(--gs-pad, 20px) * -1);
    padding-inline: var(--gs-pad, 20px);
  }
  .gsh[data-lab] .qq-tabs::-webkit-scrollbar { width: 0; height: 0; display: none; }
  .gsh[data-lab] .qq-tab { flex: 0 0 auto; }
}

/* ---------------------------------------------------------------------------
   39. Shelf V2 — Lovable's visual elevation (Step 1: /games)
   ---------------------------------------------------------------------------
   Lovable's block, applied with the five decisions they approved on 6 Nov.

   DECISION 1 — `[data-lab]` dropped from every selector. Their block was
   scoped `.gsh[data-lab] .v2-*`. Measured on production: the `.gsh` wrapping
   the content of `/games` and `/tools` carries NO `data-lab` — only
   `.gsh-chrome` (header) and `.tk-mqwrap` (marquee) do. So every V2 rule
   would have matched nothing on those two pages while working on the two
   exported ones, from one stylesheet. Adding `data-lab` instead was not an
   option: it gates 270 selectors in the locked sheet and 165 of ours, and
   switching it on for two pages that never had it applies all 435 at once.
   `v2-` appears zero times in gs-site-lab.css, gs-static-fix.css and
   site.css, so the attribute bought nothing and cost two pages.

   DECISION 3 — the "the call" lines stay sentence case in the markup and are
   uppercased HERE, by CSS. Their HTML had them typed in capitals, which
   hard-codes the shouting into the content: unreversible from CSS and
   announced as capitals by screen readers.

   DECISION 4 — SUPERSEDED BY MEASUREMENT, and Lovable are told why. They
   asked us to add the `.gsh` prefix to the five rules they left unprefixed.
   We did, then built Step 2 and found the opposite problem to Step 1: on
   `/tools` the hero, the spec bays and the dock all sit OUTSIDE `.gsh`
   entirely — only the tool consoles are wrapped, in a `.gsh.tk-toolwrap`
   island. So `.gsh`-scoped V2 rules reach nothing there, and dropping
   `[data-lab]` alone was necessary but not sufficient.

   Every V2 selector is therefore unscoped. `v2-` collides with nothing in
   any of the three stylesheets, so the prefix was defensive rather than
   load-bearing, and the two page families place `.gsh` too differently for
   any single ancestor to serve both: `/games` wraps its whole main in `.gsh`,
   `/tools` wraps only an island inside it. The `--v2-*` tokens move to
   `:root` for the same reason, with `html[data-theme="dark"]` beside the
   `.gsh[data-theme="dark"]` variant so both families theme correctly. The
   `.gs-dock` restyle is unscoped on the same evidence — §32's base rule is
   unscoped too, so ours matches its specificity and wins on order.

   DECISION 2 — the dock is NOT duplicated. `.v2-dock` would have been a
   second component beside the `.gs-dock` already on all four pages — the one
   whose invisible CTA Lovable caught and §32 fixed. Instead the V2 treatment
   is applied to the existing `.gs-dock` container, heading and paragraph
   below. `.gs-dock-cta` is deliberately untouched: it already computes the
   right pill colours in both themes and that fix is not worth re-testing for
   a restyle that does not need it.

   DECISION 5 — `--gs-head` holds a Helvetica Neue stack site-wide and stays
   that way. Lovable confirmed Helvetica Neue Bold for all headings is a
   locked brand rule, so V2 inheriting it is correct, not a bug. Recorded here
   because the token name invites "fixing" it to Space Grotesk later.

   Live after Step 1: the arcade cabinets, the pill, the dock restyle and the
   mobile carousel. The plate, slip, tape, reel and tier rules below are
   Lovable's spec for Steps 2-4 and are inert until those pages are converted.
   --------------------------------------------------------------------------- */
:root, .gsh {
  --v2-ochre: #9A5B12;
  --v2-shell: #2A2620;
  --v2-shell-ink: #F2ECE4;
  --v2-screen: #14161E;
  --v2-screen-ink: #E9F5EC;
  --v2-glow: #5FD08A;
  --v2-bezel: #2A2620;
  --v2-border: rgba(255, 255, 255, .12);
  --v2-rim: transparent;
}
html[data-theme="dark"], .gsh[data-theme="dark"] {
  --v2-ochre: #E2A65A;
  --v2-shell: #161822;
  --v2-shell-ink: #F2ECE4;
  --v2-screen: #0A0C12;
  --v2-screen-ink: #E9F5EC;
  --v2-glow: #5FD08A;
  --v2-bezel: #222634;
  --v2-border: rgba(255, 255, 255, .08);
  --v2-rim: rgba(255, 255, 255, .08);
}

.v2-kick { display: block; font-size: 12px; letter-spacing: .2em; text-transform: uppercase; color: var(--gs-ink-2); margin-bottom: 12px; }
.v2-kick b { color: var(--gs-green-active); font-weight: 600; }


.v2-priv, .v2-anat { margin: 18px 0 0; font-size: 12px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-green-active); }
.v2-anat span { color: var(--gs-ink-2); margin: 0 .5em; }

.v2-pill { display: inline-flex; align-items: center; gap: .5em; justify-content: center; padding: 12px 20px; border-radius: 999px; border: 0; cursor: pointer; font: 700 15px/1 var(--gs-head); color: var(--gs-ground); background: var(--gs-ink); text-decoration: none; min-height: 44px; }
.v2-ghost { display: inline-flex; align-items: center; gap: .5em; background: none; border: 0; cursor: pointer; font-size: 12px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-ink-2); min-height: 44px; }

/* --- Steps 2-4 spec, inert until those pages convert --- */
.v2-plate { border-radius: 18px; background: var(--gs-surface); border: 1px solid var(--gs-rule); box-shadow: inset 0 1px 0 color-mix(in oklab, var(--gs-raised) 80%, transparent), 0 14px 30px -24px var(--gs-ink); overflow: hidden; margin-top: 24px; }
.v2-plate-top { display: flex; justify-content: space-between; padding: 12px 20px; font-size: 11px; letter-spacing: .18em; text-transform: uppercase; color: var(--gs-ink-2); border-bottom: 1px dashed var(--gs-rule); }
.v2-plate-row { display: grid; grid-template-columns: repeat(4, 1fr); }
.v2-bay { display: grid; grid-template-rows: auto 1fr; gap: 10px; padding: 20px; border-left: 1px solid var(--gs-rule); }
.v2-bay:first-child { border-left: 0; }
.v2-bay-k { font-size: 11px; letter-spacing: .16em; text-transform: uppercase; color: var(--gs-ink-2); }
.v2-bay-k b { color: var(--gs-green-active); }
.v2-bay strong { font: 600 17px/1.3 var(--gs-ui); color: var(--gs-ink); }
@media (max-width: 900px) { .v2-plate-row { grid-template-columns: 1fr 1fr; } .v2-bay:nth-child(3) { border-left: 0; } .v2-bay:nth-child(n+3) { border-top: 1px solid var(--gs-rule); } }
@media (max-width: 520px) { .v2-bay { padding: 16px; } .v2-bay strong { font-size: 15px; } }


.v2-qq-grid { display: grid; grid-template-columns: repeat(2, 1fr); gap: 16px; }
.v2-slip { display: flex; flex-direction: column; gap: 12px; padding: 18px 20px; border-radius: 14px; background: var(--gs-raised); border: 1px solid var(--gs-rule); }
.v2-slip-bar { display: flex; justify-content: space-between; font-size: 11px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-ink-2); }
.v2-slip-bar b { color: var(--gs-red-text, #B23A31); font-weight: 600; }
.v2-ping { margin: 0; font: 700 19px/1.3 var(--gs-head); letter-spacing: -.01em; color: var(--gs-ink); }
.v2-sub { margin: 0; padding-left: 12px; border-left: 2px solid var(--v2-ochre); font: 400 italic 15px/1.45 var(--gs-serif); color: var(--gs-ink-2); }
.v2-sub span, .v2-reply span { display: block; font: 500 10px/1 var(--gs-mono); letter-spacing: .18em; text-transform: uppercase; font-style: normal; margin-bottom: 6px; color: var(--v2-ochre); }
.v2-reply span { color: var(--gs-green-active); }
.v2-reply p { margin: 0; font: 400 15px/1.5 var(--gs-ui); color: var(--gs-ink); }
.v2-slip-act { display: flex; align-items: center; justify-content: space-between; gap: 12px; margin-top: auto; padding-top: 12px; border-top: 1px dashed var(--gs-rule); }
@media (max-width: 760px) { .v2-qq-grid { grid-template-columns: 1fr; gap: 12px; } .v2-slip { padding: 16px; } .v2-ping { font-size: 17px; } }

.v2-snap { display: grid; grid-template-columns: repeat(3, 1fr); gap: 16px; margin: 20px 0 40px; }
.v2-tier-head { display: flex; align-items: baseline; justify-content: space-between; gap: 20px; flex-wrap: wrap; padding-top: 20px; border-top: 1px solid var(--gs-rule); }
.v2-tier-head .mono { font-size: 12px; letter-spacing: .2em; text-transform: uppercase; color: var(--gs-ink-2); }
.v2-tier-head b { color: var(--gs-green-active); }
.v2-tier-head blockquote { margin: 0; font: 400 italic clamp(17px, 2vw, 22px)/1.35 var(--gs-serif); color: var(--gs-ink); max-width: 40ch; }

.v2-tape, .v2-reel { display: flex; flex-direction: column; gap: 8px; padding: 18px 16px 16px; border-radius: 14px; background: var(--v2-shell); border: 1px solid var(--v2-border); color: var(--v2-shell-ink); text-decoration: none; }
.v2-tape h4, .v2-reel h4 { margin: 6px 0 0; font: 700 20px/1.2 var(--gs-head); letter-spacing: -.01em; color: var(--v2-shell-ink); }
.v2-tape p { margin: 0; color: color-mix(in oklab, var(--v2-shell-ink) 74%, transparent); line-height: 1.45; }
.v2-link { margin-top: auto; padding-top: 8px; display: inline-flex; align-items: center; gap: .5em; font-size: 12px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-green-active); }
.v2-shell { aspect-ratio: 16/9; border-radius: 10px; background: var(--v2-screen); padding: 10px; gap: 8px; }
.v2-label { padding: 6px 10px; border-radius: 4px; background: color-mix(in oklab, var(--v2-screen-ink) 12%, transparent); color: var(--v2-screen-ink); font-size: 10px; letter-spacing: .16em; text-transform: uppercase; border-top: 3px solid var(--v2-glow); }
.v2-window { flex: 1; align-items: center; justify-content: space-between; gap: 10px; padding: 0 14%; border-radius: 999px; background: color-mix(in oklab, var(--v2-screen-ink) 12%, var(--v2-screen)); }
.v2-window i { width: 30px; height: 30px; border-radius: 50%; border: 5px dotted var(--v2-screen-ink); animation: v2spin 6s linear infinite; }
.v2-window span { flex: 1; height: 2px; background: color-mix(in oklab, var(--v2-screen-ink) 40%, transparent); }
.v2-slot { flex: 1; display: grid; place-items: center; border: 1px dashed color-mix(in oklab, var(--v2-screen-ink) 40%, transparent); border-radius: 6px; color: color-mix(in oklab, var(--v2-screen-ink) 70%, transparent); font-size: 11px; letter-spacing: .16em; text-transform: uppercase; }
.v2-tape.is-empty { border-style: dashed; background: color-mix(in oklab, var(--v2-shell) 88%, transparent); }
@keyframes v2spin { to { transform: rotate(360deg); } }
@media (prefers-reduced-motion: reduce) { .v2-window i { animation: none; } }

.v2-frame { aspect-ratio: 2.39/1; justify-content: space-between; align-items: flex-end; padding: 10px 12px; border-radius: 8px; background: var(--v2-screen); color: var(--v2-ochre); font-size: 10px; letter-spacing: .18em; text-transform: uppercase; }
.v2-why { font: 400 italic 16px/1.4 var(--gs-serif); color: color-mix(in oklab, var(--v2-shell-ink) 82%, transparent); }
.v2-meta { font-size: 11px; letter-spacing: .14em; text-transform: uppercase; color: color-mix(in oklab, var(--v2-shell-ink) 62%, transparent); }

/* --- Step 1 live: the CRT arcade --- */
.v2-arc-grid { display: grid; grid-template-columns: repeat(4, 1fr); gap: 16px; margin: 20px 0 40px; }
.v2-arc { display: flex; flex-direction: column; gap: 10px; padding: 16px; border-radius: 18px; background: var(--gs-raised); border: 1px solid var(--gs-rule); border-bottom: 4px solid var(--gs-ink); color: var(--gs-ink); text-decoration: none; transition: transform .15s; }
.v2-arc:hover { transform: translateY(-2px); }
@media (prefers-reduced-motion: reduce) { .v2-arc { transition: none; } .v2-arc:hover { transform: none; } }
.v2-arc-bar { display: flex; justify-content: space-between; font-size: 11px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-ink-2); }
.v2-arc-bar b { color: var(--gs-green-active); }
.v2-arc-who { display: flex; gap: 10px; align-items: center; }
.v2-arc-who img { width: 48px; height: 48px; border-radius: 50%; object-fit: cover; object-position: 50% 0%; border: 2px solid var(--gs-rule); background: var(--gs-surface); }
.v2-arc-who strong { display: block; font: 600 15px/1.2 var(--gs-ui); }
.v2-arc-who span { font-size: 11px; color: var(--gs-ink-2); display: block; }
.v2-arc h4 { margin: 4px 0 0; font: 700 21px/1.15 var(--gs-head); letter-spacing: -.01em; }
.v2-arc h4 em { font-family: var(--gs-serif); font-weight: 400; color: var(--gs-indigo); }
.v2-arc > p { margin: 0; color: var(--gs-ink-2); line-height: 1.45; font-size: 15px; }
.v2-call { margin-top: auto; padding: 12px; border-radius: 10px; background: var(--v2-screen); color: var(--v2-screen-ink); }
.v2-call span { display: block; font-size: 10px; letter-spacing: .18em; text-transform: uppercase; color: var(--v2-glow); margin-bottom: 6px; }
/* Decision 3: the caps live here, not in the markup. */
.v2-call p { margin: 0; font-size: 12px; line-height: 1.5; font-family: var(--gs-mono); text-transform: uppercase; }
.v2-play { width: 100%; }

/* Locked baseline envelopes — stops the cards staggering across columns.
   Decision 4: scoped, where Lovable left these five unprefixed. */
@media (min-width: 760px) {
  .v2-arc-who { min-height: 48px; }
  .v2-arc h4  { min-height: 52px; }
  .v2-arc > p { min-height: 66px; }
  .v2-call    { min-height: 108px; }
  .v2-play    { margin-top: auto; }
}
@media (max-width: 1000px) and (min-width: 760px) {
  .v2-arc-grid { grid-template-columns: repeat(2, 1fr); }
}

/* Mobile carousel */
@media (max-width: 759px) {
  .v2-snap, .v2-arc-grid {
    display: flex; overflow-x: auto; scroll-snap-type: x mandatory;
    margin: 0 -20px; padding: 4px 20px 12px; scroll-padding: 0 20px; scrollbar-width: none;
    overscroll-behavior-x: contain;
  }
  .v2-snap::-webkit-scrollbar, .v2-arc-grid::-webkit-scrollbar { width: 0; height: 0; display: none; }
  .v2-snap > *, .v2-arc-grid > * { flex: 0 0 82%; scroll-snap-align: start; }
  .v2-arc-who, .v2-arc h4, .v2-arc > p, .v2-call { min-height: auto; }
}

/* Decision 2: the V2 dock treatment lands on the EXISTING .gs-dock rather
   than adding .v2-dock beside it. `.gs-dock-cta` is untouched — it already
   computes correct pill colours in both themes (§32) and a restyle is no
   reason to re-open a contrast fix Lovable had to catch for us once. */
.gs-dock {
  padding: 28px; border-radius: 18px;
  background: var(--gs-raised);
  border: 1px solid var(--gs-rule);
  border-top: 3px solid var(--gs-green-active);
  margin-top: 40px; gap: 24px;
}
.gs-dock-h { margin: 0 0 8px; font: 700 clamp(22px, 2.6vw, 28px)/1.15 var(--gs-head); letter-spacing: -.02em; }
.gs-dock-h em { font-family: var(--gs-serif); font-weight: 400; font-style: italic; color: var(--gs-indigo); }
.gs-dock-p { margin: 0; color: var(--gs-ink-2); max-width: 58ch; line-height: 1.5; font-size: 15px; }
@media (max-width: 720px) {
  .gs-dock { flex-direction: column; align-items: stretch; padding: 22px; }
}

/* Step 2 additions. The dock kicker was first built as `.gs-dock-kick`;
   Lovable then specced `.v2-kick` and asked for it on all four docks, so the
   markup uses their class and §39's `.v2-kick` rule carries it. The bespoke
   rule is gone rather than left as a second way to style the same thing.
   The dock's own kicker needs no green fallback: `.v2-kick b` already has
   one, and all four docks sit where `--gs-green-active` resolves. */
.gs-ic-lock { vertical-align: -0.12em; margin-right: .45em; }

/* The /tools hero sits outside both `.gsh` AND `#tools`, so §21's pins do not
   reach it. The V2 privacy line asks for `--gs-green-active`, which is
   undefined out there — invalid at computed-value time, so it silently
   inherited `--gs-ink` and shipped as near-white instead of green. Measured:
   rgb(237,239,245) where the spec-plate kickers next to it resolved green.
   Same two values as §21, extended to the hero rather than widened to
   `:root`, for the same reason §21 gives. */
.tk-hero { --gs-green-active: #1B6B38; }
html[data-theme="dark"] .tk-hero { --gs-green-active: oklch(78% .16 150); }

/* Step 3 additions. The ping stays a <blockquote> rather than becoming a <p>:
   it is a quoted message, and Lovable's `.v2-ping` rule styles whatever
   carries the class, so the semantics cost nothing. These two rules zero the
   default margins their spec assumed away. */
.v2-ping p, .v2-sub p { margin: 0; }
.v2-slip-who { color: var(--gs-ink-2); }
@media (max-width: 420px) { .v2-slip-who { display: none; } }
/* The copied-state span is absolutely placed over the label in §30; inside a
   V2 button it needs the same relative parent. */
.v2-pill, .v2-ghost { position: relative; }

/* The site's `.gs-arr` is `display: block`, which is fine inside the flex
   buttons it was built for and wrong in running text: in the /quick-q anatomy
   line each arrow forced a line break, stacking "The Ping / The Subtext /
   The Reply" into three rows 88px tall in a 720px container. Inline only
   where the arrow sits in flow text. */
.v2-anat svg, .v2-priv svg { display: inline-block; vertical-align: -0.08em; }
.v2-anat span { white-space: nowrap; }

/* ---------------------------------------------------------------------------
   40. Shelf V2 Step 4 — /playlists hi-fi archive
   ---------------------------------------------------------------------------
   Tier accent hairlines, per Lovable's spec: green for the cassettes, ochre
   for the film reels, indigo for long-form audio, red for screen sessions.
   Applied to the tier heading rather than each card, so the accent reads as
   a shelf label rather than decoration repeated twelve times. */
#music  .dt-grp-h { border-top: 2px solid var(--gs-green-active, #1B6B38); }
#films  .dt-grp-h { border-top: 2px solid var(--v2-ochre); }
#audio  .dt-grp-h { border-top: 2px solid var(--gs-indigo, #485289); }
#watch  .dt-grp-h { border-top: 2px solid var(--gs-red, #C8443A); }
#music .dt-grp-h, #films .dt-grp-h, #audio .dt-grp-h, #watch .dt-grp-h { padding-top: 18px; }

/* The cassette shell and its parts are spans inside an <a>, so they need the
   box behaviour their spec assumed from divs. */
.v2-shell, .v2-label, .v2-window, .v2-slot, .v2-frame { display: flex; }
.v2-shell { flex-direction: column; }
.v2-label { justify-content: space-between; }
.v2-tape, .v2-reel { height: 100%; }
.v2-reel h4 .v2-meta { font: inherit; font-weight: 400; color: color-mix(in oklab, var(--v2-shell-ink) 62%, transparent); }
/* Lovable's `.v2-why` is a shared italic; the first line of a reel is the
   one-liner and the second the editorial note, so the second sits quieter.

   The quiet tone has to be chassis-relative. This rule is (0,3,0) and used to
   set `--gs-ink-2`, which beat the (0,1,0) `.v2-why` when §143 moved the card
   onto the hardware chassis — leaving #5D5348 on #2A2620 at a measured 2.00:1,
   failing AA on all four film cards in light mode. Page-ground ink on a
   chassis is always wrong here; mix down from --v2-shell-ink instead. */
.v2-reel .v2-why + .v2-why { color: color-mix(in oklab, var(--v2-shell-ink) 72%, transparent); margin-top: 2px; }

/* The cassette label and slot carry `.mono` for the typeface, and `.mono`
   also sets a colour — it beat Lovable's `.v2-label` / `.v2-slot` at equal
   specificity and loaded later, so the standby text took `--gs-ink-3` instead
   of the chassis ink. Measured 2.63 against the dark chassis, under AA.
   Scoping to the shell restores the intended pair.

   The value changed in §143 and the reason is worth keeping. This rule is
   (0,2,0) and pinned `--gs-ink`, which beat the (0,1,0) `.v2-label` when the
   cassette window moved onto `--v2-screen`. `--gs-ink` flips with the theme,
   so dark mode stayed fine (12.94) while LIGHT mode put #1A1D2B ink on the
   dark well at a measured 1.27:1. `--v2-screen-ink` does not flip — the
   screen is dark in both themes — so it is the correct pair for anything
   sitting inside the window. */
.v2-shell .v2-label { color: var(--v2-screen-ink); }
/* Passing today, because --v2-shell-ink is light in both themes since §143.
   Moved to --v2-screen-ink anyway: the slot sits INSIDE the window, so it
   belongs to the screen pair, and this would break silently the next time
   the chassis ink changes. Not a bug fix — a consistency one. */
.v2-shell .v2-slot  { color: var(--v2-screen-ink); }

/* `.v2-frame` is a cinema screen: `--v2-screen` is dark in BOTH themes, but
   `--v2-ochre` flips — #9A5B12 in light, #E2A65A in dark. So in light mode
   the frame put dark ochre on a dark screen and measured 3.16, under AA,
   while dark mode was fine at 6.91. Same shape as the dock-CTA bug: a token
   that flips with the theme, used on a surface that does not.

   The screen always needs screen-bright ink, so the frame gets an ochre that
   does not flip. 7.49 in both themes. */
:root, .gsh { --v2-screen-ochre: #E2A65A; }
/* (0,2,0) to match `.gsh .mono` in the locked sheet, which was winning at
   (0,2,0) against a bare `.v2-frame` at (0,1,0) and painting the frame
   `--gs-ink-3`. The ochre film-frame treatment was never rendering in EITHER
   theme — dark merely happened to pass at 6.91 with the wrong colour, which
   is why only the light-mode number looked wrong. Loading later was not
   enough; it never is against the locked file. */
.v2-reel .v2-frame { color: var(--v2-screen-ochre); }

/* Tier 03 / 04 broadcast cards. They reuse Lovable's shared `.v2-reel` base
   (flex column, raised, 16px radius) with a mono kicker and a tactile arrow
   pill, and carry locked baselines so four cards hold one line. */
.v2-bcast { gap: 6px; }
.v2-bcast-k { font-size: 11px; letter-spacing: .16em; text-transform: uppercase; }
.v2-reel.v2-bcast .v2-bcast-k { color: var(--gs-ink-2); }
.v2-bcast p { margin: 0; color: var(--gs-ink-2); line-height: 1.45; font-size: 15px; }
.v2-bcast h4 { margin: 4px 0 0; }
.v2-bcast-acts { display: grid; gap: 8px; margin-top: auto; padding-top: 10px; }
.v2-bcast-acts .v2-link { margin-top: 0; padding-top: 0; }
@media (min-width: 760px) {
  .v2-tape h4, .v2-reel h4, .v2-bcast h4 { min-height: 50px; }
  .v2-tape p, .v2-bcast p { min-height: 77px; }
  .v2-tier-films .v2-reel .v2-why:last-of-type { min-height: 77px; }
}
@media (max-width: 759px) {
  .v2-tape h4, .v2-reel h4, .v2-bcast h4,
  .v2-tape p, .v2-bcast p,
  .v2-tier-films .v2-reel .v2-why:last-of-type { min-height: auto; }
}

/* ---------------------------------------------------------------------------
   41. The white band behind the floating header (6 Nov)
   ---------------------------------------------------------------------------
   Lovable's diagnosis is exactly right. `gs-site-lab.css:2392` sets
   `.gsh:not([data-theme="dark"]) { --gs-ground: #fff; background: #fff }`,
   so the canvas is pure white in light mode while the first fold below is
   warm. The header is a floating pill on a transparent bar, so the canvas
   shows through as a stark band above it.

   Measured, light, 1280: at y=10 the painted colour is `.gsh` rgb(255,255,255)
   and at y=90 it is rgb(247,243,238) — a 74px white strip.

   Affected pages are exactly the four they named. `/` and `/tools` are not:
   the homepage hero extends up under the header, and `/tools` is a ported
   page whose body carries the warm ground already. Verified all six.

   ONE CHANGE TO THEIR CSS, and it matters. They also set
   `--gs-ground: #F7F3EE`. That token is used as a FOREGROUND colour in 17
   places — `color: var(--gs-ground)` on `.gs-dock-cta`, `.v2-pill`,
   `.mn-cta` and others — so retokenising it would turn every pill label from
   white to warm cream and drop their contrast. Only the canvas needs to
   change, so only the canvas is changed.

   `!important` IS required here, and we were wrong to drop it first time.
   The judgement was that matching the locked rule's (0,2,0) and loading later
   would be enough. It is not: `.gsh.ab` carries an INLINE
   `style="...background:var(--gs-ground)..."`, and an inline declaration beats
   every stylesheet rule that is not `!important`. Measured after the first
   attempt — still rgb(255,255,255) at y=10. Lovable's original CSS had this
   right; the usual "match specificity instead of shouting" rule does not
   apply against an inline style.

   Their `.hd` / `.ab-head` background rules are still not applied. Those
   elements are transparent by design, and once the canvas is warm they reveal
   the right colour on their own. Their rules also lean on `--fold-a`, which
   is not defined in this file. */
.gsh:not([data-theme="dark"]) { background: #F7F3EE !important; }

/* ---------------------------------------------------------------------------
   42. /about closing fold — headline balance and marquee breathing room
   ---------------------------------------------------------------------------
   Lovable raised three things here. Measured all three first; two were real.

   1. HEADLINE — real. "GrowthSanity is the thinking. Acquret is where we put
      it to work." ran to THREE lines (132px over a 43.2px line-height),
      stranding "to work." alone. The band was near-even at 521/497px.
      Widening the left column balances it onto two.

   2. STATUS STRIP — NOT a defect, and nothing is changed. They asked us to
      anchor `.lab-close-strip` as "the official base of the card enclosure
      with a matching hairline". It already is: measured as the LAST CHILD of
      `.lab-close-card`, painted on that card's own white background, with
      `border-top: 1px solid rgb(232,225,216)` already present, and its bottom
      edge flush with the card's at 629px. There is nothing to connect — it is
      already the card's base.

   3. MARQUEE — real, but the opposite way round. They reported "almost zero
      vertical margin" above it. Measured, there is **56px above** it and
      **0px below**, with the footer flush against it. So the crowding is
      underneath, not on top.

      Their `padding-top: clamp(48px, 6vh, 64px)` would stack on the existing
      56px and open a 104-120px hole, so only the bottom padding is applied.
      The top is left alone and the numbers are reported back for them to
      judge. Their `background: transparent !important` is also skipped — the
      strip is already transparent.

      `!important` is required and we were wrong to drop it first time: the
      locked sheet carries
      `.gsh .mq-shelf.mq-end { padding-block-start: 0 !important;
      padding-block-end: 0 !important }`. Without it the rule matched, sat
      last in the cascade with the right value, and still computed 0px — even
      an inline `padding-bottom` on the element computed 0px, which is what
      finally pointed at an `!important` upstream. Lovable's CSS had this
      right too.

   4. HEADLINE, second pass. Widening the column from 521 to 566px was not
      enough on its own — still three lines. `text-wrap: balance` lets the
      browser even the lines rather than filling greedily. */
/* `text-wrap: balance` was the wrong tool and is not used: it evens line
   LENGTHS, while the ask is one SENTENCE per line. Those fight — balanced
   wrapping would happily give "GrowthSanity is the thinking. Acquret is" on
   line one. The split is explicit in the markup instead, each sentence a
   block span, with the column widened so the longer one fits on a single
   line: measured 598px against a 566px column, a 32px shortfall, so 1.5fr
   gives 611px. Below that each sentence wraps within itself, which is the
   right behaviour on a phone. */
/* CORRECTED 6 Nov. This rule shipped unscoped and that was a live bug.

   The locked sheet collapses the band to one column on phones:

     @media (width <= 760px) {
       .gsh[data-lab] .lab-close-band { grid-template-columns: minmax(0px, 1fr); gap: 22px }
     }                                                              -- line 3881

   That is (0,3,0). The rule below is ALSO (0,3,0) and loads later, and a
   media query adds nothing to specificity — so this beat the mobile rule
   INSIDE its own media query and the band stayed two columns at every width.
   Measured on /about at 390px before the fix: columns 167px and 111px, the
   headline shattered across 106px line boxes, the paragraph squeezed into a
   111px ribbon. The comment below cheerfully claimed "below that each
   sentence wraps within itself, which is the right behaviour on a phone" —
   it never got the chance, because this rule removed the collapse.

   Seventh specificity loss recorded in this file, and the first that broke a
   layout rather than quietly doing nothing. The lesson is now narrower than
   "count the selectors": an override that exists to fix ONE width has to be
   scoped to that width, or it silently follows the element everywhere.

   Scoped to the desktop side of the locked breakpoint so the phone rule gets
   its column back. `align-items: start` is added in the same query — see §48.
SUPERSEDED 18 Nov. The wide left column existed to carry the 36px `is-lead`
   headline, and `is-lead` is now off the band on /about and /shelf, so the
   headline is the base 23px and the column no longer binds. Reverted to the
   base 1.05fr/1fr for parity with the homepage card. The specificity lesson
   above still stands and is why the rule was scoped to >=761px in the first
   place; it is removed, not re-scoped. */
.gsh.ab .ab-close-line { display: block; }
.gsh .mq-shelf.mq-end { padding-bottom: clamp(20px, 3vh, 32px) !important; }

/* ---------------------------------------------------------------------------
   43. Lovable's post-deploy batch — the `.mono` trap, one more time (6 Nov)
   ---------------------------------------------------------------------------
   1. `/games` "The call" body was unreadable in light. CONFIRMED at **3.16**
      (they reported 1.08; either way it is under AA). `.v2-call` carries
      `.mono`, `.gsh .mono` colours it `--gs-ink-3`, and `.v2-call p` never
      declared its own colour — so the paragraph INHERITED dark ink onto the
      #14161E screen.

      This is ours to own. It is the fourth instance of the same trap — the
      film frame, the cassette label, the standby slot, and now this — and
      after fixing the first three we did not sweep for the rest. A sweep of
      every `v2-` element against its painted background across all four rooms
      now shows this as the only remaining failure; `.v2-shell` also flagged
      at 1.11 but is a container with no direct text, all of it in children
      that measure 14.26 and 12.81.

      No `!important` needed here, unlike the last two: the colour arrives by
      INHERITANCE from the parent, and any direct rule on the child beats
      inheritance regardless of specificity. Verified rather than assumed.
      Their `.v2-call span` fix is already in §39 and measures 9.35.

   2. Dock spacing — the dock sat flush against the marquee, measured 0px.

   3. `/quick-q` desktop stagger — real, but NOT from the ping. All four pings
      measure 61px; their reported "#002 wraps to two lines" does not happen.
      The stagger is in the action row (790 vs 810) because card 1 is 429px
      tall against the row's 448px, and the cause is OUR OWN §30 rule
      `.gsh[data-lab] .qq-grid { align-items: start }`, still matching because
      the container kept its `qq-grid` class alongside `v2-qq-grid`.

      So the fix is one property, not three min-heights. Stretching the cards
      lets the existing `margin-top: auto` on `.v2-slip-act` do its job.
      Their envelopes would also work but bake 54/110/96px into the layout,
      which breaks the moment the copy changes — the same brittleness the hero
      props taught us. */
.v2-call p { color: var(--v2-screen-ink); }
.gs-dock { margin-bottom: clamp(48px, 6vh, 64px); }
/* (0,4,0) to beat §30's `.gsh[data-lab] .qq-grid { align-items: start }` at
   (0,3,0). A bare `.v2-qq-grid` at (0,1,0) lost silently — the cards stayed
   429px against a 448px row and the action rows kept their 20px stagger.
   Same trap as every other time: loading later is not enough. */
.gsh[data-lab] .qq-grid.v2-qq-grid { align-items: stretch; }

/* ---------------------------------------------------------------------------
   45. /playlists Tiers 03 + 04 — broadcast tuner and studio monitor
   ---------------------------------------------------------------------------
   Lovable, 6 Nov: Tiers 03 and 04 "still read as plain cards" next to the
   cassette decks in Tier 01 and the film reels in Tier 02, and should become
   a broadcast/hi-fi deck and a CRT/studio-monitor chassis respectively.

   Both reuse the `--v2-shell` / `--v2-screen` token pair the decks and reels
   already use, so this is the same instrument family rather than two new
   looks. Worth restating how that pair behaves, because it is unusual and
   easy to get wrong: `--v2-shell` INVERTS between themes (dark chassis on the
   light site, light chassis on the dark one) while `--v2-screen` stays dark in
   both.

   SUPERSEDED 18 Nov. `--v2-shell` no longer inverts. QA on the full-bleed
   build showed the cream chassis the dark theme used to swap in read as
   glowing white sheets once Tiers 01 and 02 joined the quartet, so the
   chassis is now dark in BOTH themes (#2A2620 light, #161822 dark) and
   `--v2-shell-ink` is #F2ECE4 in both. `--v2-screen` still stays dark in
   both, which was always correct — a powered screen is dark whatever the
   room is doing. Anything sitting on the chassis still takes its colour from
   `--v2-shell-ink` and anything on the screen from `--v2-screen-ink`; the two
   are not interchangeable and a literal hex is wrong in both places.

   Because the chassis no longer inverts, `--v2-dial` no longer needs two
   values. #E2A65A measures 7.5:1 on #2A2620 and holds on #161822, so one
   value now serves both themes. The old dark value (#8A4F0C) existed only to
   survive the cream chassis that no longer exists.

   Note both tiers' cards carry `.mono` on `.v2-bcast-k` and `.v2-link`, which
   `.gsh .mono` at (0,2,0) colours `--gs-ink-3` — the trap that has now bitten
   four times. Every one of them is re-coloured explicitly below.
   --------------------------------------------------------------------------- */

:root, .gsh, html[data-theme="dark"], .gsh[data-theme="dark"] { --v2-dial: #E2A65A; }

/* --- the band stamp under each tier title --- */
/* `display: flex` + `width: fit-content` rather than `inline-flex`: the
   locked `.gsh .dt-grp-h` is `display: block`, so an inline badge would trail
   the subtitle text on whatever line it ended on. This gives it its own row
   while still shrinking to the label. */
.gsh .v2-band {
  display: flex; width: fit-content; align-items: center; gap: .5ch;
  margin-top: 12px; padding: 5px 11px;
  border: 1px solid color-mix(in oklab, var(--v2-dial) 55%, transparent);
  border-radius: 4px;
  font: 500 11px/1 var(--gs-mono); letter-spacing: .18em; text-transform: uppercase;
  color: var(--gs-ink);
  background: color-mix(in oklab, var(--v2-dial) 12%, var(--gs-ground));
}
.gsh .v2-band i { font-style: normal; opacity: .55; }

/* --- Tier 03: the tuner chassis ---
   Chassis body, with the dial scale drawn across the head as a repeating
   gradient and a single amber marker sitting on it. Ticks and marker are
   decoration, so they carry no contrast obligation; all text is shell ink. */
.gsh .v2-tier-audio .v2-bcast {
  position: relative;
  padding: 18px 16px 16px;
  background: var(--v2-shell);
  border: 1px solid var(--v2-border);
  border-radius: 14px;
  overflow: hidden;
}
.gsh .v2-tier-audio .v2-bcast::before {
  content: ""; position: absolute; inset: 0 0 auto 0; height: 14px;
  background:
    repeating-linear-gradient(90deg,
      color-mix(in oklab, var(--v2-shell-ink) 38%, transparent) 0 1px,
      transparent 1px 9px),
    color-mix(in oklab, var(--v2-shell-ink) 7%, transparent);
  border-bottom: 1px solid color-mix(in oklab, var(--v2-shell-ink) 16%, transparent);
}
.gsh .v2-tier-audio .v2-bcast::after {
  content: ""; position: absolute; top: 0; left: 50%; width: 2px; height: 14px;
  background: var(--v2-dial);
  transform: translateX(-85px);
  animation: v2seek 9s ease-in-out infinite;
}
.gsh .v2-tier-audio .v2-bcast > * { position: relative; }
.gsh .v2-tier-audio .v2-bcast .v2-bcast-k { margin-top: 10px; color: var(--v2-dial); }
.gsh .v2-tier-audio .v2-bcast h4 { color: var(--v2-shell-ink); }
.gsh .v2-tier-audio .v2-bcast p  { color: color-mix(in oklab, var(--v2-shell-ink) 74%, transparent); }
.gsh .v2-tier-audio .v2-bcast .v2-link { color: var(--v2-shell-ink); }
.gsh .v2-tier-audio .v2-bcast.is-static {
  background: color-mix(in oklab, var(--v2-shell) 92%, var(--gs-ground));
  border-style: dashed;
}

/* --- Tier 04: the studio monitor ---
   Inverted from Tier 03: here the CARD is the screen and the chassis is the
   bezel around it, which is what separates a monitor from a tuner at a
   glance. Scanlines overlay the whole face; `pointer-events: none` so they
   never intercept the card's own link. The screen is dark in both themes, so
   unlike Tier 03 nothing here needs a per-theme value. */
.gsh .v2-tier-watch .v2-bcast {
  position: relative;
  padding: 16px;
  background: var(--v2-screen);
  border: 6px solid var(--v2-bezel);
  border-radius: 14px;
  overflow: hidden;
  padding: 18px 16px 16px;
}
.gsh .v2-tier-watch .v2-bcast::after {
  content: ""; position: absolute; inset: 0; pointer-events: none;
  background: repeating-linear-gradient(180deg,
    color-mix(in oklab, var(--v2-screen-ink) 7%, transparent) 0 1px,
    transparent 1px 4px);
}
.gsh .v2-tier-watch .v2-bcast > * { position: relative; z-index: 1; }
.gsh .v2-tier-watch .v2-bcast .v2-bcast-k {
  display: inline-flex; align-items: center; gap: .6ch;
  color: var(--v2-glow);
}
/* The tally light. Decorative, hidden from the tree. */
.gsh .v2-tier-watch .v2-bcast .v2-bcast-k::before {
  content: ""; width: 6px; height: 6px; border-radius: 50%;
  background: var(--v2-glow);
  box-shadow: 0 0 6px var(--v2-glow);
}
.gsh .v2-tier-watch .v2-bcast h4 { color: var(--v2-screen-ink); }
.gsh .v2-tier-watch .v2-bcast p  { color: color-mix(in oklab, var(--v2-screen-ink) 78%, transparent); }
.gsh .v2-tier-watch .v2-bcast .v2-link { color: var(--v2-screen-ink); }
.gsh .v2-tier-watch .v2-bcast.is-static .v2-bcast-k::before {
  background: color-mix(in oklab, var(--v2-screen-ink) 35%, transparent);
  box-shadow: none;
}

/* Tier 04 now carries four cards — Acquret, two reserved slots and the
   nomination card — so the locked `.gsh[data-lab] .pl-grid` auto-fit lays
   four even tracks on its own and needs NO override. The earlier two- and
   three-column pins were deleted rather than edited; when the real channels
   land they simply replace the slots and the grid is already right.

   The slots are placeholders and must read as such. They are not links, they
   carry no channel name, and the bezel goes dashed with the tally light dead
   — the screen-off state of the same chassis, so an empty bay still belongs
   to the rack instead of looking like a broken card. */
.gsh .v2-tier-watch .v2-bcast.is-slot {
  border-style: dashed;
  border-color: color-mix(in oklab, var(--v2-bezel) 78%, var(--v2-screen));
}
/* Standby slots read as a ghost line, not a coupon cutout. At 20% the dashed
   rim sits just above the screen without competing with the live card. */
html[data-theme="dark"] .gsh .v2-tier-watch .v2-bcast.is-slot,
.gsh[data-theme="dark"] .v2-tier-watch .v2-bcast.is-slot {
  border-color: color-mix(in oklab, var(--v2-screen-ink) 20%, transparent);
}
.gsh .v2-tier-watch .v2-bcast.is-slot h4 {
  color: color-mix(in oklab, var(--v2-screen-ink) 62%, transparent);
  font-style: italic; font-weight: 400;
}
.gsh .v2-tier-watch .v2-bcast.is-slot .v2-bcast-k { color: color-mix(in oklab, var(--v2-screen-ink) 60%, transparent); }
.gsh .v2-tier-watch .v2-bcast.is-slot .v2-bcast-k::before {
  background: color-mix(in oklab, var(--v2-screen-ink) 30%, transparent);
  box-shadow: none;
}
.gsh .v2-tier-watch .v2-link-off {
  color: color-mix(in oklab, var(--v2-screen-ink) 52%, transparent);
  letter-spacing: .16em; text-transform: uppercase; font-size: 11px;
}

/* ---------------------------------------------------------------------------
   46. Mobile drawer runs the full width of the header bar
   ---------------------------------------------------------------------------
   Lovable, 6 Nov: make the mobile nav full-width below 720px, via
   `left: 16px; right: 16px`.

   That exact declaration cannot work here, and it is worth writing down why
   so it is not reached for again. `.lab-mnav-panel` is `position: absolute`
   and its offset parent is `details.lab-mnav` — the hamburger itself, which
   measures 36px wide (confirmed at 390px: offsetParent `.lab-mnav`, left 325,
   width 36). `left` and `right` resolve against THAT box, not the viewport,
   so the spec as written would render the menu as a 4px sliver from x=341 to
   x=345. The panel reaches full width by being sized, not by being inset.

   Measured before changing anything, the panel was already bar-flush on
   phones — 390px gave panel 20→370 against a header bar of 20→370, from §35.
   What was NOT full width was everything above ~460px, where the 420px cap
   from §35 starts binding: at 600px the panel ran 160→580 inside a 20→580
   bar, and at 720px 280→700 inside 20→700. So the cap is the whole of what
   Lovable is seeing, and dropping it is the entire fix.

   The 20px gutter is kept rather than moved to their 16px: `calc(100vw-40px)`
   is what aligns the panel's edges to the header bar, and §35 existed
   specifically to fix that alignment. 16px would push it 4px proud of the bar
   on both sides and re-introduce the thing §35 removed. */
@media (360px <= width <= 720px) {
  .gsh .lab-mnav-panel { max-width: none; }
}

/* With the disclosure gone the directory is always open, so the body no
   longer needs to read as a nested panel inside the drawer — it is the
   drawer. The label keeps `.mn-shelf`, so the locked flex/`b`/`small` rules
   at gs-site-lab.css:3494-3498 still dress it; only the chevron is dropped. */
.mn-shelf-open { display: block; width: 100%; }
.mn-shelf-lbl { cursor: default; }

/* Flattening the disclosure costs height: the directory that used to be
   behind a 36px summary row is now always open, and at 390x844 — one of the
   TALLER phones — that pushed "Get in touch" 43px past the drawer's scroll
   floor. Reachable by scrolling, but the primary CTA should not need one.

   The height comes back from the body's own chrome rather than from the
   content. As a nested panel it earned a border, a tint, a radius and its own
   padding to separate it from the drawer; as the drawer's main content it is
   just nesting a box inside an identical box. Dropping it reclaims ~26px and
   is what "open layout" should look like anyway. Group padding takes the
   remainder. */
.mn-shelf-body {
  margin: 2px 0 4px; padding: 0;
  border: 0; border-radius: 0; background: none;
}
/* These three all need to clear LOCKED rules, not just load later:
     .gsh .shf-m .shf-g            (0,3,0)  padding: 10px 6px   line 3225
     .gsh .lab-mnav-panel .mn-a    (0,3,0)  padding: 13px 14px  line 3227
   The first pass wrote them at (0,2,0) and (0,1,0) and they were dropped
   silently — content height did not move by a single pixel. Sixth time this
   file has lost that way; the computed value is checked below, not assumed. */
.gsh .shf-m.mn-shelf-body .shf-g { padding: 12px 8px 5px; }  /* top raised in §53 */
.gsh .shf-m.mn-shelf-body .shf-allw { margin-top: 2px; }
/* Retuned 6 Nov: the label lost its subtitle ("Curated tools, playlists &
   spaces"), which the open directory below already says. The 2px bottom was
   sized for a two-line stack; with one line it put the title 2px off the
   directory block. Removing the subtitle freed 19px of drawer height and the
   panel sits ~56px under its max-height, so the room is spent on the title
   rather than banked. */
.gsh .lab-mnav-panel .mn-shelf-lbl { padding-block: 12px 14px; }

/* Widening the drawer has a consequence Lovable's note did not cover: the
   directory is pinned to one column by `.gsh .shf-m .shf-cols` (0,3,0), so at
   720px each row became a label at far left and an arrow at far right with
   ~600px of nothing between them. Full width is only an improvement if the
   width gets used, so from 560px the directory returns to the two columns the
   desktop dropdown already uses. Below 560 it stays single-column.

   Flagged to Lovable as a judgement call rather than a spec item — trivially
   reverted by deleting this block if they want the single column at tablet.
   (0,4,0) against the locked (0,3,0); `.shf-soonrow` has its own 1fr rule at
   gs-site-lab.css:3614 and needs covering separately. */
@media (560px <= width <= 720px) {
  .gsh .shf-m.mn-shelf-body .shf-cols,
  .gsh .shf-m.mn-shelf-body .shf-soonrow { grid-template-columns: 1fr 1fr; }
  .gsh .shf-m.mn-shelf-body .shf-cols .shf-g + .shf-g,
  .gsh .shf-m.mn-shelf-body .shf-soonrow .shf-g + .shf-g {
    border-top: 0; border-left: 1px solid var(--gs-rule);
  }
}

/* ---------------------------------------------------------------------------
   47. Drawer: "Explore the full shelf page" steps down to secondary
   ---------------------------------------------------------------------------
   Kyle, 6 Nov, from a phone: the drawer shows two filled gradient pills —
   this one and "Get in touch" — and they clash. He is right, and it is a
   hierarchy problem rather than a colour one. A drawer should have exactly
   one primary action; two identical pills make the reader choose between
   them, and the one that should win (the CTA) is the lower of the two.

   The gradient comes from the LOCKED `.gsh .shf-all` (gs-site-lab.css:3610),
   which is shared with the desktop `.shf-dd` dropdown. Scoped here to
   `.mn-shelf-body` so the desktop dropdown is untouched — only the burger
   menu changes.

   Secondary, not invisible: it keeps the pill shape and padding so it still
   reads as a tappable destination next to the arrowed directory rows, and
   takes a hairline border plus the green arrow the rest of the shelf uses.
   A plain text link was the alternative and was rejected — it would have
   been the only unbordered interactive element in the drawer.

   This also makes §24's white-ink fix inert for this element. That rule
   existed because the gradient needed light text; with no gradient the text
   is ink, set explicitly here rather than left to fall back. The §24 comment
   is kept — the undefined-token trap it documents is still worth having.

   Specificity: locked base is (0,2,0) and its `:hover` / `:focus-visible`
   (0,3,0). Base here is (0,4,0) so it clears both; the hover below is (0,5,0)
   so it clears this base in turn. Computed values read back, not assumed. */
.gsh .shf-m.mn-shelf-body .shf-all {
  background: transparent;
  border: 1px solid color-mix(in oklab, var(--gs-ink) 22%, transparent);
  border-radius: 999px;
  color: var(--gs-ink);
  box-shadow: none;
  transform: none;
}
.gsh .shf-m.mn-shelf-body .shf-all .gs-arr { color: var(--gs-green-active, #1B6B38); }
.gsh[data-theme="dark"] .shf-m.mn-shelf-body .shf-all .gs-arr { color: var(--gs-green, #35BD6B); }
.gsh .shf-m.mn-shelf-body .shf-all:hover,
.gsh .shf-m.mn-shelf-body .shf-all:focus-visible {
  background: color-mix(in oklab, var(--gs-ink) 6%, var(--gs-ground));
  border-color: color-mix(in oklab, var(--gs-ink) 38%, transparent);
  box-shadow: none;
  transform: none;
}

/* ---------------------------------------------------------------------------
   48. /about closing fold — the two columns share a top edge
   ---------------------------------------------------------------------------
   Kyle, 6 Nov: the closing band reads messy. It does, and §42 did not touch
   the cause — it fixed the column RATIO and the sentence splits, which is why
   the fold still looked wrong after that work landed.

   The cause is `align-items: center` on the locked band (gs-site-lab.css:3842).
   The two columns hold different amounts — left is kicker + headline, right is
   paragraph + CTA — so centring them means NOTHING lines up at either edge.
   Measured at 1280 before this change:

     left  (kicker + headline)   top 2892   bottom 3007   height 115
     right (paragraph + CTA)     top 2857   bottom 3043   height 186

   The right block started 35px ABOVE the left kicker, floating against
   nothing. `align-items: start` puts the kicker and the paragraph's first
   line on one baseline — measured top delta 0 at 1024, 1280 and 1440.

   `end` was tested too and is worse (delta -71): it hangs the headline off
   the CTA's bottom edge, which is not a line anyone reads to.

   Scoped to the same >=761px query as §42, so the phone layout — now one
   column again — is untouched, where `align-items` has nothing to align.

   The 40px gutter is deliberately LEFT ALONE. Widening it reads better at
   1280 but narrows the left column, and the column width is already the
   binding constraint on the headline between ~1025 and ~1190 (see below).
   56px was the most it could take before the headline broke at 1280; 64px
   broke it outright. Not worth spending the headline's margin on air.
SUPERSEDED 18 Nov, same cause. `align-items: start` was fixing a 35px gap
   that only appeared because the 36px headline made the left column far
   taller than the right. With the headline at 23px the two columns are close
   in height and the base `center` is correct again — re-measured below. */

/* ---------------------------------------------------------------------------
   49. /quick-q cards become a swipe carousel on phones
   ---------------------------------------------------------------------------
   Kyle: on a phone the Quick Q cards are a long vertical list while /tools,
   /games and /playlists all swipe. He is right that it is the odd one out —
   these cards are the tallest on the site (ping + subtext + reply + actions),
   so a stack of four is a very long scroll to reach the fourth.

   Reuses `.v2-snap`, the carousel helper the other V2 pages already use,
   declared on the grid in the markup so the intent is visible there rather
   than hidden in a selector. But `.v2-snap` alone could never have worked:
   it is (0,1,0) against `.gsh[data-lab] .qq-grid` (0,3,0) and the
   `.v2-qq-grid` single-column media rule, both of which set the display and
   the columns. Hence the (0,4,0) rule below — the class declares the intent,
   this gives it the weight to land. Same lesson as §43 and §46; the class
   being present in the HTML is not the same as the rule applying.

   Breakpoint is 768px, which is neither the 759px the other V2 carousels use
   nor the 760px of `.v2-qq-grid`. THREE rules drop this grid toward one
   column and they disagree:

     .v2-qq-grid            (0,1,0)  1fr at <= 760   line 2308
     .gsh[data-lab] .qq-grid (0,3,0)  1fr at <= 768   line 378
     other V2 carousels              -       <= 759   -

   The (0,3,0) rule wins, so the real single-column region is **<= 768**, and
   measuring 761px confirmed it: one 721px track, four rows — a stack of tall
   cards, exactly the thing being fixed, just at tablet width. Scoping the
   carousel to 759 or 760 would have left 761-768 as that stack and looked
   like the fix simply did not work on a tablet.

   So the breakpoint tracks the rule that actually governs, not the nearest
   convention. If line 378 ever moves, this moves with it.

   `grid-template-columns: none` is explicit because the element keeps its
   grid classes; switching `display` to flex leaves the column track
   definition inert but present, and leaving it set invites confusion later. */
/* Breakpoint lowered 768 -> 759 on 12 Nov (Sprint 2), so all four hubs flip
   together at the locked /tools boundary. Between 760 and 768 the locked
   `.gsh[data-lab] .qq-grid` 1fr rule now takes this grid, which is the same
   stacked layout the other hubs show in that band. See §59. */
@media (width <= 759px) {
  .gsh[data-lab] .qq-grid.v2-snap {
    display: flex;
    grid-template-columns: none;
    gap: 12px;
    overflow-x: auto;
    scroll-snap-type: x mandatory;
    overscroll-behavior-x: contain;
    scrollbar-width: none;
    margin-inline: -20px;
    padding: 4px 20px 12px;
    scroll-padding: 0 20px;
  }
  .gsh[data-lab] .qq-grid.v2-snap::-webkit-scrollbar { width: 0; height: 0; display: none; }
  .gsh[data-lab] .qq-grid.v2-snap > .v2-slip {
    flex: 0 0 82%;
    scroll-snap-align: start;
  }
}

/* The persona filter hides cards with the `hidden` attribute, and `.v2-slip`
   carries an author `display: flex`. Author rules beat the UA sheet's
   `[hidden] { display: none }` on origin, not specificity, so whether the
   filter works at all rests on a UA detail — it does today in Chrome, which
   is why this was never visible. Stating it explicitly removes the question,
   and matters more now: in a flex row a card that failed to hide would not
   just appear, it would occupy a scroll slot. */
.gsh .v2-slip[hidden] { display: none; }

/* ---------------------------------------------------------------------------
   50. Mobile drawer clears the header bar
   ---------------------------------------------------------------------------
   Kyle, from an iPhone 17 Pro Max: the header and the drawer nearly touch.
   Measured at 430x932 — they do not nearly touch, the gap is exactly **0px**.

   The locked `top: calc(100% + 10px)` (gs-site-lab.css:1922) is measured from
   the panel's offset parent, which is the hamburger BUTTON, not the header
   bar. At 430px the hamburger ends at y=58 and the bar at y=68, so the bar's
   own 10px of padding below the button consumes the entire offset and the
   panel lands flush against the bar's edge.

   Measured options: 18px -> 8px gap, 20px -> 10px, 24px -> 14px. Taking 20px
   for a 10px visual gap, which is the separation the original 10px was
   plainly reaching for before the offset parent ate it.

   `max-height` drops by the same 10px so the drawer keeps the bottom room it
   had — pushing the top down without this would spend the gap twice, once at
   the top and again off the bottom of the screen. */
@media (width <= 720px) {
  .gsh .lab-mnav-panel {
    top: calc(100% + 20px);
    max-height: calc(100vh - 94px);
    max-height: calc(100svh - 94px);
  }
}

/* ---------------------------------------------------------------------------
   51. /about "Built by • marketers" — baseline and gradient
   ---------------------------------------------------------------------------
   Kyle: the pill's word and dot sit lower than "Built by", on every device.
   Measured by range rather than by eye: identical font size and identical
   67px text-rect heights, with "marketers" **6.5px lower** at 56px type. So
   it is purely a baseline offset, not a size or font difference.

   The cause is the locked `vertical-align: middle` on an inline-flex box that
   is taller than its text (padding plus line-height 1.1). `middle` aligns the
   BOX's centre to the parent's x-height midpoint, so any box taller than the
   text rides low — and the taller the display type, the worse it reads.

   `baseline` was tried first and is WORSE (12.6px): for an inline-flex the
   baseline comes from the first flex item, and the first item here is the dot
   `::before`, a textless box whose baseline is its bottom edge. Negative
   values were tried too and move it further down — the sign is inverted from
   what it looks like.

   0.224em zeroes it, and it is in `em` because the offset scales with the
   type. Verified at two sizes rather than assumed:

     1280px / 56.32px type:  6.46px -> -0.02px
      430px / 30px    type:  3.74px ->  0.28px

   --- the gradient -------------------------------------------------------

   The text takes the CTA's own gradient token. It must sit on an inner span:
   `background-clip: text` clips EVERY background on the element, so putting
   the gradient on `.ab-pill` itself would clip the pill's own raised fill to
   the glyphs and destroy the pill shape.

   The dot is `background: currentcolor`, and gradient text needs
   `color: transparent` — which would have made the dot vanish. It gets the
   gradient explicitly instead, so the mark and the word read as one object.

   Guarded by `@supports`: without `background-clip: text`, `color:
   transparent` is invisible text rather than a degraded colour. Outside the
   query the pill keeps its locked green, so the fallback is the design as it
   shipped. */
.gsh .ab-pill { vertical-align: 0.224em; }

@supports ((-webkit-background-clip: text) or (background-clip: text)) {
  .gsh .ab-pill-t {
    background-image: var(--gs-cta-gradient);
    -webkit-background-clip: text;
    background-clip: text;
    color: transparent;
  }
  .gsh .ab-pill::before {
    background: var(--gs-cta-gradient);
  }
}

/* ---------------------------------------------------------------------------
   52. Desktop shelf dropdown — clears the header, and the bar loses its tint
   ---------------------------------------------------------------------------
   Kyle: the desktop dropdown has the same problem as the mobile drawer, and
   the directory bar's background should be white like the rest of the panel.

   Both correct. The dropdown is the third instance of the same offset-parent
   trap: `top: calc(100% + 22px)` (gs-site-lab.css:3195) is measured from
   `details.shf-dd` — the summary — not from the header bar, so the bar's own
   padding below the summary absorbs the whole 22px and the panel lands flush
   at **0px**. (A first reading showed -6px, which was the `shfIn` animation
   caught mid-flight; measured against an explicit value it is 0.)

   32px gives a 10px gap — deliberately the same 10px §50 gave the mobile
   drawer, so the two menus sit the same distance off the same bar instead of
   each being tuned on its own.

   The bar goes `transparent` rather than being repainted white. The panel is
   `#FFFFFF` in light and `rgb(20,23,33)` in dark, so transparent means "the
   panel's colour, whatever it is" and the request holds in both themes
   without a second hardcoded value to keep in sync. The `border-bottom` stays,
   so the bar is still separated from the directory — it just stops being a
   different surface. The green glyph glow is untouched.

   (0,3,0) on both: the top rule is (0,3,0) and loads earlier; the bar's dark
   background rule `.gsh[data-theme="dark"] .shf-bar` is also (0,3,0), so
   `.gsh .shf .shf-bar` ties it and wins on order. A plain `.gsh .shf-bar`
   (0,2,0) would have fixed light and silently left dark tinted. */
.gsh .shf-dd .shf { top: calc(100% + 32px); }
.gsh .shf .shf-bar { background: transparent; }

/* --- §51 addendum: the dot centres on the word, not on the line box -------
   Kyle, after §51 shipped: the dot still is not centred against "marketers".

   He is right, and it is a different fault from the baseline one. Measured at
   1280 with real glyph metrics:

     baseline              325.38
     dot centre            304.36
     x-height centre       310.82   <- where a bullet before lowercase belongs
     cap-height centre     305.28

   The dot is **6.47px above the x-height centre** and within 1px of the CAP
   centre. That is exactly what `align-items: center` produces: it centres the
   dot on the flex line box, which spans the full em — ascender and descender
   space included — rather than on the letterforms beside it.

   For a word with capitals that reads correctly. "marketers" is all
   lowercase, so its visual mass is the x-height band and the dot floats high
   above it. The reference has to be the x-height, not the line box.

   0.115em brings it down onto that centre, in `em` so it tracks the type.
   Verified at two sizes rather than assumed:

     56.32px type:  6.47px off -> -0.01px
     30px    type:            -> -0.21px

   For the record the BASELINE is not the problem and was not touched again:
   "marketers" to "Built by" measures -0.02px at 1280 and 0.28px at 430, i.e.
   0.009em — below the threshold of anything visible. §51's vertical-align
   already settled that; this is the remaining, and genuinely visible, half. */
.gsh .ab-pill::before { position: relative; top: 0.115em; }

/* --- §51 addendum 2: the PILL BOX, which the first two passes never checked -
   Kyle, third time on this element: the pill itself is misaligned. He is
   right, and the reason I missed it twice is that I kept measuring the things
   I had already decided to look at — the text baseline, then the dot — and
   never the box around them.

   Measured against "Built by" at 45px type:

     air above the "B" cap       12.2px
     air below the "y" descender  2.6px
     pill centre vs cap band     -0.76   <- looked fine, which is why it passed
     pill centre vs FULL band    -4.83   <- the actual fault

   So the pill is correctly centred on the cap band and still rides high over
   the text as a whole, because "Built by" has a descender and the cap band
   ignores it. Nearly four times as much air above the capital as below the
   tail reads as a box floating up and to the right of its line.

   Centring on the full band — cap top to descender bottom — gives 7.4px on
   both sides.

   The box has to move WITHOUT the contents moving, since the baseline and the
   dot are both already correct. A transform does that: it shifts the box
   visually without touching layout or the baseline computation, and the two
   children take an equal counter-shift so they stay exactly where they are.
   One constant, applied three times, rather than retuning `vertical-align`
   and then compensating it twice with different numbers.

   `em` so it tracks the type. Verified the baseline does not move: -0.12px.

   The lesson, and it is the third variant of the same one on this element:
   "aligned" has to name its reference. Baseline, x-height centre and full-band
   centre are three different answers, and this element needed all three to be
   right at once. */
.gsh .ab-pill { transform: translateY(0.107em); }
.gsh .ab-pill .ab-pill-t { transform: translateY(-0.107em); }
.gsh .ab-pill::before { transform: translateY(-0.107em); }

/* ---------------------------------------------------------------------------
   53. Drawer directory — redistribute the space it already has
   ---------------------------------------------------------------------------
   Kyle, on a phone: the group headers sit too close to the rule above them and
   look untidy, the section has space that is not being used properly, and the
   item titles must not crowd their own descriptions.

   Measured inside one group before this change:

     divider rule -> "01 THE OPERATOR KIT"       5px   <- too tight
     header -> first item                        8px
     item padding-top                           10px
     "Tools" -> "Calculators & funnel audits"    2px   <- too tight
     one item -> the next                       26px   <- the wasted space

   So the complaint is exactly right, and it is a distribution problem rather
   than a shortage: 26px sits between two items while a title and its own
   description are 2px apart. The eye groups by proximity, so at 2px the
   description reads as part of the title, and at 26px the items read as
   further apart than they are.

   Both tight values are OURS. The 5px is §46, where the group padding was cut
   from 10px to reclaim drawer height so the CTA would fit; §85 later removed
   the label's subtitle and gave that height back, but the padding was never
   restored. The 2px is §24's `row-gap`.

   Three changes, netting +22px:

     group padding-top   5 -> 12px   (+7 x 4 groups  = +28)
     row-gap             2 -> 5px    (+3 x 6 items   = +18)
     item padding-block  9 -> 7px    (-4 x 6 items   = -24)

   The third is what makes the first two affordable without the drawer
   scrolling. It is not a squeeze: the item box keeps roughly its height while
   its contents breathe, and the space between one item and the next drops
   from 26px to 22px — still comfortably more than the 5px now separating a
   title from its description, so the grouping reads correctly.

   Budget checked at the tightest common phone, 390x844: cap 750, content was
   718, so 32px spare. +22 leaves the CTA reachable without a scroll, which is
   what §85 established and should not be given back.

   (0,4,0) on the item padding: the locked `.gsh .lab-mnav-panel .shf-i`
   is (0,3,0), so a two-class selector would have lost silently. */
.gsh .shf-m.mn-shelf-body .shf-i {
  padding-block: 7px;
  row-gap: 5px;
}

/* ---------------------------------------------------------------------------
   54. Drawer: About first, and the CTA pinned so it can never read as cut
   ---------------------------------------------------------------------------
   Kyle, two observations on a phone.

   1. "Get in touch" looks cut at the bottom, and the scroll that would reveal
      it is not something everyone notices.
   2. About gets lost under ~500px of shelf directory.

   **The CTA.** Nothing was actually clipped: measured, it has 9px between it
   and the panel's bottom edge at every size — inside a panel with a **20px
   corner radius**. A square-shouldered button sitting 9px inside a 20px curve
   reads as cut, because its corners are visually inside the radius. At
   375x667 the drawer also overflows by 169px, so the bottom genuinely is a
   scroll boundary, and a flush edge there looks terminal rather than
   continuing.

   Padding alone would only soften the symptom. `position: sticky` answers
   what was actually said: the button stays on screen while the directory
   scrolls behind it, so it is always visible, never reads as cut, and the
   user never has to discover the scroll to reach the primary action.

   A side effect worth stating because it relaxes an earlier constraint: §85
   and §92 both had to fit the whole drawer above its cap so the CTA would not
   need a scroll. **That constraint is now gone** — the CTA is reachable at any
   drawer height — which is why the panel can afford 14px of bottom padding
   here rather than the 8px it was rationed to.

   The upward shadow is not decoration: it is the cue that content passes
   behind the button. Without it a pinned bar on a matching ground reads as
   the end of the list.

   **About first.** Moved in the MARKUP on all 15 pages, not with CSS `order`.
   `order` moves only what is seen; keyboard focus and screen readers would
   still have run shelf -> About, which is exactly the kind of split that
   makes a menu unusable without sight. The drawer is mobile-only — the
   hamburger is hidden from 768 up and the desktop nav is separate markup — so
   desktop ordering is untouched. */
@media (width <= 720px) {
  .gsh .lab-mnav-panel { padding-bottom: 14px; }
  .gsh .lab-mnav-panel .mn-cta {
    position: sticky;
    bottom: 0;
    z-index: 2;
    box-shadow: 0 -10px 16px -10px color-mix(in oklab, var(--gs-ink) 26%, transparent);
  }
}

/* ---------------------------------------------------------------------------
   55. Clock offset chip — restore the font that was lost in transit
   ---------------------------------------------------------------------------
   Kyle: the BASE chip looks oversized and heavy next to Lovable's.

   Not a design difference. Lovable's source says:

     .lab-ck-off { flex:none; font:500 10px/1 var(--gs-mono); ... }

   What reached `gs-site-lab.css` (line 2477) is that rule with the `font:`
   shorthand serialised into EMPTY longhands — `font-weight: ;` `font-size: ;`
   `font-family: ;` and so on, names present, values gone. A blank declaration
   is dropped, so the chip inherits the page default and renders **600 / 16px
   / Helvetica Neue** instead of **500 / 10px / JetBrains Mono**.

   Proof it is a genuine fallback and not equivalent notation: `--gs-ui` is
   Space Grotesk, it is loaded, and a rule whose shorthand survived
   (`.shf-i small`) renders Space Grotesk correctly. Only the blanked rules
   come out Helvetica.

   Restored as LONGHANDS, not the shorthand, and deliberately without
   `font-weight`. Line 2515 sets `font-weight: 600` on purpose — Lovable sets
   600 there too — and a `font:` shorthand here would reset it to 500 and
   silently undo their own later rule. Only the three properties that are
   actually missing are written back.

   Mobile is untouched: `.gsh .lab-ck-off.is-base { font-size: 8.5px }` is
   (0,3,0) inside a media query and still beats this (0,2,0), so the phone
   keeps the smaller base chip Lovable specified.

   NOT changed here, on Kyle's instruction to stay inside what he raised:
   the same blanking affects `.tl-stats b`, `.dk-idx strong` and
   `.lab-sh-fmt`. Listed for a separate decision rather than widened into
   this fix. `.clock .tm` was checked and is NOT affected — Lovable's own
   later rules set its family, size and weight, so the blank base rule never
   mattered. `.shf-i b` is likewise already correct. */
.gsh .lab-ck-off {
  font-family: var(--gs-mono);
  font-size: 10px;
  line-height: 1;
}

/* ---------------------------------------------------------------------------
   56. Phones: the offset chip moves onto the time row
   ---------------------------------------------------------------------------
   Kyle's call, and a deliberate change to Lovable's layout rather than a
   restoration — flagged to them as a specific case, not a blanket override.

   Lovable hide every offset on phones (`.lab-ck-off { display: none }`,
   gs-site-lab.css:2520) and re-show only `.is-base`. The reason is plain once
   measured: beside the city name there is no room. At 390px the cards are
   170px wide and "San Francisco" plus a chip overruns — at 10px AND at 8.5px,
   so shrinking the chip does not rescue it.

   Kyle spotted the room is on the TIME row instead. It is: the time text is
   ~60px inside a 148px row, leaving **88-90px** against the 47px the widest
   chip needs. With the chip there, nothing clips on any card.

   --- the trap, found by trying it ---------------------------------------

   The chip CANNOT be a child of `.tm`. `gs-runtime.js` does

     qs(".tm", c).textContent = new Intl.DateTimeFormat(...)

   on every tick, and assigning `textContent` destroys all child nodes. Moving
   the chip inside `.tm` looked right and then silently vanished within a
   second. A screenshot taken fast enough would have passed it.

   So the chip stays exactly where Lovable put it in the markup, and is placed
   visually instead. `top: 100%` on an absolutely positioned chip inside the
   header means "just below the header", so it follows the header's real
   height rather than a number copied from a measurement — if the header ever
   wraps or changes height, the chip travels with it.

   Desktop is untouched by construction: everything here is inside the same
   `width <= 719px` query Lovable used to hide the chips, so above 720px not a
   single declaration applies.

   All twelve chips move together, `.is-base` included. Leaving BASE in the
   header while the others sat on the time row would have put the same
   component in two places on one screen. */
@media (width <= 719px) {
  .gsh .lab-ck-top { position: relative; }
  .gsh .lab-ck-off,
  .gsh .lab-ck-off.is-base {
    display: inline-flex;
    position: absolute;
    top: 100%;
    right: 0;
    font-size: 8.5px;
    /* 11px, not 6: measured at 6px the chip sat 5px above the time's vertical
       centre. This lands it level with the digits. */
    margin-top: 11px;
  }
}

/* ---------------------------------------------------------------------------
   57. The remaining blanked fonts, restored on Lovable's ruling
   ---------------------------------------------------------------------------
   Lovable, 7 Nov: restore the rest. Their words — "our design system reserves
   Space Grotesk for metrics and UI; Helvetica on stats was unintended
   minifier fallout." That confirms §55's diagnosis: a build step mangled the
   `font:` shorthands into empty longhands, it was never a design change.

   Two restored, one deliberately not:

     .tl-stats b      700 15px Helvetica   -> 700 clamp(18,2vw,24) Space Grotesk
     .dk-idx strong   700 18px Helvetica   -> family only; 700/18px already right
     .lab-sh-fmt      NO CHANGE — see below

   `.lab-sh-fmt` is left alone. Their ruling asks for "500 11px", and it
   already renders exactly 500 11px JetBrains Mono. Their own source file says
   `600 10px`, so the ruling and the source disagree — and the ruling matches
   what ships today. Changing it to 600/10px would contradict the instruction
   we were just given, so the live value stands and the discrepancy is in the
   reply back to them. Checked, not assumed: this is the second time a Lovable
   spec has described a value the build already had.

   Longhands again, not the `font:` shorthand, for the same reason as §55 —
   a shorthand resets every sub-property, including ones later rules set on
   purpose. Only what is actually wrong is written back. */
.gsh .tl-stats b {
  font-family: var(--gs-ui);
  font-size: clamp(18px, 2vw, 24px);
  font-weight: 700;
  line-height: 1;
}
.gsh .dk-idx strong { font-family: var(--gs-ui); }

/* ---------------------------------------------------------------------------
   58. Sprint 1 — marquee separator, and the lean shelf nav (12 Nov)
   ---------------------------------------------------------------------------
   Two rules, both approved by Lovable after the inspection pass. Task 3 of
   that sprint needed no work: the footer phone alignment has been live since
   §22 and measures flush at 375-720, so nothing here touches it.

   1. THE MARQUEE SEPARATOR

   The locked sheets set an indigo four-pointed star between marquee items:

     .gsh .marquee-track span::before { content: "✦"; color: var(--gs-indigo); margin-right: 2.5rem; }
       -- gs-site-lab.css:1320  (locked, governs 6 of the 7 marquee pages)
       -- gs-home.css:99        (not locked, governs /games)

   Declared TWICE, which is why this override is written at the same selector
   rather than something narrower: one rule has to beat both, and both are
   (0,2,2). Matching specificity plus loading later is what wins; a lighter
   selector would have silently lost on six pages and worked on none.

   The marquee is on 7 pages, not 15 -- index, about, shelf, playlists,
   quick-q, tools-home and games. `gs-static-fix.css` is the last stylesheet
   on all of them, verified on the live site.

   Colour is a flat `#C24A3A` rather than `var(--gs-red)`, and that is
   deliberate. Lovable asked for even weight in both themes, and measured
   against the real grounds the flat value gives 4.39:1 on #F7F3EE and 4.13:1
   on #07080D. `--gs-red` is #C8443A light but #E05D52 dark, which climbs to
   5.58:1 -- brighter in dark, which is the thing they were avoiding. The
   token is six points of red and six of green away from this value; close
   enough to look like a duplicate, far enough to behave differently.

   For the record on what was rejected: the brief first proposed
   `var(--gs-ink-2)` at 0.4 opacity, which composites to 1.88:1 light and
   2.12:1 dark, and terracotta at 0.45, which is 1.87:1 and 1.66:1. Both are
   near-invisible rather than subtle, and the 0.45 terracotta is the weakest
   of the set on dark despite reading as the warmer choice.

   `margin-right: 2.5rem` is restated because the override replaces the whole
   declaration block; dropping it would collapse the spacing to the track's
   own 2.5rem gap alone.

   2. THE LEAN SHELF NAV

   Four nav items each carry a `<small>` subtitle, in both the desktop
   dropdown and the mobile drawer -- 8 per page across 15 pages, 120 in all.
   Hidden in CSS rather than deleted from the markup, on Lovable's ruling:
   one declaration in one file against 120 deletions across two source trees,
   and reversible if the subtitles come back.

   Scope checked before writing it. `.shf-i` is nav-only: the live /shelf page
   carries exactly 8, the same as every other page, and its own cards use
   different classes. The locked sheet does carry
   `.gsh .shp-wrap .shf-i small` (0,3,1), which would outrank this rule, but
   `.shp-wrap` appears nowhere in the shipped markup -- so that rule matches
   nothing today. If a future shelf layout introduces `.shp-wrap`, this hide
   will stop applying inside it, which is the safe direction to fail.

   Specificity: (0,2,1) against the locked `.gsh .shf-i small` (0,2,1) at
   gs-site-lab.css:3216 and gs-lab.css:1483. Matched, loading later. The
   `.shf-dd` variant (0,3,1) only sets white-space and overflow, so it has
   nothing to say about `display`. --- */
.gsh .marquee-track span::before {
  content: "·";
  color: #C24A3A;
  margin-right: 2.5rem;
}

.gsh .shf-i small { display: none; }

/* --- §59: Sprint 2 Navigation Lean & Mobile Carousel Uniformity -------------
   Tasks 1 and 3 were dropped after the inspection: the nav already reads
   "Tools" and "Games" (the only matches were prose in the homepage meta and
   og descriptions), and Lovable confirmed the baseline work was done in
   earlier passes. What follows is Task 2 + 2b only.

   THE TARGET IS /tools, AND IT IS LOCKED. `gs-lab.css` already ships exactly
   the block the brief specifies — `grid-auto-flow: column`,
   `grid-auto-columns: 86%`, `gap: 12px`, snap on container and children — at
   `@media (max-width:759px)`. So this section does not invent a pattern; it
   brings the other three hubs onto the one already in the locked sheet, and
   759 is used throughout because that boundary cannot be changed.

   Why not the 760px in the brief: at exactly 760 the locked
   `@media (min-width:720px) and (max-width:999px)` rule puts /tools into a
   two-column row grid at (0,4,0), so a 760 rule would be the one width where
   ours applies and the locked carousel does not. /quick-q also ran to 768, so
   760 would have left 761-768 non-uniform — the same 8px window that has
   caught this grid before. §49 is lowered to 759 in the same pass.

   1. /PLAYLISTS — two live bugs, not one

   (a) Snapping was inert. The container sets `scroll-snap-type: x mandatory`
       but the align rule targeted `> .pl-card`, and there are ZERO `.pl-card`
       elements on the page — the children are `.v2-tape`. Every child
       computed `scroll-snap-align: none`. The old rule was also scoped
       `<= 719px`, so it would have missed 720-759 even with the right class.

   (b) The first card in every tier is narrower than the rest. Measured at
       390px the tracks were `260px 287px 287px 287px`. The locked
       `.gsh[data-lab] .pl-grid` (gs-site-lab.css:3955) sets
       `grid-template-columns: repeat(auto-fit, minmax(min(100%,260px),1fr))`
       at (0,3,0), which defines ONE explicit track; `grid-auto-columns` only
       ever sized the implicit tracks after it. So track 1 came from the
       locked rule and tracks 2-4 from ours.

       This is why the rule below is written at `.gsh[data-lab]` (0,3,0) and
       not `.gsh` (0,2,0): `grid-template-columns: none` has to WIN, or
       `grid-auto-columns: 86%` stays inert on the first card and the tier
       stays ragged. The same locked rule is why `gap` measured 14px against
       the 12px we declared.

   The older `@media (width <= 719px)` block above is fully subsumed by this
   one and can be deleted in a tidy-up pass; it is left in place for now
   because it restates the same structural properties and removing it is a
   separate, unrelated risk.

   2. /QUICK-Q and /GAMES — flex, so flex-basis is the width

   Both run `display: flex` rather than a column grid, so 86% is set as
   `flex-basis`, not `grid-auto-columns`. Only the basis is overridden, never
   the `flex` shorthand, so the existing `0 0` grow/shrink survive — a card
   that could grow would defeat the peek.

   /games also carried `gap: 16px` from the base `.v2-arc-grid` rule at
   (0,1,0); `.gsh` prefixes take these to (0,2,0) to bring it to 12px. --- */
@media (max-width: 759px) {

  /* /playlists — (0,3,0), matching the locked rule it has to beat */
  .gsh[data-lab] .pl-grid {
    display: grid;
    grid-auto-flow: column;
    grid-template-columns: none;
    grid-auto-columns: 86%;
    gap: 12px;
    overflow-x: auto;
    scroll-snap-type: x mandatory;
    overscroll-behavior-x: contain;
    scrollbar-width: none;
    margin-inline: -20px;
    padding: 0 20px 12px;
    /* Needed the moment snapping starts working, and only then. The grid is
       pulled out to the page edge with `margin-inline: -20px` and given the
       gutter back as padding, so card 1 sits 20px into the scrollable area.
       Without `scroll-padding` the snapport starts at the padding box, the
       browser snaps card 1's start to it, and the carousel loads already
       scrolled to 20 — eating the left gutter. /tools and /quick-q both carry
       `scroll-padding: 20px` and rest at 0; this brings /playlists in line.
       It was invisible before because snap-align matched nothing. */
    scroll-padding: 0 20px;
  }
  .gsh[data-lab] .pl-grid::-webkit-scrollbar { width: 0; height: 0; display: none; }
  /* `> *`, not a named class. Lovable's ruling said `.v2-tape`, which came
     from my inspection — and that inspection read only the FIRST of the four
     grids. The tiers do not share a child class:
         music  .v2-tape
         films  .v2-reel
         audio  .v2-reel.v2-bcast
         watch  .v2-reel.v2-bcast
     Naming `.v2-tape` would have fixed one tier in four and left the other
     three free-scrolling, which is the bug we were sent to fix. Matching
     every direct child also survives the next card type. Same (0,3,0), so
     it still beats the locked template rule. */
  .gsh[data-lab] .pl-grid > * { scroll-snap-align: start; }

  /* /quick-q — (0,5,0) against §49's own rule, basis only */
  .gsh[data-lab] .qq-grid.v2-snap > .v2-slip { flex-basis: 86%; }

  /* /games — (0,2,0) over the base (0,1,0) */
  .gsh .v2-snap,
  .gsh .v2-arc-grid { gap: 12px; }
  .gsh .v2-snap > *,
  .gsh .v2-arc-grid > * { flex-basis: 86%; }
}

/* --- 60. Sprint 3 — Homepage fold flow & polish ----------------------------
   Four decisions from the inspection, all signed off, all narrower than the
   brief originally scoped.

   1. MARQUEE STRIP — top padding only

   The brief described this strip as sitting on linen Fold B (#F2ECE4) with
   hairline borders. It does not, and never did in anything measurable: on
   production it computes #F7F3EE in light and #07080D in dark, inheriting
   the body ground, with border-top and border-bottom both 0px. There is no
   background rule on it to neutralise.

   What reads as a band is that it is the one linen strip between two white
   surfaces — `.sec.lab-close` above and the `.lft` footer below. That is the
   opposite of the diagnosis and is a composition question, not a border one.

   The one real gap was clearance: padding resolved to 0 top / 30px bottom,
   so the strip had no breathing room above it. `padding-block` sets both.
   Lovable confirmed: keep the ground, fix the padding.

   `!important` is carried over from the rule this supersedes, not added — the
   locked sheet sets `padding` (shorthand) on this element, which would
   otherwise reset the block axis after us regardless of order.

   2. KICKERS — 0.12em, and why Act 01 is not in this list

   Five folds get the two-span pattern; the hero does not. Its kicker is five
   children, and `tk-live` / `tk-scene` are rewritten every second by
   `initKicker()` in gs-runtime.js, which is LOCKED. Restructuring it would
   mean changing markup that locked JS queries by class. Lovable's ruling:
   leave Act 01 alone. It renders identically, so the inconsistency is in the
   DOM only.

   Letter-spacing was genuinely inconsistent before this — measured 1.32px on
   the base `.mono`, 1.54px on `.lab-sh-kick` and **normal (0)** on
   `.lab-close-kick`. 0.12em at 11px is 1.32px, so this normalises two
   outliers rather than changing all five.

   The two spans wrap as a unit on narrow screens: `flex-wrap: wrap` with the
   delimiter kept on the first span, so a break falls after the `//` and
   never orphans it onto a line of its own.

   3. THE TOOLS SCROLL CTA

   `.tl-all` is gone and `.tl-next` replaces it. Removal was safe for layout —
   its parent is `div.wrap` at `display: block`, not a grid, and it was the
   last of three siblings.

   Worth recording what it cost: `.tl-all` was the homepage's only
   in-content link to /tools. The remaining routes are the header dropdown,
   the mobile drawer and one prose link. Lovable weighed that and chose the
   scroll chain; noting it because it is a navigation change, not a styling
   one, and it will not be obvious from the diff.

   `btn btn-solid` matches #desk exactly, per the ruling. `.btn-pill` was
   dropped from the spec — it has zero definitions anywhere on the site and
   would have been inert. --- */

/* 1. Marquee strip — clearance on both sides */
.gsh .mq-shelf.mq-end { padding-block: clamp(20px, 3vh, 32px) !important; }

/* 2. The Act kickers, Acts 02-06 */
.gsh .dk-kick,
.gsh .tl-kick,
.gsh .world-kick,
.gsh .lab-sh-kick {
  display: flex;
  flex-wrap: wrap;
  gap: 0 0.36em;
  font-family: var(--gs-mono);
  font-size: 11px;
  line-height: 1.4;
  letter-spacing: 0.12em;
  text-transform: uppercase;
  color: var(--gs-ink-3);
  margin: 0 0 10px;
}

/* Act 06 takes the same box but NOT the colour, and it is a separate block
   for exactly that reason.

   Scoping to `.lab-close-bench` to spare the note band lifted this selector
   from (0,2,0) to (0,3,0) — which TIES the locked
   `.gsh[data-lab] .lab-close-kick { color: var(--gs-green-active); ... }`
   and then wins on load order. So the scoping fix silently took the green
   as well, in light only; dark survived because its rule is
   `.gsh[data-lab][data-theme="dark"]` at (0,4,0).

   Shipped that way for one deploy. I re-measured tracking after scoping and
   not colour, so the check I ran could not have caught it. Omitting `color`
   here lets the locked green through in both themes, which is what Lovable
   confirmed they want for the closing roadmap band. */
.gsh .lab-close-bench .lab-close-kick {
  display: flex;
  flex-wrap: wrap;
  gap: 0 0.36em;
  font-family: var(--gs-mono);
  font-size: 11px;
  line-height: 1.4;
  text-transform: uppercase;
  margin: 0 0 10px;
}

/* The heads these sit in are centred on three of the five folds, and a flex
   child does not inherit `text-align` the way the old inline span did. */
.gsh .sec-head:not(.left) .dk-kick,
.gsh .sec-head:not(.left) .tl-kick,
.gsh .sec-head:not(.left) .world-kick,
.gsh .sec-head:not(.left) .lab-sh-kick { justify-content: center; }

/* #vote needs (0,3,0). The locked sheet carries
     `.gsh[data-lab] .lab-close-kick { color: var(--gs-green-active); letter-spacing: 0px; }`
   at gs-site-lab.css:3845, which outranks the (0,2,0) block above — caught in
   verification, where four kickers measured 1.32px and this one measured
   `normal`.

   Letter-spacing only. That rule also sets a deliberate green, and a second
   theme-scoped rule below it swaps the shade in dark; the closing band is
   meant to read differently from the other four. The brief specifies the
   font and the tracking, not the colour, so the green stays and #vote keeps
   its own ink while matching everything else on type.

   SCOPED TO `.lab-close-bench`. `.lab-close-kick` appears TWICE on the
   homepage, and the two are structurally identical — both the first child of
   a `.lab-close-say`, both followed immediately by an `<h2>`. Only the
   parent band tells them apart:

       .lab-close-band.lab-close-bench  -> Act 06, the community roadmap
       .lab-close-band.lab-close-note   -> "GrowthSanity x Acquret"

   Unscoped, the first build gave the note band the Act tracking and
   `display: flex` as well. Visually that was tracking-only — both sit
   left-aligned on one line, so flex changed nothing anyone would see — but
   it is not an Act kicker and was not in scope. Caught by counting the class
   in the packaged index.html and finding two. */
.gsh[data-lab] .lab-close-bench .lab-close-kick { letter-spacing: 0.12em; }

/* 3. Tools fold -> #world scroll CTA */
.gsh .tl-next {
  display: flex;
  justify-content: center;
  margin-top: clamp(20px, 3vh, 32px);
}

/* --- 61. Sprint 4 — About polish, marquee box, kicker levelling -----------
   Tasks 1 and 3a were dropped: the operator card already stacks above the
   form below 900px, and the footer axis has been flush since §22. Both
   verified on production again before signing off.

   1. THE MARQUEE BOX

   Sprint 3 checked `.mq-shelf.mq-end`, the outer section, and found it clean
   — correctly, but it was the wrong element. The borders and beige come from
   the INNER ticker, `gs-site-lab.css:1318`:

     .gsh .marquee { border-block: 1px solid var(--gs-rule); background: var(--gs-surface); … }

   Lovable caught that from a screenshot. Matching (0,2,0) and loading later
   is enough; no `!important` needed, and it is left out so a future rule can
   still win on merit.

   2. ACT KICKER COLOUR — a token-scope problem, not a cascade one

   The `#world` kicker measured `rgb(154,162,184)` in dark against
   `rgb(142,151,171)` on Acts 02/03/05, and the obvious read is that
   something outranks our colour rule. It does not. Every rule matching that
   element sets `var(--gs-ink-3)`, ours included — checked by walking every
   stylesheet and testing `el.matches()` on each.

   The cause is that `section.sec.world` REDEFINES the token locally:

     .gsh            --gs-ink-3: #8E97AB   (dark)
     section.world   --gs-ink-3: #9AA2B8   ← lighter, deliberate for that fold

   So `var(--gs-ink-3)` is correct everywhere and simply resolves to a
   different value inside `#world`. Hardcoding `#8E97AB` on the kicker would
   work and would also plant a literal that silently diverges the next time
   the palette moves.

   Instead the value is captured once at `.gsh`, where the token is still the
   global one, and the kickers use the capture. Custom properties resolve at
   the declaring element, so `--gs-act-ink` inherits the already-computed
   `#8E97AB` down into `#world` untouched by the local override — and it
   tracks the palette in both themes with no literal anywhere.

   3. FLIP CARD KICKERS

   10px -> 11px, matching the homepage Acts. The middot stays: the back face
   lives in the locked `gs-runtime.js`, so changing the front separator alone
   would make it flip from `·` to `//` mid-interaction.

   The four dead `.ab-card-tag` spans are stripped from `about.html`. They
   never rendered — our own flip handler overwrites that element on every
   render — so this is a markup cleanup with no visual change. Copies remain
   inside `gs-runtime.js` and cannot be removed; that file is locked. --- */

/* 1. Marquee — no borders, no box */
.gsh .marquee { border-block: 0; background: transparent; }

/* 2. One kicker ink, captured before #world can redefine it */
.gsh { --gs-act-ink: var(--gs-ink-3); }

.gsh .dk-kick,
.gsh .tl-kick,
.gsh .world-kick,
.gsh .lab-sh-kick { color: var(--gs-act-ink); }

/* --- 62. ACT KICKERS 02-06: THE THREE-ROLE SYSTEM -------------------------
   Kyle, 13 Nov: the other Acts read as "bland grey text" next to Act 01.
   Correct — Act 01 is NOT flat. It carries a two-tone hierarchy that the
   other five never had:

     ACT 01   11:10:11 PM LOCAL // Act 01:   Desk          >
              <-------- green 700 -------->  grey 500   blinking green
     ACT 02   LIVE INBOX // ACT 02: THE SANITY DESK _
              <----------- all grey 500, one tone ----->  static underscore

   So this is a ROLE system lifted from Act 01, not a copy of it. Each
   kicker already contained the same three parts; index.html now marks them
   up, and the roles map 1:1 onto Act 01's own elements:

     .ak-stat   status label   <- analogue of Act 01's .tk-live  (green 700)
     .ak-sep    the "//"       <- analogue of Act 01's <i>       (green 500)
     b          act number     <- Act 01 already uses <b>        (green 700)
     .ak-scene  scene name     <- analogue of Act 01's .tk-scene (grey 500)
     .gs-cursor blinking block <- already defined and animated in the locked
                                  sheet at gs-site-lab.css:3082, and already
                                  disabled under prefers-reduced-motion at
                                  :3094. Nothing to add here.

   WHY THERE IS NO FIGHT WITH ACT 06's LOCKED GREEN. The locked sheet sets
   `.gsh[data-lab] .lab-close-kick { color: var(--gs-green-active) }` at
   (0,3,0) on the PARENT, so every child inherits green. Inheritance loses
   to any direct rule on the child, whatever its specificity — so colouring
   the children below simply never lets that green through, and the locked
   rule is left untouched rather than overridden. Act 06 ends up two-tone
   like the rest, which is Kyle's call of 13 Nov. It does reverse Lovable's
   13 Nov ruling ("keep the green on #vote exactly as built"), and that goes
   to them as an open item — Act 06 still carries the same green as the
   other five on its label and act number; what it loses is being green
   edge to edge.

   WHY `--gs-act-ink` AND NOT `--gs-ink-3` for the scene. `section.sec.world`
   locally redefines `--gs-ink-3`, so the raw token would hand Act 04 a
   different grey from the other four. §61 captured `--gs-act-ink` on `.gsh`
   for exactly this reason. Reuse it; never reach for the raw token here.

   SIZE STAYS 11px. Act 01 is 11.5px, but Kyle asked for consistency with
   each other rather than a copy of Act 01, and 11px is the section-kicker
   size Lovable signed off in Sprint 3. Changing it would reopen a settled
   decision for no design gain.

   WRAP BEHAVIOUR IS UNCHANGED. Sprint 3 established that at 390 the two
   long kickers break after the "//". The markup keeps exactly TWO flex
   children — .ak-head and .ak-body — so the only break point is still the
   same one. The new parts are inline inside them.

   THE SECOND `.lab-close-kick`. "GrowthSanity x Acquret" lower down the
   page carries the same class with no act structure. Every selector here
   is keyed to the new child classes, which exist only in the five Acts, so
   that one is untouched without needing a .lab-close-bench scope. --- */

/* 1. Status label and act number — the accent pair */
.gsh .ak-stat,
.gsh .ak-body b {
  color: var(--gs-green-active, #1b6b38);
  font-weight: 700;
}

/* 2. The separator — accent, lighter, and never italic (it is an <i>) */
.gsh .ak-sep {
  color: var(--gs-green-active, #1b6b38);
  font-style: normal;
  font-weight: 500;
}

/* 3. The scene name — the quiet half of the pair */
.gsh .ak-scene {
  color: var(--gs-act-ink);
  font-weight: 500;
}

/* 4. Dark. The locked sheet carries TWO different dark greens: the hero
      kicker uses rgb(53,189,107) at gs-site-lab.css:3714 and `.lab-sh-kick b`
      uses rgb(34,197,94) at :3442. Standardising on the hero's, so all six
      Acts share one accent in dark as well as light.
      `.gsh[data-theme="dark"] .ak-body b` is (0,3,1), matching the locked
      `.gsh[data-theme="dark"] .lab-sh-kick b`; this file loads last, so a
      match is enough. */
.gsh[data-theme="dark"] .ak-stat,
.gsh[data-theme="dark"] .ak-sep,
.gsh[data-theme="dark"] .ak-body b {
  color: rgb(53, 189, 107);
}

/* 5. Act 05 alone had a decorative chevron on its <b> via a locked
      ::before (gs-site-lab.css:3441). Act 01 has none, so it would now be
      the one Act out of six carrying it. Suppressed at matching
      specificity, later file. */
.gsh .ak-body b::before { content: none; }

/* --- 63. PERSONA v5 CONTAINERS -------------------------------------------
   Lovable approved Option (c) on 14 Nov: keep every box exactly as it is and
   let `cover` fill it, so the discarded 25% of a 3:4 source comes off the
   BOTTOM (legs) rather than the head. No geometry moves, so §34's hero
   measurements are untouched.

   The crop value is `50% 0%`, not the `50% 15%` the brief first proposed.
   Measured across the alpha bounding boxes of all 16 files: headroom is 2.2%
   at the tightest, 4.5% mean. Under `cover` a square box shows a 75% vertical
   slice starting at `object-position Y x 25%`, so 15% starts the slice at
   3.75% and cuts the crown on 7 of the 16, all four Felix states among them.
   The ceiling is 8%; 0% is the only value with margin.

   `!important` as specified. The locked rule it has to beat is
   `.gsh .hero .hero-c .stage .stage-peek` at (0,5,0) — gs-site-lab.css:1857,
   which is also the ONLY rule giving .stage-peek an aspect-ratio. The
   homepage uses `hero-c`; the `hero-center` and `hero-split` variants carry
   no aspect-ratio and are not used here, so this one rule governs. --- */

.gsh .hero .hero-c .stage .stage-peek {
  object-fit: cover !important;
  object-position: 50% 0% !important;
}

/* The idle card. Locked rule is `.gsh .dk-zero img` (0,2,1) at :2040, which
   also pins width/height to 96px — left alone, only the fit changes. */
.gsh .dk-zero img {
  object-fit: cover !important;
  object-position: 50% 0% !important;
}

/* The /games console badges are NOT `.ib-card-avatar` — that class does not
   exist anywhere on the site. The real selector is `.v2-arc-who img`, and it
   lives in THIS file at §-rule 2394, not in a locked sheet. So the
   object-position went in at source rather than as an !important override of
   our own rule. These were the only centre-anchored surface on the site, and
   a 3:4 source under `50% 50%` would have cropped them to the torso. */

/* --- 64. THE TRANSLATOR CAPTION ------------------------------------------
   Kyle flagged the `#world` caption as a duplicate kicker; Lovable approved
   stripping the green lead and muting what remains.

   TWO CORRECTIONS TO THE INSTRUCTION, both measured.

   1. There is only ONE variant in the markup, not two. `.lab-ck-hint-d` and
      `.lab-ck-hint-m` are defined in the locked sheet at :3387-:3390 but
      appear NOWHERE in index.html. Only `.lab-ck-hint-lead` is real, and
      `@media (width <= 719px)` already hides it — so this was a desktop-only
      problem and mobile already rendered what we are now making universal.
      Removing the span is the whole change; no mobile counterpart exists.

   2. `color: var(--gs-ink-3)` here would have been a NO-OP. `section.sec.world`
      redefines `--gs-ink-3` locally, and inside #world it resolves to
      `#9aa2b8` — byte-identical to `--gs-ink-2`, which is what the caption
      already computes to via `.gsh .lab-clock-sec .lab-ck-hint` at :3384.
      Setting the token would have changed nothing on screen.

      `--gs-act-ink` is the token §61 captured on `.gsh` for exactly this
      reason. It holds the real ink-3 (`#8e97ab`) and actually mutes the line,
      which is what the instruction was for. Same token the Act scene names
      use, so the muted tier stays one colour across the fold.

   Specificity: `.gsh .lab-clock-sec .lab-ck-hint` is (0,3,0); this matches it
   and loads later. --- */

.gsh .lab-clock-sec .lab-ck-hint { color: var(--gs-act-ink); }

/* ===========================================================================
   §65  GAMES HUB — the operational desk               added 16 Nov 2026
   ---------------------------------------------------------------------------
   Scoped `.gh-*` throughout. Nothing above this block is edited.

   Why not reuse /quick-q's `.qq-tab` rail: every one of its 22 rules is
   scoped `.gsh[data-lab]`, and this page's main shell has no `data-lab`.
   Adding the attribute would switch on 118 rules, 11 of whose class families
   are already in this markup (.sec, .mono, .lft, .gs-dock, .gsh-chrome,
   .dt-kick, .mq-shelf, .tk-mqwrap, .v2-arc-grid, .v2-snap). So: new rules.

   The rail is pure CSS. `662288b3` is not touched and no JS runs here.
   ========================================================================= */

/* ---- fold clearance ----------------------------------------------------- */
/* gs-pages.css sets `.gsh .pg-sec{padding-block:clamp(40px,7vw,80px) ...}`,
   which is 80px of top padding at desktop widths and pushed the deck to
   start at 360px and finish at 798px - slicing the lower Play bars on a
   1366x768 laptop. Element+class here is (0,2,1) against that rule's
   (0,2,0), so it wins on specificity and needs no !important. */
.gsh main.gh-main { padding-top: 28px; padding-bottom: 48px; }

/* ---- hero -------------------------------------------------------------- */
.gsh .gh-hero { margin-bottom: 14px; }
.gsh .gh-hero .dt-kick { margin-bottom: 8px; }
.gsh .gh-h1 { margin: 0 0 10px; font-size: clamp(34px, 3.8vw, 44px); }
.gsh .gh-sub {
  font: 400 16px/1.5 var(--gs-ui, "Space Grotesk", sans-serif);
  color: var(--gs-ink-2);
  margin: 0;
  max-width: 54ch;
}
/* The terracotta wave only ever existed as `.hero-h1 em.ink::after`, so it
   could never reach this page's `.pg-h` h1. Re-pointed here, reusing the
   gsh-ink keyframe that gs-home.css already ships. `em.g` was specified but
   resolves to plain green text and would have dropped the Caslon italic. */
@media (prefers-reduced-motion: no-preference) {
  .gsh .gh-h1 em.ink { position: relative; }
  .gsh .gh-h1 em.ink::after {
    content: ""; position: absolute; left: 0; right: 0; bottom: -0.06em;
    height: 0.22em; clip-path: inset(0 100% 0 0);
    animation: gsh-ink 1.1s .4s var(--gs-ease) forwards;
    background: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 200 14' preserveAspectRatio='none'%3E%3Cpath d='M2 9 C 40 3, 90 12, 130 6 S 185 4, 198 8' fill='none' stroke='%23C24A3A' stroke-width='3' stroke-linecap='round'/%3E%3C/svg%3E") no-repeat center/100% 100%;
  }
}

/* ---- filter rail ------------------------------------------------------- */
.gsh .gh-tabs { display: flex; flex-wrap: wrap; gap: 8px; margin: 0 0 16px; }
/* sr-only, but still focusable: display:none and visibility:hidden both drop
   the input out of the tab order and the rail becomes mouse-only. */
.gsh .gh-in {
  position: absolute; width: 1px; height: 1px;
  clip-path: inset(50%); overflow: hidden; white-space: nowrap;
}
.gsh .gh-tab {
  display: inline-flex; align-items: center; cursor: pointer;
  font: 600 11px/1 var(--gs-mono, "JetBrains Mono", monospace);
  letter-spacing: .1em; text-transform: uppercase;
  color: var(--gs-ink-2);
  padding: 9px 14px; border: 1px solid var(--gs-rule); border-radius: 999px;
  background: var(--gs-raised);
  transition: color .18s var(--gs-ease), border-color .18s var(--gs-ease), background .18s var(--gs-ease);
}
.gsh .gh-tab:hover { color: var(--gs-ink); border-color: var(--gs-rule-2); }
.gsh .gh-in:checked + .gh-tab {
  color: var(--gs-on-accent); background: var(--gs-ink); border-color: var(--gs-ink);
}
/* the input is invisible, so the pill carries the focus ring */
.gsh .gh-in:focus-visible + .gh-tab {
  outline: 2px solid var(--gs-green-active); outline-offset: 2px;
}

/* ---- deck -------------------------------------------------------------- */
.gsh .gh-deck {
  display: grid; grid-template-columns: repeat(2, 1fr);
  max-width: 1080px; margin: 0 auto; gap: 14px;
}
/* Hiding siblings does not recentre a grid, so the filtered state needs its
   own track definition. */
.gsh .gh-hub:has(.gh-in:not([value="all"]):checked) .gh-deck {
  grid-template-columns: minmax(0, 532px);
  justify-content: center;
}
.gsh .gh-hub:has(#gh-marketing:checked)  .gh-card:not([data-seat="marketing"]),
.gsh .gh-hub:has(#gh-sales:checked)      .gh-card:not([data-seat="sales"]),
.gsh .gh-hub:has(#gh-product:checked)    .gh-card:not([data-seat="product"]),
.gsh .gh-hub:has(#gh-leadership:checked) .gh-card:not([data-seat="leadership"]) { display: none; }

/* ---- card -------------------------------------------------------------- */
.gsh .gh-card {
  display: flex; flex-direction: column; padding: 18px;
  background: var(--gs-raised); border: 1px solid var(--gs-rule);
  border-radius: 16px; text-decoration: none; color: var(--gs-ink);
  transition: transform .2s var(--gs-ease), border-color .2s var(--gs-ease);
}
.gsh .gh-card:hover { border-color: var(--gs-rule-2); }
@media (prefers-reduced-motion: no-preference) {
  .gsh .gh-card:hover { transform: translateY(-2px); }
  .gsh .gh-card:hover .gh-ar { transform: translateX(3px); }
}
.gsh .gh-card:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }

/* seat tone, light then dark. Set per seat rather than inline, because an
   inline custom property cannot respond to [data-theme]. */
.gsh .gh-card[data-seat="marketing"]  { --c: #1B6B38; }
.gsh .gh-card[data-seat="sales"]      { --c: #8E2F2A; }
.gsh .gh-card[data-seat="product"]    { --c: #3B4A5E; }
.gsh .gh-card[data-seat="leadership"] { --c: #8A4F0C; }   /* was #8C6D37: 4.70 on the
   #FDFCF9 card, the lowest kicker on the page. #8A4F0C is the canonical homepage
   leadership token and measures 6.39 here — note that is on #FDFCF9, not the
   5.58 the brief quoted, which is the figure for the #F2ECE4 homepage tile. */
.gsh[data-theme="dark"] .gh-card[data-seat="marketing"]  { --c: #5FBF84; }
.gsh[data-theme="dark"] .gh-card[data-seat="sales"]      { --c: #E07C74; }
.gsh[data-theme="dark"] .gh-card[data-seat="product"]    { --c: #8FA6C4; }
.gsh[data-theme="dark"] .gh-card[data-seat="leadership"] { --c: #D2AE6A; }

.gsh .gh-split { display: grid; grid-template-columns: 96px 1fr; gap: 16px; align-items: start; }
.gsh .gh-col   { display: block; min-width: 0; }
.gsh .gh-av {
  width: 96px; height: 96px; border-radius: 12px;
  object-fit: cover; object-position: top center;
  border: 1px solid var(--gs-rule); background: var(--gs-surface);
}
.gsh .gh-meta {
  display: block;
  font: 600 11px/1 var(--gs-mono, "JetBrains Mono", monospace);
  letter-spacing: .1em; text-transform: uppercase;
  color: var(--gs-ink-2); margin-bottom: 6px;
}
.gsh .gh-meta .seat { color: var(--c); font-weight: 700; }
/* 50px, not the specified 48. Two lines at 21px/1.15 measure 48.3px, so 48
   had no headroom: at exactly 768 (the narrowest 2-column width) card 02's
   title rendered 49px and the row spread by 1px. 50 holds at every width
   from 360 to 1366 and costs 4px of page height. */
.gsh .gh-card h2 {
  font: 700 21px/1.15 var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  letter-spacing: -.02em; margin: 0 0 8px; min-height: 50px;
  color: var(--gs-ink);
  /* NOT display:flex. The spec had `display:flex; align-items:flex-start`
     here, which turns the text node and the <em> into separate flex items
     and collapses the whitespace between them - the titles rendered as
     "That Can't BeRight". Block already top-aligns inside the min-height
     envelope, which is all the flex was doing. */
}
.gsh .gh-card h2 em {
  font-family: var(--gs-serif, "Libre Caslon Text", Georgia, serif);
  font-style: italic; font-weight: 400; color: inherit;
}
/* 38px locks both action-line wraps to two lines */
.gsh .gh-line {
  display: block; font-size: 13.5px; line-height: 1.4;
  color: var(--gs-ink-2); margin: 0; min-height: 38px;
}
.gsh .gh-go {
  margin-top: 16px; width: 100%; height: 44px;
  display: flex; align-items: center; justify-content: center; gap: 8px;
  border-radius: 999px;
  background: var(--gs-cta-gradient, #1A1D2B); color: var(--gs-on-accent);
  font: 700 14px/1 var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  letter-spacing: -.01em;
}
.gsh .gh-ar { flex: none; transition: transform .2s var(--gs-ease); }

/* ---- mobile: horizon snap-deck ----------------------------------------- */
@media (max-width: 767px) {
  .gsh .gh-deck {
    display: flex; overflow-x: auto; scroll-snap-type: x mandatory;
    gap: 12px; padding: 4px 16px 20px; margin: 0 -16px;
    scrollbar-width: none; -webkit-overflow-scrolling: touch;
    overscroll-behavior-x: contain;
  }
  .gsh .gh-deck::-webkit-scrollbar { display: none; }
  .gsh .gh-card { flex: 0 0 85%; scroll-snap-align: start; }
  /* filtering a flex deck: the centring track above does not apply here */
  .gsh .gh-hub:has(.gh-in:not([value="all"]):checked) .gh-deck {
    grid-template-columns: none; justify-content: flex-start;
  }
  /* The portraits are 480x640 (3:4). A full-width 120px-tall plate was a
     2.16 ratio box, so `cover` showed about 35% of the image height -
     forehead and eyes only. A square tile keeps the whole face. */
  .gsh .gh-split { grid-template-columns: 80px 1fr; gap: 12px; align-items: start; }
  .gsh .gh-av { width: 80px; height: 80px; border-radius: 10px; }
}

/* ---- §3.5 display typography ------------------------------------------- */
/* The numerics were tagged var(--gs-ui) (Space Grotesk). Primary metrics are
   Helvetica Neue Bold; HUD telemetry is JetBrains Mono. (0,2,0) beats the
   existing (0,1,0), which matters because that rule lives in each game page's
   own inline <style>. */
.gsh .tile-v { font-family: var(--gs-head); font-weight: 700; }
.gsh .hud-v  { font-family: var(--gs-mono); }

/* ---- §66 Operator Clearance Slip (all 4 game result panels) ------------- */
/* Replaces the ad-hoc Game Over box with one docket across seats 01-04.
   Scoped `.gsh #g-result`, which carries an ID, so every rule here clears the
   (0,2,0) declarations in each game page's own inline <style> without needing
   to match them one for one.

   Token note, measured on the live pages rather than taken from the brief:
   `--gs-green-active` is NOT defined on any game page - only `--gs-green-text`
   (#1B6B38 light / #3FD08A dark). The brief's token would have computed to
   nothing and left the eyebrow inheriting body colour, so every green below
   reads --gs-green-text with the brand hex as an explicit fallback.
   Likewise `--gs-rule` resolves to rgba(255,255,255,.09) in dark, not the
   #232738 the brief quotes; we use the token so both themes stay correct. */
.gsh #g-result {
  border: 1px solid var(--gs-rule, #E8E1D8);
  background: var(--gs-raised, #FDFCF9);
  border-radius: 12px;
  padding: 18px 20px 14px;
  text-align: left;
}

/* Seat identity kicker */
.gsh #g-result #g-r-eyebrow {
  font-family: var(--gs-mono, "JetBrains Mono", monospace);
  font-size: 11px;
  letter-spacing: .08em;
  text-transform: uppercase;
  color: var(--gs-green-text, #1B6B38);
  margin: 0 0 8px;
}

/* Verdict headline */
.gsh #g-result #g-r-title {
  font-family: var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  font-weight: 700;
  font-size: 24px;
  line-height: 1.2;
  letter-spacing: -.035em;
  color: var(--gs-ink, #1A1D2B);
  margin: 0 0 8px;
}

/* Telemetry row: big score, then mono run stats */
.gsh #g-result #g-r-score {
  font-family: var(--gs-head, "Helvetica Neue", Helvetica, Arial, sans-serif);
  font-weight: 700;
  font-size: 30px;
  line-height: 1;
  font-variant-numeric: tabular-nums;
  color: var(--gs-ink, #1A1D2B);
  margin: 0 0 6px;
}
.gsh #g-result .slip-tele {
  font-family: var(--gs-mono, "JetBrains Mono", monospace);
  font-size: 12px;
  letter-spacing: .02em;
  color: var(--gs-ink-2, #5D5348);
  margin: 0 0 10px;
  font-variant-numeric: tabular-nums;
}

/* 3-state pip matrix. The engine writes `sq`, `sq ok`, `sq bad`; a round that
   expired keeps the bare `sq`, which is why the third state is selected by
   absence rather than by a class of its own. */
.gsh #g-result #g-r-grid {
  display: flex;
  flex-wrap: wrap;
  gap: 6px;
  margin: 0 0 10px;
}
.gsh #g-result #g-r-grid .sq {
  width: 14px; height: 14px; border-radius: 4px;
}
.gsh #g-result #g-r-grid .sq.ok  { background: var(--gs-green-text, #1B6B38); }
.gsh #g-result #g-r-grid .sq.bad { background: var(--gs-red-fill, #C8443A); }
/* expired / unanswered: hatched, so it reads as "no answer" rather than as a
   third verdict. Outline keeps it visible on both grounds where a 40% fill
   would wash out on #1A1D2B. */
.gsh #g-result #g-r-grid .sq:not(.ok):not(.bad) {
  background: transparent;
  border: 1px solid var(--gs-rule, #E8E1D8);
  background-image: repeating-linear-gradient(
    -45deg,
    var(--gs-ink-3, #6E655B) 0 1px,
    transparent 1px 4px);
  opacity: .55;
}

.gsh #g-result #g-r-best {
  font-family: var(--gs-mono, "JetBrains Mono", monospace);
  font-size: 11px;
  color: var(--gs-ink-3, #6E655B);
  margin: 0 0 4px;
}

/* Perforated hairline above the action row - the receipt tear */
.gsh #g-result .panel-actions {
  border-top: 1px dashed var(--gs-rule, #E8E1D8);
  margin-top: 10px;
  padding-top: 12px;
  display: flex;
  flex-wrap: wrap;
  gap: 10px;
  align-items: stretch;
}
.gsh #g-result .panel-actions .pbtn { flex: 1 1 auto; min-width: 0; }

/* #g-next is an <a>, so it arrives underlined while its two <button>
   siblings do not. Without this the three pills do not read as one set. */
.gsh #g-result .panel-actions a.pbtn { text-decoration: none; }
.gsh #g-result .panel-actions a.pbtn:hover { text-decoration: none; }

/* Share demoted to a text link so the bar keeps three equal pills while the
   WhatsApp / LinkedIn / X row stays reachable from inside the panel. */
.gsh #g-result .slip-share {
  display: block;
  width: 100%;
  margin: 10px 0 0;
  padding: 0;
  border: 0;
  background: none;
  font-family: var(--gs-mono, "JetBrains Mono", monospace);
  font-size: 11px;
  letter-spacing: .04em;
  color: var(--gs-ink-3, #6E655B);
  text-align: left;
  text-decoration: underline;
  text-underline-offset: 3px;
  cursor: pointer;
}
.gsh #g-result .slip-share:hover { color: var(--gs-ink, #1A1D2B); }
.gsh #g-result .slip-share:focus-visible {
  outline: 2px solid var(--gs-green-text, #1B6B38);
  outline-offset: 2px;
}

/* Anomaly recap rows */
.gsh #g-recap #g-recap-list .recap-k {
  font-family: var(--gs-mono, "JetBrains Mono", monospace);
  font-size: 11px;
  letter-spacing: .04em;
  color: var(--gs-red-fill, #C8443A);
}

@media (max-width: 479px) {
  .gsh #g-result { padding: 18px 16px 16px; }
  .gsh #g-result #g-r-title { font-size: 21px; }
  .gsh #g-result #g-r-score { font-size: 30px; }
  /* three pills stack rather than squeeze to unreadable widths */
  .gsh #g-result .panel-actions .pbtn { flex: 1 1 100%; }
}

/* ---- §69 /about flip cards: horizontal single-fold carousel ------------- */
/* Lovable's 17 Nov 1:1 port. Replaces the §68 / `.ab-card` grid entirely —
   that markup no longer exists on the page, so every `.ab-card` rule above
   (including all of §68) is now inert. Left in place rather than pruned, per
   the pre-launch CSS freeze; flagged for the cleanup pass.

   Scoped `.gsh` throughout, which the brief's selectors were not. House
   convention, and it keeps these rules above the page's own `.gsh` resets.
   No appearance change — `.fc*` collides with nothing on the site (checked).

   Three DEVIATIONS from the brief's CSS, each marked at its rule. */

/* Single-fold pass. `.ab-why` is padding-block 88px (72 base, 56 below
   560px); this reclaims 84px of it. The controls bar loses 12px of its
   top margin. `.fc-track` margin-top drops to 16 too - the brief left it
   at 24, but adjacent sibling margins COLLAPSE to the larger of the two,
   so a 16px bottom margin on .fc-ctrls alone would reclaim nothing. */
.gsh .ab-why { padding-block: 56px 60px; }
.gsh .fc-ctrls { display: flex; align-items: center; justify-content: space-between; margin: 12px 0 16px; gap: 16px; }
.gsh .fc-flip-all {
  display: inline-flex; align-items: center; gap: 8px; padding: 10px 18px; border-radius: 999px;
  background: var(--gs-raised); border: 1px solid var(--gs-rule-2); font: 700 13px/1 var(--gs-head);
  color: var(--gs-ink); cursor: pointer; transition: background .15s ease, border-color .15s ease;
}
.gsh .fc-flip-all:hover { background: var(--gs-surface); border-color: var(--gs-ink); }
.gsh .fc-flip-all:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }
.gsh .fc-nav { display: inline-flex; align-items: center; gap: 10px; }
.gsh .fc-counter { font: 600 12px/1 var(--gs-mono); letter-spacing: .08em; color: var(--gs-ink-3); }
.gsh .fc-btn {
  all: unset; display: inline-flex; align-items: center; justify-content: center; width: 34px; height: 34px;
  box-sizing: border-box; border-radius: 50%; background: var(--gs-raised); border: 1px solid var(--gs-rule-2);
  font: 700 14px/1 var(--gs-head); color: var(--gs-ink); cursor: pointer; transition: background .15s ease;
}
.gsh .fc-btn:hover:not(:disabled) { background: var(--gs-surface); border-color: var(--gs-ink); }
.gsh .fc-btn:disabled { opacity: .35; cursor: not-allowed; }
.gsh .fc-btn:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }

/* Horizontal snap track. */
/* The carousel runs at every width. A 4-column desktop grid was built and
   measured first; it cannot work. `.wrap` caps at 1140, so four columns are
   273px, and the footer needs 377:

       206 widest tag + 10 gap + 117 pill + 44 face padding = 377

   At 273 all eight tags were cut and ran under the CTA pill, headlines went
   to 3-4 lines, and the section grew to 707. Four columns of 377 would need
   a 1556px container, which a 1440 screen cannot give even full-bleed.

   380 is the card width: it clears the 377 floor with 3px to spare and puts
   the section at 671. Card 03 keeps a 32px sliver at the right edge, which
   is the accepted cost.

   padding-block gives the -2px hover lift room; `overflow-x: auto` makes
   this a scroll container, so overflow-y resolves to auto and anything
   outside the padding box is clipped. margin-block cancels it, with 8 on
   top rather than -8 so that 8 margin + 8 padding reproduces the 16px gap
   under the controls bar. */
.gsh .fc-ctrls {
  display: flex; align-items: center; justify-content: space-between;
  gap: 16px; margin: 12px 0 16px;
}
.gsh .fc-track {
  display: grid; grid-auto-flow: column;
  grid-auto-columns: min(380px, calc(100vw - 48px));
  gap: 16px; width: 100%;
  overflow-x: auto; scroll-snap-type: x mandatory;
  padding-block: 8px 14px; margin-block: 8px -14px;
  -webkit-overflow-scrolling: touch; scrollbar-width: none;
}
.gsh .fc-track::-webkit-scrollbar { display: none; }

/* Phones and tablets: one card fills the track, so centre its snap point. */
@media (width <= 1023px) {
  .gsh .fc { scroll-snap-align: center; }
}

/* Outer card container. Both faces occupy grid cell 1/1, so the card's height
   is max(front, back) and flipping cannot move anything by construction. */
.gsh .fc {
  all: unset; box-sizing: border-box; display: grid; cursor: pointer; border-radius: 16px;
  background: var(--gs-raised); border: 1px solid var(--gs-rule-2);
  box-shadow: 0 1px 0 var(--gs-rule), 0 2px 6px -4px rgba(26, 29, 43, .18);
  transition: transform .15s ease, border-color .15s ease; scroll-snap-align: start;
}
.gsh .fc:hover { transform: translateY(-2px); }
.gsh .fc:active { transform: translateY(0); }
.gsh .fc:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }
.gsh .fc.is-on { border-color: color-mix(in oklab, var(--gs-green-active) 55%, var(--gs-rule-2)); }

/* Face and the five locked slots. */
.gsh .fc-face { grid-area: 1 / 1; display: flex; flex-direction: column; padding: 22px 22px 18px; min-width: 0; box-sizing: border-box; }
.gsh .fc .fc-back, .gsh .fc.is-on .fc-front { visibility: hidden; pointer-events: none; }
.gsh .fc.is-on .fc-back { visibility: visible; pointer-events: auto; }

.gsh .fc-kick { font: 500 11px/1.3 var(--gs-mono); letter-spacing: .08em; text-transform: uppercase; color: var(--gs-ink-3); }
.gsh .fc-head {
  display: flex; align-items: flex-start; min-height: 56px; margin-top: 12px;
  font: 700 20px/1.22 var(--gs-head); letter-spacing: -.01em; color: var(--gs-ink);
}
.gsh .fc-body { display: block; min-height: 52px; margin-top: 8px; font: 400 14px/1.5 var(--gs-ui); color: var(--gs-ink-2); }

/* DEVIATION 1: `.fc-viz > div` -> `.fc-viz > *`. Every <div> in the brief's
   markup is a <span> here, because <button> takes phrasing content only and
   <div> inside it is invalid HTML. One child either way. */
.gsh .fc-viz { display: block; margin-top: auto; padding-top: 14px; width: 100%; }
.gsh .fc-viz > * {
  border-radius: 10px; background: var(--gs-surface); border: 1px solid var(--gs-rule);
  padding: 10px 12px; height: 94px; box-sizing: border-box; display: flex; flex-direction: column; justify-content: center;
}
/* Dark: the brief asked for the panel at --gs-raised (#1A1D2B) over #0E1018.
   That is the wrong way round here - the CARD is already --gs-raised #1A1D2B
   and the section behind it is --fold-b #0E1018, so a #0E1018 panel would
   match the fold and a #1A1D2B panel would vanish into the card. The panel
   stays the inset (same direction as light, where #F2ECE4 insets into
   #FDFCF9); what was actually muddy is the rule, at #ffffff17 / 9%. */
.gsh[data-theme="dark"] .fc-viz > * { border-color: rgba(255, 255, 255, .14); }

/* `flex-wrap` rather than a tuned breakpoint. The tag is nowrap and the pill
   is flex:none, so on a card too narrow for both the tag was silently cut -
   at 360 that was 5 of 8 footers, by up to 23px. Wrapping drops the pill to
   its own row instead, which cannot fail on string length the way a width
   threshold can. Inert at 768 and above: nothing clips there, so nothing
   wraps, and the desktop footer is byte-identical to before. */
.gsh .fc-foot {
  display: flex; align-items: center; justify-content: space-between;
  flex-wrap: wrap; gap: 10px; row-gap: 8px;
  margin-top: 14px; padding-top: 12px; border-top: 1px solid var(--gs-rule);
}
.gsh .fc-tag { display: inline-flex; align-items: center; gap: 8px; font: 500 11px/1.3 var(--gs-mono); color: var(--gs-ink-2); min-width: 0; white-space: nowrap; }
.gsh .fc-pip { flex: none; width: 7px; height: 7px; border-radius: 50%; background: var(--gs-red); }
.gsh .fc-back .fc-pip { background: var(--gs-green-active); }

.gsh .fc-pill {
  flex: none; display: inline-flex; align-items: center; gap: 6px; padding: 7px 12px; border-radius: 999px;
  font: 700 12.5px/1 var(--gs-head); border: 1px solid var(--gs-ink); color: var(--gs-ink); background: transparent;
  transition: background .15s ease;
}
.gsh .fc:hover .fc-front .fc-pill { background: var(--gs-surface); }
/* The CTA glyphs are now inline SVG. `flex: none` stops the arrow being
   squeezed when the footer is tight, and the pill is already inline-flex
   with a 6px gap, so no margin is needed. */
.gsh .fc-arr { flex: none; display: block; }
/* --gs-on-accent rather than --gs-raised: same rendered colour in both
   themes, but it is the token that means "text on a filled surface". */
.gsh .fc-back .fc-pill { background: var(--gs-ink); color: var(--gs-on-accent); border-color: var(--gs-ink); }

/* --- the six artifacts --- */

/* 01 front: flatline telemetry */
.gsh .fc-bars { display: grid; gap: 7px; width: 100%; }
.gsh .fc-bar { display: grid; grid-template-columns: 82px 1fr 88px; align-items: center; gap: 8px; font: 500 10.5px/1 var(--gs-mono); color: var(--gs-ink-2); }
.gsh .fc-bar-v { text-align: right; color: var(--gs-ink); }
.gsh .fc-bar-t { height: 9px; border-radius: 3px; background: repeating-linear-gradient(90deg, var(--gs-rule-2) 0 2px, transparent 2px 6px); }
.gsh .fc-bar-t i { display: block; height: 100%; border-radius: 3px; background: var(--gs-red); }
.gsh .fc-bar:nth-child(2) .fc-bar-t i { opacity: .7; }
.gsh .fc-bar:last-child .fc-bar-v { color: var(--gs-red-text); font-weight: 700; }

/* 01 back: cartridge bay */
.gsh .fc-bays { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 6px; width: 100%; }
.gsh .fc-bay { display: flex; flex-direction: column; justify-content: space-between; gap: 6px; padding: 8px; border-radius: 6px; background: var(--gs-raised); border: 1px solid var(--gs-rule-2); border-top: 3px solid var(--gs-green-active); }
.gsh .fc-bay b { font: 700 13px/1.1 var(--gs-head); color: var(--gs-ink); }
.gsh .fc-bay small { font: 500 10px/1.2 var(--gs-ui); color: var(--gs-ink-2); }

/* 02 front: DM toasts */
.gsh .fc-toasts { display: grid; gap: 6px; width: 100%; }
.gsh .fc-toast { display: flex; align-items: center; gap: 8px; padding: 6px 8px; border-radius: 6px; background: var(--gs-raised); border: 1px solid var(--gs-rule-2); font: 500 11.5px/1.25 var(--gs-ui); color: var(--gs-ink); }
.gsh .fc-toast:nth-child(2) { margin-left: 12px; }
.gsh .fc-toast i { flex: none; font: 600 9.5px/1 var(--gs-mono); font-style: normal; letter-spacing: .06em; text-transform: uppercase; padding: 3px 5px; border-radius: 3px; color: var(--gs-red-text); font-weight: 700;
  background: color-mix(in oklab, var(--gs-red-text) 8%, transparent); }

/* 03 front: unread channel noise */
.gsh .fc-chan { display: grid; gap: 8px; width: 100%; }
.gsh .fc-chips { display: flex; flex-wrap: wrap; gap: 5px; }
.gsh .fc-chip { display: inline-flex; align-items: center; gap: 5px; padding: 4px 6px; border-radius: 5px; background: var(--gs-raised); border: 1px solid var(--gs-rule-2); font: 500 10.5px/1 var(--gs-mono); color: var(--gs-ink-2); }
.gsh .fc-chip i { font-style: normal; min-width: 16px; text-align: center; padding: 1px 3px; border-radius: 999px; background: var(--gs-red); color: #fff; font-weight: 700; font-size: 9.5px; }
/* The struck-through line becomes a query/result pair: query in ink-3, the
   null result in red-text. No CSS came with the brief for these three
   classes, so this is written to match the rest of the card. */
.gsh .fc-search-box { display: flex; flex-direction: column; gap: 2px; min-width: 0; }
.gsh .fc-search-q { font: 500 10.5px/1.25 var(--gs-mono); color: var(--gs-ink-3); white-space: nowrap; overflow: hidden; text-overflow: ellipsis; }
.gsh .fc-search-res { font: 700 10.5px/1.25 var(--gs-mono); color: var(--gs-red-text); }

/* 03 back: Quick Q incident docket */
.gsh .fc-dock { display: grid; gap: 5px; width: 100%; }
.gsh .fc-dock-r { display: grid; grid-template-columns: 18px 68px 1fr; align-items: baseline; gap: 6px; font: 500 11px/1.25 var(--gs-ui); color: var(--gs-ink); }
.gsh .fc-dock-r i { font: 600 9.5px/1 var(--gs-mono); font-style: normal; color: var(--gs-green-active); }
.gsh .fc-dock-r b { font: 600 9.5px/1 var(--gs-mono); letter-spacing: .06em; text-transform: uppercase; color: var(--gs-ink-3); }
.gsh .fc-dock-r:last-child span { color: var(--gs-green-active); font-weight: 600; }

/* 04 front: conflicting department scorecards */
.gsh .fc-coll { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 5px; width: 100%; }
.gsh .fc-coll-c { display: flex; flex-direction: column; gap: 3px; padding: 6px 8px; border-radius: 6px; background: var(--gs-raised); border: 1px solid var(--gs-rule-2); }
.gsh .fc-coll-c small { font: 500 9.5px/1 var(--gs-mono); letter-spacing: .04em; text-transform: uppercase; color: var(--gs-ink-3); white-space: nowrap; }
.gsh .fc-coll-c b { font: 700 14.5px/1 var(--gs-head); color: var(--gs-ink); }

/* 04 back: synchronized room telemetry */
.gsh .fc-matrix { display: flex; flex-direction: column; justify-content: space-between; height: 100%; width: 100%; }
.gsh .fc-mat-seats { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 4px; }
.gsh .fc-mat-seats span { display: inline-flex; align-items: center; gap: 5px; font: 600 10px/1 var(--gs-mono); color: var(--gs-ink); padding: 4px 6px; border-radius: 4px; background: var(--gs-raised); border: 1px solid var(--gs-rule-2); }
.gsh .fc-mat-seats i { color: var(--gs-green-active); font-style: normal; font-size: 8px; }
/* DEVIATION 2: `display: block` added. The brief had this as a <div>, which is
   block by default; as a <span> it would go inline and lose its centring,
   border-top and padding. */
.gsh .fc-mat-rule { display: block; font: 600 10px/1.2 var(--gs-mono); letter-spacing: .04em; text-transform: uppercase; color: var(--gs-green-active); text-align: center; border-top: 1px solid var(--gs-rule); padding-top: 4px; }

/* DEVIATION 3: reduced motion. The brief specified none; the rest of the site
   honours it (`.ab-card`, `.gsh .fc` hover lift, smooth scroll in §8 JS). */
@media (prefers-reduced-motion: reduce) {
  .gsh .fc, .gsh .fc-btn, .gsh .fc-flip-all, .gsh .fc-pill { transition: none; }
  .gsh .fc:hover { transform: none; }
  .gsh .fc-track { scroll-behavior: auto; }
}

/* DEVIATION 4: the unread-count badge on the channel chips fails AA in dark.
   White on --gs-red is 4.83:1 in light (#C8443A) but only 3.59:1 in dark
   (#E05D52), at 9.5px/700 — small text, so it needs 4.5. Measured on painted
   pixels, not estimated. Light is left exactly as specified; dark takes the
   ground colour instead of white, which reads 5.30:1. */
.gsh[data-theme="dark"] .fc-chip i { color: var(--gs-surface); }

/* ---- §70 Footer nav tap targets on phones ------------------------------ */
/* WCAG 2.5.8 (AA) wants 24x24. The footer column links render 21px tall at
   phone widths, so they miss by 3.

   The selector in the approved brief was `.gsh .lft nav li a`. There is no
   <nav> in this footer - the lists sit in `.lft-cols > div > ul > li` - so
   that rule would have matched nothing and shipped as a silent no-op.
   Checked against the live DOM before writing this.

   Scoped to `.lft-cols` only, and deliberately NOT applied to:
     .lft-legal a   "Privacy & Terms", 16px but INLINE INSIDE A SENTENCE,
                    which 2.5.8 exempts. It is also `display: inline`, where
                    vertical padding does not grow the line box, so padding
                    would inflate the hit area into the lines above and
                    below without changing what you see.
     .lft-share a   already 46x46
     .lft-brand a   already 38px

   The column links are `inline-flex`, so padding-block does increase the
   box: 21 + 8 = 29px. Shared footer, so this lands on every page. */
@media (width <= 640px) {
  .gsh .lft .lft-cols li a { padding-block: 4px; }
}

/* ---- §71 Homepage Acts 02-05: one header type scale -------------------- */
/* Lovable 17 Nov. Three things the brief's selectors could not do on their
   own, each checked against the live DOM and cascade first:

   1. `.dk-h1` is scoped `.gsh #desk .dk-head h2 .dk-h1` in gs-site-lab.css,
      so putting the class on Act 03 and Act 05 does nothing. The break rule
      is generalised here to all four acts.
   2. `.world-sec` and `#shelf` do not exist. The section is
      `.lab-clock-sec` / `#world`, and Act 05 is `#picks`, not `#shelf`.
   3. `.lab-sh-h` did not exist either - it is added to Act 05's h2 by this
      release, so it is styled here rather than assumed.

   Specificity: every rule below is one class deeper than the gs-site-lab
   rule it replaces - (0,3,1) against (0,2,1), (0,3,0) against (0,2,0) - so
   it wins regardless of file order or media query. The rules it has to beat:
     .gsh .dk-head h2                     base, and @media (width >= 720px)
     .gsh .tl-head h2                     base, and @media (width <= 420px)
     .gsh .lab-clock-h                    base, and @media (width >= 1000px)
     and a shared landscape rule at (orientation: landscape) and
     (height <= 520px) and (width >= 640px), preserved below. */

/* The phone break, generalised off #desk. 719px is the existing breakpoint. */
@media (width <= 719px) {
  .gsh .dk-sec .dk-head h2 .dk-h1,
  .gsh .tl-sec .tl-head h2 .dk-h1,
  .gsh .lab-shelf-sec .lab-sh-head h2 .dk-h1 { display: block; }
}

/* Desktop: one scale for all four acts. */
.gsh .dk-sec .dk-head h2,
.gsh .tl-sec .tl-head h2,
.gsh .lab-clock-sec .lab-clock-h,
.gsh .lab-shelf-sec .lab-sh-head h2 {
  font-size: clamp(34px, 2.6vw, 40px);
  line-height: 1.15;
  letter-spacing: -0.025em;
}

/* Phone. */
@media (width <= 719px) {
  .gsh .dk-sec .dk-head h2,
  .gsh .tl-sec .tl-head h2,
  .gsh .lab-clock-sec .lab-clock-h,
  .gsh .lab-shelf-sec .lab-sh-head h2 {
    font-size: 24px;
    line-height: 1.2;
  }
}

/* Short landscape screens keep their own smaller size. gs-site-lab.css
   already did this for Acts 02 and 03; without it a 800x500 landscape phone
   would take the desktop clamp and lose most of the fold to a headline.
   Extended to Acts 04 and 05 so all four behave alike. */
@media (orientation: landscape) and (height <= 520px) and (width >= 640px) {
  .gsh .dk-sec .dk-head h2,
  .gsh .tl-sec .tl-head h2,
  .gsh .lab-clock-sec .lab-clock-h,
  .gsh .lab-shelf-sec .lab-sh-head h2 {
    font-size: clamp(22px, 3.2vw, 28px);
    line-height: 1.15;
  }
}

/* ---- §72 Ghost header leak: #world, #picks, #vote ---------------------- */
/* The same two declarations §36 already gives #tools and Lovable gives
   #desk at gs-site-lab.css:3811. Measured before, on the live page:

                  desktop 1440            phone 390
                  scroll-margin  pad      scroll-margin  pad
     #desk          0px          96px       0px          90px   treated
     #tools         0px          96px       0px          90px   treated
     #world       100px        58.5px      94px          40px   leaked
     #picks       100px        58.5px      94px          40px   leaked
     #vote        100px          56px      94px          36px   leaked

   The three untreated ids fall through to `.gsh[data-lab] [id]`
   (gs-site-lab.css:3810) and land 94-100px down, so the previous fold shows
   through the floating pill's transparent gaps.

   `!important` is required for #world and #picks, not a shortcut:
   gs-site-lab.css:3722 sets `padding-block … !important` on
   `section.sec.lab-clock-sec` and `section.sec.lab-shelf-sec` at (0,4,1).
   #vote is `.lab-close` and not in that list, but is kept here for one
   readable rule. Our selector is (1,2,0) - an id outranks any number of
   classes - so the cascade is won on specificity; the !important only
   matches theirs.

   Also beaten, by the same id specificity: gs-site-lab.css:3500, inside
   `@media (width >= 1000px)`, which gives .lab-clock-sec its own
   scroll-margin-top and a 36px padding-top. That rule is why #world behaved
   differently above and below 1000px. */
.gsh[data-lab] #world,
.gsh[data-lab] #picks,
.gsh[data-lab] #vote {
  scroll-margin-top: 0 !important;
  padding-top: calc(var(--hd-h, 72px) + 24px) !important;
}

/* ---- §73 The forward CTA chain, completed ------------------------------ */
/* Acts 04 and 05 had no onward CTA, so the chain that runs hero -> #desk ->
   #tools -> #world stopped there. Both new wrappers mirror `.lab-desk-next`
   exactly, which is the existing centred forward-CTA container.

   The buttons are `.btn.btn-solid` with the arrow copied verbatim from the
   three existing CTAs - `0.85em` at stroke-width 1.9, with aria-hidden and
   focusable="false". The brief's variant was 16px at stroke-width 1.5 and
   carried neither attribute; matching the existing three keeps the chain
   visually identical and keeps the arrow out of the a11y tree. */
.gsh .lab-ck-next,
.gsh .lab-sh-next {
  border-top: 1px solid var(--gs-rule);
  justify-content: center;
  padding: 16px;
  display: flex;
}
@media (width <= 719px) {
  .gsh .lab-ck-next,
  .gsh .lab-sh-next { padding: 12px; }
}

/* ---- §74 The bottom overscroll band: two-stop canvas -------------------- */
/* Lovable's option (b), chosen over flattening the footer, because §6 above
   is Kyle's 29 Oct call that the footer is a LIFTED panel and §1 sets `body`
   to the page ground so no pale band shows at the page edges. One flat
   colour cannot serve both ends; a gradient can.

   The surfaces that were disagreeing:

                 footer             html / body
     light    rgb(255,255,255)    rgb(247,243,238)
     dark     rgb(11,14,28)       rgb(7,8,13)

   `.gsh` covers the whole document — measured 1440x5733 against a 5732px
   document — and `body` paints its own box, so this gradient is NEVER seen
   inside the page. It is only visible where the canvas extends past the
   document: iOS rubber-band overscroll. Top overscroll takes the first
   stop (the ground, unchanged), bottom overscroll takes the last (the
   footer panel). The hard stop at 50% is never rendered.

   Specificity: the page's own inline <style> carries
   `:root{background-color:#F7F3EE}` at (0,1,0), and being inline it also
   comes last in document order. `html:root` is (0,1,1) and
   `html[data-theme="dark"]:root` is (0,2,1), so both win on specificity
   rather than order. `background-image` is used, not the `background`
   shorthand, so the inline background-color stays as the fallback. */
html:root {
  background-image: linear-gradient(to bottom, #F7F3EE 0 50%, #FFFFFF 50% 100%);
}
html[data-theme="dark"]:root {
  background-image: linear-gradient(to bottom, #07080D 0 50%, #0B0E1C 50% 100%);
}

/* ---- §75 Forward-CTA containers: match .lab-desk-next's REAL state ------ */
/* My §73 said it mirrored `.lab-desk-next` and it did not. I copied the
   first declaration in gs-site-lab.css (border-top 1px, padding 16px);
   later rules in that file override it, and the effective computed state at
   390 is `border-top: 0px none; padding: 0px; margin-top: 20px`. So §73
   shipped a 1px rule across the fold that the fold it was copying had not
   had for a long time. That grey line is mine. */
.gsh .lab-ck-next,
.gsh .lab-sh-next {
  border-top: 0 !important;
  padding: 0 !important;
  margin-top: clamp(20px, 3vh, 32px);
}
@media (width <= 719px) {
  .gsh .lab-ck-next,
  .gsh .lab-sh-next { margin-top: 20px; }
}

/* ---- §76 Acts 04 and 05: one-line headlines on desktop ----------------- */
/* Both were capped, and both caps were deliberate until now:
     .lab-clock-h     max-width 30ch   (gs-site-lab.css:2524)
     .lab-sh-head     max-width 720px
   Lifting them from 1000px up lets each headline sit on one line; the caps
   still apply below that, so the phone break is untouched. */
@media (width >= 1000px) {
  .gsh .lab-clock-sec .lab-clock-h { max-width: none; }
  .gsh .lab-shelf-sec .lab-sh-head { max-width: none; }
  .gsh .lab-shelf-sec .lab-sh-head h2 { max-width: none; }
}

/* ---- §77 Mobile anchor clearance: the overshoot, reverted --------------- */
/* This section used to set `scroll-margin-top: -14px` with an inflated
   `padding-top: calc(var(--hd-h,72px) + 38px)` to chase a 10-15px sliver of
   the previous fold that Kyle and Lovable saw on real iPhone Safari and that
   never reproduced here. It was shipped reasoned, not reproduced, and it made
   production worse: the stranded pill Lovable reported was this rule.

   The mechanism, now understood. A negative scroll-margin lands the section
   top ABOVE the viewport top, which pulls the previous section's last 14px
   into the scroll position. In a desktop preview pane that band hides under
   the 68px sticky header, so it measured clean. On iOS Safari the dynamic
   toolbar resizes the VISUAL viewport independently of the layout viewport
   and exposes the band - and what lives at the bottom of each previous
   section is its forward-CTA pill.

   This is a revert, not a deletion. Deleting the block would drop #tools back
   to gs-site-lab.css:3810 (`scroll-margin-top: calc(var(--hd-h,72px) + 28px)`
   = 94px), landing it 94px down rather than at 0. Both properties are set
   explicitly on all five, which is the configuration measured at 390x844:

     anchor    section top    heading clears the 68px header by
     #desk          0px       54px
     #tools         0px       64px
     #world         0px       64px
     #picks         0px       74px
     #vote          0px       85px

   `--hd-h` resolves to 66px live, so padding-top computes to 90px: the header
   overlays its first 68px, leaving 22px before the heading box. No negative
   offset means nothing from the previous section can sit above the fold. */
@media (width <= 719px) {
  .gsh[data-lab] #desk,
  .gsh[data-lab] #tools,
  .gsh[data-lab] #world,
  .gsh[data-lab] #picks,
  .gsh[data-lab] #vote {
    scroll-margin-top: 0 !important;
    padding-top: calc(var(--hd-h, 72px) + 24px) !important;
  }
  /* Item 4: Act 02's own bottom padding was the larger half of the 198px gap
     between "Explore the tools" and "Pick a tool." 90px of what remains is
     the anchor clearance restored above, so 159px is the floor worth taking. */
  .gsh[data-lab] #desk { padding-bottom: 28px !important; }
}

/* ---- §78 Act 04 mobile: reclaim what padding can reclaim --------------- */
/* See the note: the stated goal - 6 clocks plus the CTA inside one 390x844
   fold - is not reachable by tightening padding. Measured breakdown before:

     padding-top        90     (the §72 clearance; reducing it re-opens §72)
     sec-head          163     kicker 15 + h2 58 + note 48 + margins
     zones row          51     + 12 margin
     clocks grid       577     3 rows x 186 + 2 x 9.6
     CTA                65
     padding-bottom     40
                     -----
                      1017  against an 844 viewport, before Safari's bar

   Even at zero gaps the grid alone is 558 and the head 163, so the floor is
   around 870. The lever is card height or row count, not padding. What
   follows takes the ~45px that padding honestly gives. */
@media (width <= 719px) {
  /* The cards are `.clock.lab-ck-card`, styled at gs-site-lab.css:3392 at
     (0,3,0). One more class to out-specify it - a (0,2,0) rule here was
     silently losing, which is how it shipped doing nothing the first time. */
  .gsh .lab-clock-sec .clocks .clock.lab-ck-card { padding: 6px 10px; }
  .gsh .lab-clock-sec .clocks { gap: 6px; }
  .gsh .lab-clock-sec .sec-head { margin-bottom: 10px; }
  .gsh .lab-clock-sec .lab-world-zones { margin-bottom: 8px; }
}

/* ---- §79 Act 06 and the footer: mobile vertical spacing ---------------- */
@media (width <= 719px) {
  .gsh[data-lab] #vote { padding-bottom: 28px; }
  .gsh .lft { padding-block: 2.5rem 2rem; }
}

/* ---- §80 Act 04 mobile fold: four clocks, compact head ----------------- */
/* §78 took everything padding could give and the fold was still 151px over.
   The remaining levers were content, and Lovable picked two: four clocks
   instead of six, and the note paragraph out of the way.

   The zone toggle is untouched. gs-runtime.js still owns `.zone-on`
   (ZONES holds two sets of six), and gs-site-lab.css:3766 still hides
   anything not in the active zone below 719px. This adds a second
   condition: a card also has to be marked `data-m`. Four cards in each zone
   carry it, so mobile shows four of the active six and the toggle keeps
   swapping which four.

   A positional selector cannot do this - DOM order interleaves the zones
   (Austin, zone 1, is the sixth card), so `nth-child` cannot mean "the
   first four of the active zone". Hence the marker.

   Dropped on mobile only, and still there on desktop and on the toggle's
   other half: Toronto, San Francisco, Singapore, Sydney. */
@media (width <= 719px) {
  /* The `:not([data-m])` filter that used to sit here is gone. It rendered
     four cards against a zone tab that reads "6 hubs" verbatim, and the
     four cards it hid were Toronto, San Francisco, Singapore and Sydney -
     so zone 2 ("The global arc") showed no APAC hub at all on mobile. The
     height it bought is recovered by the §81 compaction instead.

     The note stays hidden. It costs 48px and pushes Act 04's CTA from 752
     to 830, which defeats the fold budget the compaction just bought. */
  .gsh .lab-clock-sec .sec-head .note { display: none; }
}

/* ---- §81 Act 04 mobile: Option B card compaction ----------------------- */
/* Restoring the sixth clock per zone costs a third grid row. Measured:
   6 cards with no compaction is a 556px grid and a 931px section. This takes
   the card from 180px to 140px, the grid to 433px and the section to 792px,
   with the CTA bottom at 752px.

   What drives the card height is `.lab-ck-flip > .lab-ck-face`: 77px of
   content, 95px on the four cards whose quote wraps to three lines at 390px
   (Bengaluru, New York, Toronto, San Francisco). It is NOT `.lab-ck-msg` -
   that rule (gs-site-lab.css:3111, min-height 4.2em) matches ZERO elements on
   this page and belongs to a different clock implementation. Anything aimed
   at it is a silent no-op.

   Why ~350px was asked for and not delivered: 3 rows inside 350px is 116px a
   card, and the fixed chrome above the flip is already 66px, leaving 33px for
   a face that needs 77-95px. Reaching it means clamping the quote under two
   lines, i.e. truncating mid-sentence. 433px is the no-clipping floor.

   Every visible card in both zones was checked for clipping on the front
   face, the back face, and the quote in both axes (scrollHeight AND
   scrollWidth, which is the only check that catches truncation): none. */
@media (width <= 719px) {
  .gsh[data-lab] .lab-clock-sec .clock .tm {
    margin-top: 2px !important;
    font-size: 19px !important;
    line-height: 22px !important;
  }
  .gsh[data-lab] .lab-clock-sec .clock .lab-ck-flip {
    margin-top: 5px !important;
    padding-top: 5px !important;
  }
  .gsh[data-lab] .lab-clock-sec .clock .lab-ck-k { margin-bottom: 5px !important; }
  .gsh[data-lab] .lab-clock-sec .clock .lab-ck-q {
    font-size: 12px !important;
    line-height: 16px !important;
    min-height: 32px !important;
  }
}

/* ---- §82 Acts 04 and 05: desktop vertical budget ----------------------- */
/* At 1366x768 Act 04's "Decompress" pill sat at 762px against a 768px
   viewport - 16px below the fold, a real failure on the most common laptop
   height. At 1440x900 both CTAs were already reachable and #world simply
   overshot the 820px target by 29px.

   `padding-top: 96px` is NOT available: it is the anchor clearance
   (--hd-h + 24), and spending it re-opens exactly what §77 above reverts.
   padding-block carries !important at gs-site-lab.css:3722 (0,4,1), so the
   bottom trim needs !important to land at all.

   Measured after:
     1440x900   #world 849 -> 808 (CTA 768)   #picks 817 -> 788 (CTA 748)
     1366x768   #world 834 -> 802 (CTA 762)   #picks 802 -> 782 (CTA 742)
   All four homepage forward CTAs inside the fold at 1366x768; no clipping on
   any of the 12 desktop cards.

   Note for anyone reading this later: `.lab-world-zones` computes
   `display: none` above 719px. The zone tabs are a mobile-only control, so
   desktop deliberately renders all 12 cities in 3 rows x 4 columns, and
   filtering desktop to 6 is not an option - nothing would reach the other
   six. The "6 hubs" label is a mobile label. */
@media (width >= 720px) {
  .gsh[data-lab] #world,
  .gsh[data-lab] #picks { padding-bottom: 40px !important; }
  .gsh[data-lab] #world .sec-head { margin-bottom: 10px !important; }
  .gsh[data-lab] #picks .sec-head { margin-bottom: 28px !important; }
  .gsh[data-lab] #world .clocks { gap: 8px !important; }
  .gsh[data-lab] #world .clock .lab-ck-flip {
    margin-top: 4px !important;
    padding-top: 4px !important;
  }
}

/* ---- §83 Forward-CTA containers: text-align parity --------------------- */
/* Everything that positions the pill already matched `.hero-ctas` exactly -
   height 40, margin-top 27, border-top 0, display flex, justify-content
   center - on all four. `text-align` was the only divergence (hero center,
   the four `start`), and under `justify-content: center` it has no effect on
   the pill itself; it would only surface if a block wrapped to loose text.
   Set for consistency, not because anything was visibly off. */
.gsh .lab-desk-next,
.gsh .tl-next,
.gsh .lab-ck-next,
.gsh .lab-sh-next { text-align: center; }


/* ---- §84 The analog quartet: motion, and the standby states ------------ */
/* Four effects, one per tier, all CSS. Three are new; Tier 01's spool spin
   (`v2spin`, line ~2175) already shipped and is left alone.

   Where each one hangs, because the pseudo-element budget is tight:
     Tier 01  `.v2-window i`        real element, already animated
     Tier 02  `.v2-frame::before`   both pseudos were free
     Tier 03  `.v2-bcast::after`    the needle, already existed as a static bar
     Tier 04  `.v2-bcast::after`    scanlines, plus the tally on `-k::before`
   Tier 03 has NO free pseudo-element left — ::before is the ruler, ::after is
   the needle. Anything further on that tier needs real markup. */

@keyframes v2seek {
  0%, 100% { transform: translateX(-85px); }
  28%      { transform: translateX(46px); }
  54%      { transform: translateX(-22px); }
  78%      { transform: translateX(85px); }
}
@keyframes v2crawl  { to { background-position-x: -24px; } }
@keyframes v2drift  { to { background-position-y: 4px; } }
@keyframes v2tally {
  0%, 100% { box-shadow: 0 0 3px var(--v2-glow); }
  50%      { box-shadow: 0 0 9px var(--v2-glow); }
}

/* Tier 02: a 35mm sprocket strip crawling along the top edge of the frame.
   24px is one full period of the gradient, so the loop is seamless. */
.gsh .v2-tier-films .v2-frame { position: relative; overflow: hidden; }
.gsh .v2-tier-films .v2-frame::before {
  content: ""; position: absolute; inset: 0 0 auto 0; height: 7px;
  background: repeating-linear-gradient(90deg,
    color-mix(in oklab, var(--v2-screen-ink) 42%, transparent) 0 6px,
    transparent 6px 24px);
  animation: v2crawl 7s linear infinite;
}
.gsh .v2-tier-films .v2-frame::after {
  content: ""; position: absolute; inset: auto 0 0 0; height: 7px;
  background: repeating-linear-gradient(90deg,
    color-mix(in oklab, var(--v2-screen-ink) 42%, transparent) 0 6px,
    transparent 6px 24px);
  animation: v2crawl 7s linear infinite;
}

/* Tier 04: the scanline overlay drifts, and the tally pip breathes. */
.gsh .v2-tier-watch .v2-bcast::after { animation: v2drift 8s linear infinite; }
.gsh .v2-tier-watch .v2-bcast .v2-bcast-k::before { animation: v2tally 3.5s ease-in-out infinite; }
/* A dead slot does not breathe. */
.gsh .v2-tier-watch .v2-bcast.is-slot .v2-bcast-k::before,
.gsh .v2-tier-watch .v2-bcast.is-static .v2-bcast-k::before { animation: none; }

/* One consolidated stop. Each effect rests somewhere deliberate rather than
   wherever its keyframe happened to be: the needle parks mid-dial, the
   sprockets and scanlines sit at their natural offset, the tally holds its
   quiet glow. */
@media (prefers-reduced-motion: reduce) {
  .gsh .v2-window i,
  .gsh .v2-tier-films .v2-frame::before,
  .gsh .v2-tier-films .v2-frame::after,
  .gsh .v2-tier-audio .v2-bcast::after,
  .gsh .v2-tier-watch .v2-bcast::after,
  .gsh .v2-tier-watch .v2-bcast .v2-bcast-k::before { animation: none !important; }
  .gsh .v2-tier-audio .v2-bcast::after { transform: translateX(-22px); }
  .gsh .v2-tier-watch .v2-bcast .v2-bcast-k::before { box-shadow: 0 0 6px var(--v2-glow); }
}


/* ---- §85 Act 05: the dual console ------------------------------------- */
/* Replaces the four-column shelf. `#picks` and `.lab-sh-next` are deliberately
   untouched: the first is read by gs-runtime.js:263 and keyed by §72/§77/§82,
   the second carries the forward-CTA contract from §75/§83. `initPicks()` now
   finds #picks but no `.lab-sh-grid` and returns early — inert, no error, and
   no reason to open a byte-frozen runtime file.

   Height budget, which is what drove this layout. Measured content ceilings
   before `Shape the roadmap` leaves the fold: 605 at 1440x900, 479 at
   1366x768, 609 at 1024x900, 494 at 390x844. 1366 is the binding one. A tile
   is a 44px bay beside two text rows, so it lands near 72px; two rows plus a
   gap is ~156, and the console closes around 290 — comfortably inside 479.
   A 2x2 of full game cards would have been ~606 and missed 1366 by 127. */

.gsh .ac-consoles { display: grid; grid-template-columns: 1fr 1fr; gap: 16px; align-items: stretch; }
.gsh .ac-console {
  display: flex; flex-direction: column; gap: 14px;
  padding: 18px 16px 16px; border-radius: 14px;
  background: var(--gs-raised); border: 1px solid var(--gs-rule);
}
.gsh .ac-head { display: flex; flex-direction: column; gap: 4px; }
.gsh .ac-console .ac-kick {
  font-size: 11px; letter-spacing: .18em; text-transform: uppercase;
  color: var(--gs-green-active);
}
.gsh .ac-sub { font-size: 14px; line-height: 1.4; color: var(--gs-ink-2); }

.gsh .ac-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 12px; }

.gsh .ac-tile {
  display: flex; align-items: center; gap: 10px;
  padding: 10px 12px; border-radius: 12px;
  background: var(--gs-surface);
  border: 1px solid color-mix(in oklab, var(--gs-ink) 12%, transparent);
  text-decoration: none; color: var(--gs-ink);
}
html[data-theme="dark"] .gsh .ac-tile, .gsh[data-theme="dark"] .ac-tile {
  background: #1A1D2B; border-color: rgba(255, 255, 255, .08);
}
.gsh .ac-tile:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }

.gsh .ac-bay {
  flex: none; width: 44px; height: 44px; border-radius: 8px; overflow: hidden;
  position: relative; background: var(--v2-screen);
}
.gsh .ac-bay img { width: 100%; height: 100%; object-fit: cover; object-position: top center; display: block; }

.gsh .ac-body { display: flex; flex-direction: column; gap: 3px; min-width: 0; }
.gsh .ac-spec { display: flex; align-items: center; justify-content: space-between; gap: 8px; }
/* `.gsh .mono` is (0,2,0) and paints --gs-ink-3; both of these are (0,3,0)
   and set their own colour. That trap has bitten five times now. */
.gsh .ac-tile .ac-seat { font-style: normal; font-size: 10px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-ink-2); }
/* `opacity: .6` removed 21 Nov: over the #F2ECE4 tile it composited
   #5D5348 to rgb(153,144,134) = 2.68:1, failing AA on all eight stamps
   (four games "60S", four playlists TAPE/REEL/CAST/FEED #001). Solid it
   reads 6.40. Dark was already pinned to 1 by §113, so this is light-only. */
.gsh .ac-tile .ac-time { font-style: normal; font-size: 10px; letter-spacing: .14em; text-transform: uppercase; color: var(--gs-ink-2); opacity: 1; }
.gsh .ac-title { font: 700 14px/1.25 var(--gs-head); color: var(--gs-ink); letter-spacing: -.01em; }
.gsh .ac-title em { font: 400 italic 1em/1.25 var(--gs-serif); }

/* Seat accents, one per discipline, measured per theme. */
.gsh .ac-tile[data-seat="marketing"] .ac-seat { color: #1B6B38; }
.gsh .ac-tile[data-seat="sales"]      .ac-seat { color: #B23A31; }
.gsh .ac-tile[data-seat="product"]    .ac-seat { color: #3B5CCC; }
.gsh .ac-tile[data-seat="leadership"] .ac-seat { color: #8A4F0C; }
html[data-theme="dark"] .gsh .ac-tile[data-seat="marketing"] .ac-seat,
.gsh[data-theme="dark"] .ac-tile[data-seat="marketing"] .ac-seat { color: #5FD08A; }
html[data-theme="dark"] .gsh .ac-tile[data-seat="sales"] .ac-seat,
.gsh[data-theme="dark"] .ac-tile[data-seat="sales"] .ac-seat { color: #FF7A70; }
html[data-theme="dark"] .gsh .ac-tile[data-seat="product"] .ac-seat,
.gsh[data-theme="dark"] .ac-tile[data-seat="product"] .ac-seat { color: #818CF8; }
html[data-theme="dark"] .gsh .ac-tile[data-seat="leadership"] .ac-seat,
.gsh[data-theme="dark"] .ac-tile[data-seat="leadership"] .ac-seat { color: #FBBF24; }

/* Card 02's bays are the four instruments in miniature, reusing the tier
   keyframes so the teaser and /playlists move in the same language. The
   needle needs its own sweep: v2seek throws +/-85px, which is wider than a
   44px bay. */
@keyframes v2seekmini { 0%, 100% { transform: translateX(-13px); } 50% { transform: translateX(13px); } }
.gsh .ac-art { background: var(--v2-screen); }
.gsh .ac-art::before, .gsh .ac-art::after { content: ""; position: absolute; }
.gsh .ac-art-tape::before {
  width: 11px; height: 11px; border-radius: 50%; border: 3px dotted var(--v2-screen-ink);
  top: 16px; left: 6px; animation: v2spin 6s linear infinite;
}
.gsh .ac-art-tape::after {
  width: 11px; height: 11px; border-radius: 50%; border: 3px dotted var(--v2-screen-ink);
  top: 16px; right: 6px; animation: v2spin 6s linear infinite;
}
.gsh .ac-art-reel::before {
  inset: 0 0 auto 0; height: 6px;
  background: repeating-linear-gradient(90deg, color-mix(in oklab, var(--v2-screen-ink) 45%, transparent) 0 5px, transparent 5px 20px);
  animation: v2crawl 7s linear infinite;
}
.gsh .ac-art-reel::after {
  inset: auto 0 0 0; height: 6px;
  background: repeating-linear-gradient(90deg, color-mix(in oklab, var(--v2-screen-ink) 45%, transparent) 0 5px, transparent 5px 20px);
  animation: v2crawl 7s linear infinite;
}
.gsh .ac-art-cast::before {
  inset: 0 0 auto 0; height: 10px;
  background: repeating-linear-gradient(90deg, color-mix(in oklab, var(--v2-screen-ink) 38%, transparent) 0 1px, transparent 1px 7px);
}
.gsh .ac-art-cast::after {
  top: 0; left: 50%; width: 2px; height: 10px; background: var(--v2-dial);
  animation: v2seekmini 9s ease-in-out infinite;
}
.gsh .ac-art-feed::before {
  inset: 0; background: repeating-linear-gradient(180deg, color-mix(in oklab, var(--v2-screen-ink) 9%, transparent) 0 1px, transparent 1px 4px);
  animation: v2drift 8s linear infinite;
}
.gsh .ac-art-feed::after {
  width: 6px; height: 6px; border-radius: 50%; bottom: 7px; right: 7px;
  background: var(--v2-glow); animation: v2tally 3.5s ease-in-out infinite;
}

.gsh .ac-cta { margin-top: auto; width: 100%; justify-content: center; }

/* Phones: the pair becomes a snap deck rather than stacking, which would run
   past 1200px against a 494px budget. Inside each console the 2x2 unfolds to
   four compact rows. */
@media (width <= 719px) {
  .gsh .ac-consoles {
    display: flex; grid-template-columns: none;
    gap: 12px; overflow-x: auto; scroll-snap-type: x mandatory;
    margin-inline: -16px; padding-inline: 16px;
    scrollbar-width: none;
  }
  .gsh .ac-consoles::-webkit-scrollbar { display: none; }
  .gsh .ac-console { flex: 0 0 86%; scroll-snap-align: start; }
  .gsh .ac-grid { grid-template-columns: 1fr; gap: 8px; }
}

@media (prefers-reduced-motion: reduce) {
  .gsh .ac-art::before, .gsh .ac-art::after { animation: none !important; }
  .gsh .ac-art-cast::after { transform: translateX(0); }
}


/* ---- §86 /about: section-by-section polish ---------------------------- */
/* All of the base styling for these lives in `gs-site-lab.css`, which stays
   locked, so each item here is either a new class the markup now carries or
   an override that has to out-specify the base. Specificities are noted
   where they are load-bearing. */

/* S2 — the Why subhead is desktop-only. Below 1024 it costs ~80px that Card
   01 needs to clear the fold with its footer buttons visible. */
@media (width <= 1023px) {
  .gsh.ab .ab-why-head p { display: none; }
  .gsh.ab .ab-why { padding-top: 32px; }
}

/* S3 — the founders kicker drops the pill. `.ab-eyebrow`'s border, radius,
   padding and green-dot ::before are at gs-site-lab.css:4105-4107; rather
   than fight them the markup now carries `.ab-founders-kicker`, so those
   rules simply stop matching on this page. Nothing was unset. */
.gsh.ab .ab-founders-kicker {
  display: inline-block;
  font: 500 11px/1 var(--gs-mono);
  letter-spacing: .12em;
  text-transform: uppercase;
  color: var(--gs-ink-3);
}
.gsh.ab .ab-founders-kicker i {
  font-style: normal;
  color: var(--gs-green-active);
  margin: 0 .35ch;
}

/* S4 — the operator card headline, off an inline style and onto fold
   authority matching the other section h2s. */
.gsh.ab .lab-card-h2 {
  font: 700 clamp(30px, 3.8vw, 48px)/1.08 var(--gs-head);
  letter-spacing: -.035em;
  margin: 12px 0 14px;
}
/* The card subhead's colour was a literal `rgb(93, 83, 72)` at
   gs-site-lab.css:2411, selector `.gsh:not([data-theme="dark"]) #card
   .lab-card-sub` = (1,3,0). `.gsh.ab #card .lab-card-sub` ties at (1,3,0) and
   loads later, so it wins without !important — and because it sets a token
   rather than a literal, it is now correct in both themes instead of one. */
.gsh.ab #card .lab-card-sub {
  font: 400 17px/1.6 var(--gs-ui);
  color: var(--gs-ink-2);
  max-width: 520px;
}
@media (width <= 719px) {
  .gsh.ab #card .lab-card-sub { font: 400 16px/1.55 var(--gs-ui); }
}

/* S5 — the closing headline now wraps each sentence in its own block so it
   can never break after "where". The existing rule was scoped `.gsh.ab`;
   index.html and shelf.html carry the same markup now, so it is widened to
   every page that uses the band. */
.gsh .ab-close-line { display: block; }

/* S6 — the feedback line, relocated out of the old `.lab-close-strip` (which
   only existed on three pages) into the footer, which is on all fifteen. */
.gsh .lft-feedback {
  display: flex; flex-wrap: wrap; align-items: center; gap: .5ch 1ch;
  margin-bottom: 1rem; padding-bottom: 1rem;
  border-bottom: 1px solid var(--gs-rule);
  font-size: 12px; letter-spacing: .04em; line-height: 1.5;
  color: var(--gs-ink-3);
}
.gsh .lft-feedback a {
  color: var(--gs-green-active);
  text-decoration: none;
  border-bottom: 1px solid color-mix(in oklab, var(--gs-green-active) 40%, transparent);
}
.gsh .lft-feedback a:hover { border-bottom-color: var(--gs-green-active); }
@media (width <= 560px) {
  .gsh .lft-feedback { gap: .35ch .75ch; font-size: 11.5px; }
}


/* ---- §87 /about: fold coverage and split-fold baselines --------------- */
/* All three base rules are in the locked gs-site-lab.css, so each item below
   out-specifies rather than edits. Note the real breakpoint is **900px**, not
   the 1024 the brief assumes: gs-site-lab.css:3988 and :1390 both switch to
   two columns at `width >= 900px`. Scoping the alignment fixes to 1024 would
   have left the 900-1023 band misaligned, so they match the layout's own
   breakpoint instead.

   1. Zero-peek hero. The base sets `min-height: calc(-74px + 100svh)` — the
   resolved form of `100svh - var(--hd-h)`. The header is `position: sticky`
   and overlays the fold rather than occupying flow, so subtracting it
   double-counts: measured peek was exactly 74px at 1440, 1280 and 1024, which
   is --hd-h to the pixel. The homepage hero uses the full 100svh and peeks 0.
   Unscoped, because the base rule only applies >=900px and the fold needs the
   floor at every width — below 900 it was absent entirely, which is why
   768x1024 peeked 293px. */
.gsh.ab .ab-hero { min-height: 100svh; }

@media (width >= 900px) {
  /* 2. Hero CTA — REVERTED 19 Nov, and the reason matters more than the fix.
     §87 moved `.ab-hero-cta` into column 1 and stretched the art to row 3 so
     their bottoms shared a grid line. The measurement was right: delta went
     +107 -> 0. The result was wrong. Locking the CTA to the bottom of a
     100svh fold tore it away from the copy it belongs to and left ~140px of
     void between the paragraph and the button on a 14" MacBook. Baseline
     alignment was the wrong target; the left column is a reading unit —
     kicker, headline, copy, action — and it should stay tight regardless of
     where the window's bottom edge happens to fall.

     The base grid placement is restored. `.ab-hero-cta` keeps its own
     row-3 full-width slot from gs-site-lab.css:3996 and is pulled back up to
     the copy by replacing that rule's 56px padding-top with a tighter
     margin. */
  .gsh.ab .ab-hero-cta { padding-top: 0; margin-top: clamp(20px, 2.5vh, 28px); }

  /* 3. Operator card split baseline — DELIBERATELY NOT ALIGNED.
     `align-items: end` was tried and reverted on Kyle's call. It does hit a
     0px delta against the card's bottom edge, but the two columns are
     genuinely different heights (art 346px, form+buttons 209px) so the 137px
     difference only moves: `end` gives 137px above the inputs and 0 below,
     where the base `center` gives 89 above and 69 below. The balanced split
     reads better beside the card than a hard baseline with a void over the
     form. A third arrangement — stretch the column and push the buttons down
     — does not work, because the grid row is sized by the art.
     Leave `.idwrap` on the base `center`. */
}

/* ---- §88 /about: the forward CTA chain -------------------------------- */
/* Folds 2 and 3 had no terminal CTA, so the next section's kicker was the
   first thing to enter the frame — "THE BACKSTORY" bleeding into the flip
   cards, "CRAFT OVER CLOUT" into the founder portraits. Each fold now ends
   on its own downward action, matching the homepage's Act chain, and Fold 4
   gets a floor cushion instead of a button because its actions already are
   the terminal controls. */
.gsh.ab .ab-fold-cta {
  display: flex;
  justify-content: center;
  margin-top: clamp(32px, 4.5vh, 48px);
  padding-bottom: clamp(24px, 3vh, 36px);
}

/* Fold 4 keeps its form actions as the terminal control and buys the gap
   with padding instead of a third button. */
.gsh.ab #card { padding-bottom: clamp(64px, 8vh, 88px); }
@media (width <= 719px) {
  .gsh.ab #card { padding-bottom: 56px; }
}

/* ---- §89 /about: anchor clearance and desktop floor --------------------- */
/* 1. The ghost-button leak. `#why`, `#founders` and `#card` were inheriting
   `scroll-margin-top: 100px`, so a forward CTA click landed the section 100px
   down and left the CTA you just pressed sitting in the transparent band
   above the floating header pill. Measured before: previous CTA bottom at
   +40px (#founders) and +12px (#card) at 1440x820.

   Same rule already proven on #desk/#tools in §77: land flush at 0 and clear
   the pill with the section's own padding instead. This is now the recorded
   standard in CLAUDE.md. */
.gsh[data-lab].ab #why,
.gsh[data-lab].ab #founders,
.gsh[data-lab].ab #card {
  scroll-margin-top: 0 !important;
  padding-top: calc(var(--hd-h, 72px) + 24px) !important;
}

@media (width >= 1000px) {
  /* 2. Adaptive floor, desktop only. Deliberately NOT `min-height: 100svh` —
     that was declined for inner folds because it breaks iPad and phone where
     the content already exceeds the screen. A clamped cushion scales down on
     shorter screens instead of cutting the CTA off. */
  .gsh.ab #founders,
  .gsh.ab #card { padding-bottom: clamp(80px, 12vh, 120px); }

  /* 3. Hero vertical balance. The hero is a centred flex column inside a
     100svh fold, so with content shorter than the fold it split the slack
     evenly — 180px above the kicker and 180px below the CTA, dropping the H1
     to y=288 on a 14" MacBook. Top-weighting it puts the title in the
     160-190 band and hands the surplus to the floor, where it reads as
     breathing room rather than a void.

     The padding is a plain clamp, NOT `calc(var(--hd-h) + ...)`. On this page
     the header sits in normal flow and the hero's own top already starts at
     y=74, so adding --hd-h again double-counts it — the first attempt put the
     H1 at 236 instead of the target band. 7vh keeps the title between 165 and
     180 from 820 through 1024. */
  .gsh.ab .ab-hero {
    justify-content: flex-start;
    padding-top: clamp(48px, 7vh, 72px);
  }
}

/* ---- §90 /about: true pinned tab (dots retired) ----------------------- */
/* The macOS traffic lights are gone from the markup, so the pinned tab is now
   the left-most thing in the bar and anchors the edge itself. §90 previously
   restored those dots to their real colours; that styling is retired with
   them.

   ONLY THE DOT RULES WERE REPLACED. The `.ab-founder-link` rules below were
   also inside the old §90 block and are unrelated to the window controls —
   they make both founder visuals clickable and give them a focus ring.
   Swapping the whole block out would have silently dropped them.

   The `.ab-dots` rule is kept as supplied even though the element no longer
   exists, so the dots stay hidden if that markup ever comes back. It matches
   nothing today. */
.gsh.ab .ab-win-bar .ab-dots { display: none; }
.gsh.ab .ab-win-bar {
  padding-left: 14px; /* comfortable inset clearing the 22px chassis curve */
}
/* Both founder visuals are now links — Kyle's photo and Megha's initials
   placeholder. The placeholder is a div, not an img, so it needed wrapping
   separately; a blanket img rule would have linked only one of the two. */
.gsh.ab .ab-founder-link { display: block; text-decoration: none; border-radius: inherit; }
.gsh.ab .ab-founder-link:focus-visible { outline: 2px solid var(--gs-green-active); outline-offset: 3px; }

/* ---- §91 Closing card: Signal vs Noise telemetry bay ------------------ */
/* Shared component — /about, / and /shelf all get it. The band was two
   children (say | notebody); they are now wrapped in `.lab-close-left` so the
   grid is copy-and-CTA on the left, instrument on the right.

   Tokens are `--v2-*`, not the `--hw-*` the spec was written against: that
   namespace does not exist here, and the spec's fallbacks would have pinned
   the bay to literals that never respond to the theme. Same for the
   animations — `hw-scan` and `hw-tally` are defined nowhere, so the scanline
   and the tally pulse would have been silent no-ops. They reuse `v2drift`
   and `v2tally`, which is also why the bay moves in the same language as the
   hardware tiers on /playlists. */
.gsh[data-lab] .lab-close-band.lab-close-note {
  grid-template-columns: 1.15fr 0.85fr;
  align-items: center;
  gap: clamp(24px, 4vw, 48px);
}
.gsh[data-lab] .lab-close-left { display: flex; flex-direction: column; gap: 18px; }

@media (width <= 899px) {
  .gsh[data-lab] .lab-close-band.lab-close-note { grid-template-columns: 1fr; gap: 28px; }
}

.gsh[data-lab] .acq-telemetry {
  background: var(--v2-screen);
  color: var(--v2-screen-ink);
  border: 4px solid var(--v2-bezel);
  border-radius: 12px;
  padding: 14px 16px;
  display: flex; flex-direction: column; gap: 12px;
  position: relative; overflow: hidden;
  box-shadow: inset 0 0 0 1px rgba(255, 255, 255, .08), 0 12px 24px -6px rgba(0, 0, 0, .25);
}
/* Scanlines, reusing the Tier 04 drift. */
.gsh[data-lab] .acq-telemetry::after {
  content: ""; position: absolute; inset: 0; pointer-events: none;
  background: repeating-linear-gradient(180deg,
    color-mix(in oklab, var(--v2-screen-ink) 5%, transparent) 0 1px, transparent 1px 4px);
  animation: v2drift 8s linear infinite;
}
.gsh[data-lab] .acq-tel-head {
  display: flex; justify-content: space-between; align-items: center;
  font-size: 10px; letter-spacing: .14em;
}
/* `.gsh .mono` is (0,2,0) and paints --gs-ink-3; every mono child here is
   (0,3,0) and sets its own colour. Sixth time this trap has come up. */
.gsh[data-lab] .acq-telemetry .acq-tel-id {
  color: color-mix(in oklab, var(--v2-screen-ink) 65%, transparent);
}
.gsh[data-lab] .acq-telemetry .acq-tel-rec {
  display: inline-flex; align-items: center; gap: 6px; color: var(--v2-glow);
}
.gsh[data-lab] .acq-tally {
  width: 6px; height: 6px; border-radius: 50%; flex: none;
  background: var(--v2-glow); box-shadow: 0 0 4px var(--v2-glow);
  animation: v2tally 3.5s ease-in-out infinite;
}
.gsh[data-lab] .acq-tel-screen {
  display: flex; flex-direction: column; gap: 10px;
  background: rgba(0, 0, 0, .35);
  border: 1px solid rgba(255, 255, 255, .06);
  border-radius: 6px; padding: 10px 12px;
}
.gsh[data-lab] .acq-channel { display: flex; flex-direction: column; gap: 4px; }
.gsh[data-lab] .acq-telemetry .acq-ch-meta {
  display: flex; justify-content: space-between;
  font-size: 9.5px; letter-spacing: .12em;
  color: color-mix(in oklab, var(--v2-screen-ink) 70%, transparent);
}
.gsh[data-lab] .acq-telemetry .acq-tag.tag-amber { color: var(--v2-dial); }
.gsh[data-lab] .acq-telemetry .acq-tag.tag-green { color: var(--v2-glow); }
.gsh[data-lab] .acq-wave-box { height: 24px; width: 100%; overflow: hidden; position: relative; }
.gsh[data-lab] .acq-svg-wave { width: 200%; height: 100%; display: block; }
.gsh[data-lab] .wave-amber { animation: acqcrawl 7s linear infinite; }
.gsh[data-lab] .wave-green { animation: acqcrawl 14s linear infinite; }
@keyframes acqcrawl { from { transform: translateX(0); } to { transform: translateX(-50%); } }
.gsh[data-lab] .acq-telemetry .acq-tel-foot {
  display: flex; justify-content: space-between; gap: 8px;
  font-size: 8.5px; letter-spacing: .14em;
  color: color-mix(in oklab, var(--v2-screen-ink) 50%, transparent);
  border-top: 1px solid rgba(255, 255, 255, .06);
  padding-top: 8px;
}
@media (prefers-reduced-motion: reduce) {
  .gsh[data-lab] .acq-telemetry::after,
  .gsh[data-lab] .acq-tally,
  .gsh[data-lab] .wave-amber,
  .gsh[data-lab] .wave-green { animation: none !important; }
}


/* ---- §92 /about desktop fold coverage SOP, and the telemetry bay polish ----
   Brief: each desktop fold is a self-contained view; the next fold stays 100%
   off-screen until the user scrolls or clicks the forward CTA. Plus a warm
   graphite chassis so the bay stops reading as a hole punched in the cream
   card in light mode.

   TWO DEVIATIONS FROM THE BRIEF AS WRITTEN, both measured, both because the
   brief's own stated goal ("0px visible") is not what its CSS produces:

   1. The brief gives every fold `min-height: calc(100svh - var(--hd-h, 72px))`.
   That is correct for the HERO and wrong for the three anchored folds, and
   the difference is where the header sits relative to each.

   The header is `position: sticky; top: 0` — it OCCUPIES FLOW, it does not
   float. (§87 recorded the opposite and that is the error being corrected
   here; it is also why CLAUDE.md's "never subtract" line needed its scope
   fixed.) So:

     - At scroll 0 the hero begins at y=74, BELOW the header box. Its
       available space is therefore `100svh - 74px`, and subtracting is right.
       Measured overshoot: 74px before, 0px after.

     - An anchored fold lands at scroll == its own document top, so its box
       begins at viewport y=0 and the sticky header re-pins ON TOP of it.
       Its available space is the FULL `100svh`. Subtract the header there and
       the fold comes up exactly --hd-h short, so the next fold shows in the
       gap. Measured with the brief's CSS: 72px of the next fold visible on
       all three, at all four viewports. With `100svh`: 0px.

   2. `var(--hd-h)` resolves to 72px on this page while the header measures
   74px, so the token form leaves a 2px overshoot. The hero uses the literal
   74px, which is also what the locked base rule already uses
   (gs-site-lab.css: `min-height: calc(-74px + 100svh)`). Fixing the token
   itself would change the anchor-clearance padding on every page and is not
   in this brief's scope — logged instead.

   SAFETY GUARD, as the brief requires: `min-height` only, never `height`, so
   content taller than the fold expands it rather than clipping. Verified at
   1280x800, where #why measures 804px against an 800px viewport.

   Scope is >=1000px per the brief; tablets and phones keep fluid auto
   heights. NOT extended down to the 900px breakpoint where the locked sheet
   switches the hero, so the 900-999px band keeps its pre-existing 74px hero
   overshoot — reported, not fixed here, because widening a geometry change
   past its brief is exactly what §77 did. */

@media (min-width: 1000px) {

  /* Fold 1 — hero. Restores the centring the locked sheet already asks for;
     §89's top-weighting (flex-start + 7vh) is superseded by load order. It was
     introduced because centring a 982px-tall box dropped the H1 to y=288; the
     box was 74px too tall at the time, which is the actual bug being fixed. */
  .gsh.ab .ab-hero {
    min-height: calc(100svh - 74px);
    display: flex;
    flex-direction: column;
    justify-content: center;
    padding-top: clamp(24px, 4vh, 48px);
    /* Upward bias, Kyle on review: dead-centre put the content top back at
       y=291, the same place §89 was built to avoid. `justify-content: center`
       splits the slack, so an oversized bottom pad lifts the block by half the
       difference. Measured lift 46-65px across 1024-1920; content top lands at
       158/123/84/199/72 above at 982/900/800/1080/768. */
    padding-bottom: clamp(120px, 17vh, 190px);
  }

  /* Folds 2 and 3 — full 100svh, CTA pinned to the floor by `margin-top:auto`. */
  .gsh.ab #why,
  .gsh.ab #founders {
    min-height: 100svh;
    display: flex;
    flex-direction: column;
    justify-content: space-between;
    box-sizing: border-box;
  }

  /* §89 left #founders on `padding-bottom: clamp(80px, 12vh, 120px)` while
     #why sits on the base 60px, which parked the two forward CTAs at 118px
     and 60px off the floor — two "floor anchors" at different heights. Matched
     to #why so the pair reads as one system. */
  .gsh.ab #founders { padding-bottom: 60px; }

  .gsh.ab #why .ab-fold-cta,
  .gsh.ab #founders .ab-fold-cta {
    margin-top: auto;
    padding-top: clamp(24px, 3vh, 40px);
    padding-bottom: clamp(24px, 4vh, 48px);
  }

  /* Fold 4 — the card builder commands the screen; the closing band with the
     telemetry bay is only revealed on a deliberate scroll. */
  .gsh.ab #card {
    min-height: 100svh;
    display: flex;
    flex-direction: column;
    justify-content: center;
    padding-bottom: clamp(60px, 8vh, 100px);
    box-sizing: border-box;
  }
}

/* ---- §92b telemetry bay: warm graphite chassis, graticule, phosphor -------
   Light mode kept the bay on `--v2-screen` (#14161E), which against the
   #F7F3EE card read as a hole rather than an instrument. Warm graphite plus a
   1px inner highlight and a real drop shadow seats it on the card instead.

   Dark is unchanged in substance — it keeps the tokens, so the bay still
   matches the Act 05 consoles and the /playlists hardware tiers there.

   CONTRAST, re-measured because the backdrop moved: lifting the screen from
   #14161E to #161922 lowers the light-mode floor from 4.86:1 to 4.79:1 on the
   8.5px RIGOR/SCALE footer. Still AA for normal text, zero failures across all
   eight nodes, but that footer was already the thinnest margin in the bay and
   it is now thinner. It is the first thing to re-check if this chassis is ever
   lightened again. (§143 shipped three contrast defects by NOT re-auditing a
   backdrop that had just moved.)

   KNOWN COHESION COST, reported not resolved: in light mode this sets the
   bezel to #232734 (cool) while `--v2-bezel` elsewhere in light is #2A2620
   (warm). The /playlists hardware quartet keeps the warm bezel, so the
   telemetry bay becomes the only cool-bezel chassis in light mode. Lovable's
   design call to make; flagged with both values. */

.gsh[data-lab] .acq-telemetry {
  background: #161922;
  /* Bezel stays on the token (#2A2620 light / #222634 dark) rather than the
     briefed literal #232734 — Kyle's call on review, so the bay matches the
     /playlists hardware tiers instead of being the only cool bezel in light. */
  border: 4px solid var(--v2-bezel);
  box-shadow:
    inset 0 0 0 1px rgba(255, 255, 255, .12),
    0 16px 32px -8px rgba(22, 25, 34, .22);
}
.gsh[data-theme="dark"][data-lab] .acq-telemetry {
  background: var(--v2-screen, #090B10);
  border-color: var(--v2-bezel, #1A1D2B);
  box-shadow:
    inset 0 0 0 1px rgba(255, 255, 255, .08),
    0 12px 28px -6px rgba(0, 0, 0, .5);
}

/* Oscilloscope graticule behind the traces. */
.gsh[data-lab] .acq-tel-screen {
  background: rgba(0, 0, 0, .45);
  background-image:
    linear-gradient(to right, rgba(255, 255, 255, .035) 1px, transparent 1px),
    linear-gradient(to bottom, rgba(255, 255, 255, .035) 1px, transparent 1px);
  background-size: 20px 12px;
  border: 1px solid rgba(255, 255, 255, .08);
}

/* Phosphor punch. The brief gives the baseline as 1.75px for both traces;
   amber is 1.75 but green is already 2 (inline attrs in the markup). The rule
   below sets both to 2.5 regardless, so the outcome is as briefed. CSS beats
   an SVG presentation attribute, so no markup change is needed. */
.gsh[data-lab] .acq-svg-wave path { stroke-width: 2.5px; }
.gsh[data-lab] .wave-amber path {
  stroke: var(--v2-dial, #E2A65A);
  filter: drop-shadow(0 0 4px rgba(226, 166, 90, .65));
}
.gsh[data-lab] .wave-green path {
  stroke: var(--v2-glow, #5FD08A);
  filter: drop-shadow(0 0 5px rgba(95, 208, 138, .75));
}

/* Left column presence. The kicker's tracking needs (0,4,0) to clear the
   locked `.gsh[data-lab] .lab-close-kick { letter-spacing: 0px }` at
   gs-site-lab.css:3845 — measured `normal` before, 1.54px after. Its colour
   is already the deliberate green at 6.55:1 and is left alone. */
.gsh[data-lab] .lab-close-say .lab-close-kick { letter-spacing: .14em; }
.gsh[data-lab] .lab-close-actions .btn-solid { margin-top: 24px; }


/* ---- §93 /about on phones, and the trace that read as sawteeth ------------
   Kyle's audit was on an iPhone 17 Pro Max — a 440x956 layout viewport, not
   the 390x844 the brief assumes, and Safari's toolbar leaves roughly 871 of
   that visible at first paint. Every number below was measured across
   440x956, 440x871, 390x844, 390x760 and 375x667 so the fix is not tuned to
   one device.

   THREE DEVIATIONS FROM THE BRIEF, each because the briefed rule does not do
   what the brief says it does. All three were measured, not reasoned.

   1. `.ab-founders-grid { grid-template-columns: 1fr 1fr }` resizes the WRONG
   GRID. That selector is the [backstory text | photo block] wrapper, which is
   deliberately one column below 1024px. The two founders live one level
   deeper in `.ab-founders`, which is ALREADY `repeat(2, minmax(0,1fr))` at
   every width — the side-by-side layout the brief asks for is what ships
   today. Applying the briefed rule splits the wrapper instead: measured
   `296.8px 87.2px`, which collapses both founder tiles to **37x37px**. Only
   the tile sizing is kept.

   2. The closing-card order rule cannot reach `.lab-close-actions`. With
   `display: contents` on `.lab-close-left` the band's flex items become
   `.lab-close-say`, `.lab-close-notebody` and `.acq-telemetry` — but the CTA
   sits INSIDE `.lab-close-notebody`, so `order: 4` on it is a no-op, and so
   is `order: 2` on `.lab-close-notebody p`. Measured result of the briefed
   CSS: **body -> CTA -> headline -> telemetry**, i.e. the copy and the button
   jump above the headline and the telemetry is still last. `display: contents`
   is applied to `.lab-close-notebody` as well, which promotes both the
   paragraph and the CTA to flex items and yields the intended
   headline -> body -> telemetry -> CTA. Safe here because that element only
   carries `display: grid; gap: 0` — no box of its own to lose.

   3. Card width cannot set the carousel peek. `.fc-track` is a grid whose
   columns come from `grid-auto-columns: min(380px, calc(100vw - 48px))`, so
   narrowing `.fc` leaves the card shrunken inside a column that still starts
   in the same place — measured: card 364px, peek unchanged at 24px. The track
   is what moves. `calc(100vw - 76px)` lands the peek at exactly 40px at every
   phone width (76 = 20px gutter + 16px gap + 40px peek), inside the brief's
   36-40px target.

   The fold floors use `min-height: 100svh` rather than the briefed fixed
   padding, for the same reason the desktop SOP does (see CLAUDE.md): an
   anchored fold lands at viewport 0 with the sticky header over it, so it
   needs the full small-viewport height. `100svh` is the toolbar-visible
   height on iOS, which is the conservative one. The briefed `padding-bottom:
   72px` on `#why` moves a 123px bleed only to 111px; `100svh` takes it to 0.

   WHAT THIS CANNOT FIX, and why — see the note file. `#founders` carries ~400px
   of backstory copy before the tiles even start. Below 390px wide it does not
   fit headline + copy + tiles + CTA in the toolbar-visible height at ANY tile
   size: tested 150/132/120px, all still short at 375x667. Tile sizing is kept
   at the briefed 150px rather than shrinking the portraits further to chase a
   fit that copy length, not geometry, is blocking. */

@media (max-width: 599px) {

  /* 1. REVERSED 20 Nov (§107). This slot used to freeze the tab wordmark on
     phones. That was my error: `@keyframes abRest` holds "growthsanity" for
     0-42% of the cycle, collapses to "gs" for 50-92%, then expands again, and
     the "gro san" I saw was simply the transition between the two — about 8%
     of a 4.8s loop, sampled from a still. It is the designed effect, not a
     rendering fault: the tab collapses to its pinned-tab short form while
     `abPin` jiggles the pin at the same moment (abPin moves at 45-51%, which
     is exactly when the text is away). Freezing the text left the pin moving
     alone, which is what Kyle reported from his phone.

     Nothing replaces it — the rule is simply gone, so phones inherit the
     locked animation that desktop already runs. `prefers-reduced-motion` is
     still honoured by the locked sheet (gs-site-lab.css:4048). */

  /* 2. Fold 1 breathing room: the hero CTA sat 29px BELOW the fold at
     390x760. Tightening the hero's own padding and the grid gap lifts it to
     +27px there and +111 at 390x844. */
  .gsh.ab .ab-hero { padding-top: 24px; padding-bottom: 48px; }
  .gsh.ab .ab-hero-grid { gap: 20px; }

  /* 3. Intentional carousel peek — on the track, not the card. */
  .gsh.ab .fc-track { grid-auto-columns: calc(100vw - 76px); }

  /* 4. Fold floors. */
  .gsh.ab #why,
  .gsh.ab #founders,
  .gsh.ab #card { min-height: 100svh; box-sizing: border-box; }
  .gsh.ab #why      { padding-bottom: 72px; }
  .gsh.ab #founders { padding-top: 56px; padding-bottom: 72px; }
  .gsh.ab #card     { padding-bottom: 80px; }

  /* 5. Founders compaction. The 4/5 portraits rendered 193x241 and pushed the
     captions under the Safari toolbar; 1:1 at 150px recovers 91px per tile.
     `.ab-founders` is left alone — it is already two columns. */
  .gsh.ab .ab-founder-img,
  .gsh.ab .ab-founder-mono {
    aspect-ratio: 1 / 1;
    max-height: 150px;
    object-fit: cover;
    border-radius: 12px;
  }
  .gsh.ab .ab-founders-grid { gap: 24px; }
  .gsh.ab #founders .ab-fold-cta { margin-top: 24px; }
}

/* Closing card: the telemetry reads as the evidence under the claim, so it
   belongs above the button, with the CTA anchoring the card. */
@media (max-width: 899px) {
  .gsh[data-lab] .lab-close-band.lab-close-note {
    display: flex;
    flex-direction: column;
  }
  .gsh[data-lab] .lab-close-left,
  .gsh[data-lab] .lab-close-notebody { display: contents; }
  .gsh[data-lab] .lab-close-say        { order: 1; }
  .gsh[data-lab] .lab-close-notebody p { order: 2; }
  .gsh[data-lab] .acq-telemetry        { order: 3; }
  .gsh[data-lab] .lab-close-actions    { order: 4; }
}


/* ---- §94 the sliver was the toolbar, and the backstory on phones ----------
   Two items, one of which turned out not to be what it looked like.

   1. THE FOLD 4 SLIVER IS A VIEWPORT-UNIT BUG, NOT A MISSING CUSHION.
   §93 gave the mobile folds `min-height: 100svh`. `svh` is the SMALL viewport
   — the height with Safari's toolbar SHOWING. The moment the toolbar retracts
   on scroll the visible area grows to `lvh`, the section is suddenly shorter
   than the viewport, and the next fold shows in the difference.

   Reproduced exactly: with the viewport at 956 (toolbar hidden) and the
   section pinned to 871 (what `100svh` resolves to with the toolbar up), both
   `#card` and `#why` leak **85px** of the next fold — and 85px is the toolbar.
   The brief's proposed `+24px` cushion would have covered 24 of those 85.

   `100lvh` sizes the fold to the toolbar-HIDDEN height instead, so it is
   never shorter than the viewport in either state: full when the toolbar is
   away, taller than the viewport (harmless) when it is showing. Measured 0px
   in both states.

   `#why` had the identical latent defect and gets the same change — same
   mechanism, same one line. It cannot make the untested case worse, which is
   the §77 test: a taller fold can only ever cover more, never less.

   `#founders` is left on `svh` because its content exceeds both values, so
   the floor never applies there.

   2. BACKSTORY COPY ON PHONES. Two paragraphs (405px at 440 wide, 643px at
   390) are replaced by one on <600px, wrapped so desktop is untouched. The
   mobile paragraph measures 304px / 331px — a saving of 101px and 102px, and
   the whole reason Fold 3's CTA can clear the toolbar at all.

   The small `#founders` tightening below (padding-top 56->40, grid gap
   24->20, CTA margin 24->16) is NOT in the brief. The copy swap alone leaves
   the CTA at -4px at 390x760 — still under the toolbar, which fails the
   brief's own verification target. These three take it to +8. The portrait
   size Lovable specified (150px) is deliberately left alone; dropping it to
   140px would buy another 10px if they want it. */

@media (min-width: 600px) { .ab-mob-only  { display: none !important; } }
@media (max-width: 599px) { .ab-desk-only { display: none !important; } }

@media (max-width: 599px) {
  /* Sized to the toolbar-hidden height so neither toolbar state leaks. */
  .gsh.ab #why,
  .gsh.ab #card { min-height: 100lvh; }

  /* Buys the last 36px so Fold 3's CTA clears the toolbar on a 390 phone. */
  .gsh.ab #founders { padding-top: 40px; }
  .gsh.ab .ab-founders-grid { gap: 20px; }
  .gsh.ab #founders .ab-fold-cta { margin-top: 16px; }
}


/* ---- §95 /about phones: de-clutter Fold 2, unpack Fold 3 ------------------
   Visual polish only — both folds already fit. Everything is inside
   `@media (max-width: 599px)`; desktop is not touched.

   HIDING THE FLIP-ALL IS SAFE, CHECKED RATHER THAN ASSUMED. `.fc-flip-all` is
   a real button (`#fcFlipAll`) that gs-static-fix.js looks up by id. Two
   things make `display: none` safe here: the element stays in the DOM, so
   `getElementById` still resolves and no handler throws; and every `.fc` card
   carries its own click listener that toggles only itself, so a card can
   still be flipped by tapping it or its own pill. Hiding the global control
   removes a duplicate affordance, not the only one.

   ONE NUMBER IN THE BRIEF DOES THE OPPOSITE OF ITS COMMENT, kept as written
   because the end result is better either way. `#why .ab-fold-cta` is
   described as gaining "relaxed breathing room below the card", but its
   current margin-top resolves to 39px at 440x871 (from §88's
   `clamp(32px, 4.5vh, 48px)`), so setting 28px TIGHTENS the gap above the
   button by 11px. The room the brief is after arrives from the other side:
   pulling the cluster up lifts the CTA's clearance below from 162px to 183px
   at 440x871, and 19px to 34px at 390x760. Recorded so the next person does
   not read 28 as a loosening. */

@media (max-width: 599px) {

  /* Fold 2 — one forward affordance, not two. */
  .gsh.ab .fc-flip-all { display: none !important; }
  .gsh.ab .fc-ctrls { justify-content: flex-end; margin: 6px 0 14px; }
  .gsh.ab #why .ab-fold-cta { margin-top: 28px; }

  /* Fold 3 — 125px portraits and tighter captions buy the CTA its margin
     back. §94 had cut that margin to 16px to win clearance at 390x760; the
     savings here fund 32px and still leave more room than before. */
  .gsh.ab .ab-founder-img,
  .gsh.ab .ab-founder-mono { max-height: 125px; }

  .gsh.ab .ab-founder figcaption { margin-top: 8px; gap: 1px; }
  .gsh.ab .ab-founder figcaption b     { font-size: 15px; }
  .gsh.ab .ab-founder figcaption small { font-size: 13px; }

  .gsh.ab .ab-mob-only p { font-size: 15px; line-height: 1.5; }

  .gsh.ab #founders .ab-fold-cta { margin-top: 32px; }
}


/* ---- §96 cache-key remint — comment only, no rules -----------------------
   Not a design change. `6b08a187` deployed correctly to origin, but a subset
   of Hostinger CDN edges cached the PREVIOUS css bytes under the NEW stamped
   url during the propagation window, and then held them: measured 6 of 20
   fetches returning `e2f650fe` bytes from `?v=6b08a187`, with `age: 8556`
   against `age: 248` on the fresh edges. With `cache-control: max-age=604800`
   those entries had ~6.9 days left to run.

   `?v=` cannot clear that, because the stale bytes sit UNDER the new stamped
   url. A re-push cannot either: the stamp is a content hash, so identical
   content mints an identical url and matches the poisoned entries again.

   This comment changes the file's bytes and therefore its hash, which mints a
   different stamped url. A url no edge has ever seen cannot be stale — a
   never-deployed stamp was verified to MISS through to origin and return the
   correct 289,345 bytes on every attempt.

   Nothing below or above this block changed. If a future deploy ever splits
   across edges the same way, this is the lever; a CDN purge in hPanel is the
   other, and costs no deploy. */


/* ---- §97 footer sign-off, and the marquee as hardware --------------------
   Both items are sitewide, which on this repo means BOTH bundle roots. The
   footer markup move touches 15 files: 5 in static-export/ (about, index,
   playlists, quick-q, shelf) and 10 in launch-build/. Editing only one root
   is a silent no-op on the other ten pages — make-bundle.py merges the three
   roots with launch-build/ winning.

   REVERSING AN EARLIER CALL, ON PURPOSE. §61 set
   `.gsh .marquee { border-block: 0; background: transparent; }` because
   Lovable asked for "no borders, no box" off a screenshot. This brief asks
   for the opposite — a recessed carbon chassis — from the same partner. The
   old rule is left in place and out-specified rather than deleted, so the
   history stays legible and the decision stays revisitable.

   THE SPACING TRAP IN THE FEEDBACK LINE. Its markup is
   `<span>…</span><a>…</a><span>or</span><a>…</a>` with NO whitespace between
   the elements. The spacing comes entirely from `gap: .35ch .75ch` on a flex
   container. Switching it to `display: block`, as a compact sign-off would
   normally want, runs the words together — "email us athello@acquret.comor
   DM". So it stays flex-wrap, with the gap tightened and a max-width doing
   the wrapping instead. */

/* 1. The sign-off now lives under .lft-share inside .lft-brand. */
.gsh .lft-brand .lft-feedback {
  display: flex;
  flex-wrap: wrap;
  align-items: baseline;
  gap: .3ch .6ch;
  max-width: 300px;
  margin: 18px 0 0;
  padding: 0;
  border: 0;
  font: 400 11.5px/1.5 var(--gs-mono);
  letter-spacing: 0;
  text-transform: none;
  color: var(--gs-ink-3);
}
.gsh .lft-brand .lft-feedback a {
  color: var(--gs-green-active);
  text-decoration: none;
  border-bottom: 1px solid color-mix(in oklab, var(--gs-green-active) 35%, transparent);
}
.gsh .lft-brand .lft-feedback a:hover { border-bottom-color: var(--gs-green-active); }

/* 2. Marquee chassis — THE AMBER TAPE, third and locked revision.
   Pass 1 was carbon in both themes (too hard on linen). Pass 2 was
   theme-calibrated, which measured 1.06:1 against the page in light and so
   nearly reinstated the washout that started this. This pass uses the warm
   dial amber in BOTH themes: it separates from the light page at 1.93:1 and
   from the dark page at 9.37:1, with one ink colour that clears AAA on it.

   `.gsh .marquee` is (0,2,0), which beats §61's transparent rule on load
   order alone. No theme override is needed now — that is the point of the
   choice, and it is why the earlier `[data-theme="dark"]` rules are gone
   rather than left to fight it. */
.gsh .marquee {
  background: #E2A65A;
  border-block: 1px solid rgba(17, 20, 29, .18);
  padding-block: 11px;
}
.gsh .marquee-track {
  font: 600 13px/1 var(--gs-ui);
  letter-spacing: .02em;
  color: #11141D;
}
/* One ink for every phrase — the alternating colours are retired with the
   two-tone grounds that needed them. These two selectors stay, set to the
   same value, so the odd/even structure is still there to colour later
   without re-deriving which element does what. */
.gsh .marquee-track span:nth-child(odd),
.gsh .marquee-track span:nth-child(even) { color: #11141D; }

.gsh .marquee-track span::before {
  content: "//";
  color: #11141D;
  opacity: .55;
  margin-right: 2.25rem;
}

/* Measured on #E2A65A:
     phrase ink #11141D -> 8.62:1  AAA   (matches the brief exactly)
     separator at opacity .55 -> 3.21:1, i.e. BELOW AA once composited.
       It is a decorative glyph, not content, so it ships as briefed —
       .65 would give 4.13:1 and .75 would give 5.3:1 if that is wanted.
   Rail against its surroundings: 1.93:1 on the light page (#f7f3ee),
   9.37:1 on the dark page (#07080d), 2.14:1 on the white footer. */


/* ---- §98 caution tape: touch freeze, the moat, and the stamping -----------

   1. THE TAP FREEZE IS REAL AND THE BRIEF'S FIX CANNOT BE WRITTEN AS GIVEN.
   The pause rule lives in the LOCKED sheet —
   `gs-site-lab.css: .gsh .marquee:hover .marquee-track { animation-play-state:
   paused; }` — so it cannot be wrapped in a media query. Scoping has to be
   done from the other side: re-assert `running` on exactly the devices the
   brief wants excluded. `(hover: none), (pointer: coarse)` is the complement
   of `(hover: hover) and (pointer: fine)` — comma is OR in a media query —
   and the selector is matched exactly so it wins on load order alone.

   2. THE BRIEF'S MOAT FIX WOULD HAVE BEEN HALF A NO-OP. Two things build the
   moat and they need different weights:

     padding  `.gsh .mq-shelf.mq-end { padding-block: clamp(20px,3vh,32px)
              !important }`  — ours, (0,3,0), measured 27px top AND bottom
     ground   `.gsh[data-lab] .mq-shelf.mq-end { background: var(--fold-a)
              !important }`  — LOCKED, (0,4,0), !important

   The briefed selector is `.gsh .mq-shelf.mq-end` at (0,3,0). Against the
   padding that is enough; against the locked background it loses on
   specificity even with `!important`, so the ground would have stayed. The
   rule below leads with the (0,4,0) form so both actually land.

   (`--fold-a` is #f7f3ee light / #07080d dark — so the "beige" is the page
   ground showing through the shelf's own padding, not a border.)

   3. STAMPING. `var(--gs-head)` is indeed "Helvetica Neue" — the brief is
   right. The separator keeps its own mono face at 700 so the slashes stay
   thick; it would otherwise inherit the Helvetica and thin out.

   Symmetric spacing: the gap is split between two owners. `gap: 2.5rem` on
   the locked track sits to the LEFT of each `///`, and the `::before`
   margin-right sits to its RIGHT. Both go to 1.5rem, which is what makes the
   spacing even — changing only one would have moved the asymmetry rather
   than removed it. */

/* 1. Hover-pause for pointer devices only. */
@media (hover: none), (pointer: coarse) {
  .gsh .marquee:hover .marquee-track { animation-play-state: running; }
}

/* 2. Edge-to-edge tape: no shelf padding, no page ground around it. */
.gsh[data-lab] .mq-shelf.mq-end,
.gsh .mq-shelf.mq-end {
  padding: 0 !important;
  margin: 0 !important;
  background: transparent !important;
}

/* 3. Barricade-tape stamping. */
.gsh .marquee {
  border-block: 1.5px solid rgba(17, 20, 29, .28);
}
.gsh .marquee-track {
  font: 700 12.5px/1 var(--gs-head);
  text-transform: uppercase;
  letter-spacing: .05em;
  gap: 1.5rem;
}
.gsh .marquee-track span::before {
  content: "///";
  font-family: var(--gs-mono);
  font-weight: 700;
  color: #11141D;
  opacity: 1;
  margin-right: 1.5rem;
}


/* ---- §99 Option A: the hazard-edge tape -----------------------------------
   Supersedes §98's marquee chassis, typography and separator. The §98 touch
   and moat fixes are NOT repeated here — they are already in place and
   unchanged (`@media (hover: none), (pointer: coarse)` and the
   `.mq-shelf.mq-end` reset); re-declaring them would just be a second copy
   to keep in sync.

   `.marquee::before` / `::after` were free — nothing in either sheet used
   them, so the stripes claim them without displacing anything. The marquee
   already carries `overflow: hidden`, so the stripes clip to the rail.

   WHAT THE NUMBERS DO. Ink `#14161F` on `#EDB04A` is **9.37:1**, up from
   8.62 on the old amber — still AAA, slightly better. The rail separates
   from the dark page at 10.39:1 but only **1.74:1** from the light page,
   down from 1.93: the lighter amber sits closer to the cream. That is the
   stripes' job now — they are what defines the edge in light mode, not the
   ground contrast, which is why the 1.5px hairlines could go.

   Height moves 38px -> **43px** (13px line + 15px padding x2, no border).
   The brief set no target; recording it because the last three revisions all
   specified one.

   ON THE `!important` FLOOD, implemented as briefed but worth stating: of the
   ~20 `!important`s here, only the ones on `.mq-shelf.mq-end` (which fight a
   locked `(0,4,0) !important`) are actually needed. Everything else wins on
   load order alone, the way every other rule in this file does. The cost is
   that the next marquee tweak will also have to carry `!important`. Happy to
   strip them to plain declarations in a follow-up — same rendering, measured.

   SEPARATOR ARCHAEOLOGY. `.gsh .marquee-track span::before` now has four
   generations stacked in this file: `✦`, `·`, `//`, `///`, and this chip.
   Only the last applies. CSS is frozen pre-launch so they stay; worth one
   consolidation pass after launch. */

/* Chassis: solid amber, stripes instead of hairlines. */
.gsh .marquee {
  position: relative !important;
  overflow: hidden !important;
  background: #EDB04A !important;
  padding-block: 15px !important;
  border-block: none !important;
}

/* 5px continuous black-and-amber diagonal hazard stripes, top and bottom. */
.gsh .marquee::before,
.gsh .marquee::after {
  content: "" !important;
  position: absolute !important;
  left: 0 !important;
  right: 0 !important;
  height: 5px !important;
  background: repeating-linear-gradient(-45deg, #14161F 0 7px, #EDB04A 7px 14px) !important;
  z-index: 2 !important;
  pointer-events: none !important;
}
.gsh .marquee::before { top: 0 !important; }
.gsh .marquee::after  { bottom: 0 !important; }

/* Back to the site's own UI face; the Helvetica stamping of §98 is retired. */
.gsh .marquee-track {
  font: 700 13px/1 var(--gs-ui) !important;
  text-transform: uppercase !important;
  letter-spacing: .06em !important;
  color: #14161F !important;
  gap: 1.5rem !important;
}
.gsh .marquee-track span {
  display: inline-flex !important;
  align-items: center !important;
}

/* The typed slashes become a drawn hazard chip. `vertical-align` is inert
   inside the inline-flex span above — kept as briefed, harmless. */
.gsh .marquee-track span::before {
  content: "" !important;
  display: inline-block !important;
  width: 24px !important;
  height: 11px !important;
  border-radius: 2px !important;
  background: repeating-linear-gradient(-45deg, #14161F 0 4px, transparent 4px 8px) !important;
  margin-right: 1.5rem !important;
  vertical-align: middle !important;
}


/* ---- §100 refined hazard rail, and the Act 06 clip ------------------------
   Refines §99. Three of its rules have to be actively UNDONE rather than
   just out-specified, because §99 set them with `!important`:
   the two stripe pseudo-elements (`content: none`) and the chip's box
   (width/height/background/radius/display reset before the glyph goes in).

   STRIPES MOVE FROM PSEUDO-ELEMENTS TO BORDER-IMAGE, and that changes the
   height. §99's stripes were absolutely positioned — they overlaid the rail
   and added nothing. Borders occupy space, so at the same 15px padding the
   rail goes 43px -> 49px even though the stripes themselves are slimmer
   (5px -> 3px). If the intent was a slimmer tape overall, padding wants to
   drop to 12px, which lands it back at 43px. Left at 15px as briefed and
   reported instead of silently changed.

   The separator becomes a glyph again, so it inherits the track's Space
   Grotesk rather than carrying its own face — correct here, since a diamond
   is not a slash and does not need the mono weight.

   Guardrails all still in place and untouched: the `(hover: none), (pointer:
   coarse)` scope, the `.mq-shelf.mq-end` zero-moat reset, and the locked
   sheet's blanket `prefers-reduced-motion` rule. None are re-declared. */

/* 1. Retire §99's overlay stripes. */
.gsh .marquee::before,
.gsh .marquee::after { content: none !important; }

/* 2. Slim 3px hazard edges as real borders. */
.gsh .marquee {
  border-top: 3px solid transparent !important;
  border-bottom: 3px solid transparent !important;
  border-image: repeating-linear-gradient(
    -45deg,
    #0E1018 0px,
    #0E1018 6px,
    #EDB04A 6px,
    #EDB04A 12px
  ) 3 !important;
}

/* 3. Jet black ink, both themes. Measured 9.86:1 on #EDB04A — AAA, though
   not the 11.3:1 the brief states. */
.gsh .marquee-track { color: #0E1018 !important; }

/* 4. One clean diamond instead of the striped chip. The chip's box is reset
   first; §99 set each of these with `!important`. */
.gsh .marquee-track span::before {
  content: "\25C6" !important;
  display: inline !important;
  width: auto !important;
  height: auto !important;
  border-radius: 0 !important;
  background: none !important;
  color: #0E1018 !important;
  font-size: .72em !important;
  line-height: 1 !important;
  margin-right: 1.5rem !important;
}


/* ---- §100b Act 06 bench clips its own text at 390px ----------------------
   From the 19 Nov sweep. The band cut 43px off the right edge — the kicker
   rendered "...CO-BUILD & DEPI" and the headline's second line was sliced.
   Two causes, both in the locked sheet: `.lab-close-say` is a grid whose
   track computes to 368px inside its own 300px box (constraining the
   children does nothing — `max-width: 100%` was tested and changed nothing),
   and `.gsh h2 em.hl-fire` carries `white-space: nowrap`.

   Scoped deliberately to the bench: all seven other `h1/h2 em.hl-*` on the
   homepage fit with room to spare, so the global highlight treatment is left
   alone. Pre-existing, not a regression — the 18 Nov build `65115b43` shows
   the identical 52px overflow. */
@media (max-width: 760px) {
  .gsh[data-lab] .lab-close-say { grid-template-columns: minmax(0, 1fr); }
  .gsh .lab-close-bench .hl { white-space: normal; }
}


/* ---- §101 consolidation: tape height, tablet overflow, two contrast fixes -
   All approved together. The Act 06 bench fix is already in place from §100b
   and is NOT repeated here.

   1. TAPE HEIGHT. §99 set `padding-block: 15px !important`, so this needs
   `!important` to land. 12px + the 3px borders gives 13 + 24 + 6 = **43px**,
   back to the pre-border height with the slim edges kept.

   2. TABLET OVERFLOW — and the band it actually lives in. `.ac-consoles` has
   two existing rules: `1fr 1fr` at top level, and a swipe carousel
   (`display: flex; overflow-x: auto; scroll-snap`) at `width <= 719px`. So
   the overflow band is **720-899px**, not all of tablet.

   The cause is the `1fr` trap: `1fr` means `minmax(auto, 1fr)`, and `auto`
   floors the track at min-content. A console's min-content is ~407px, so the
   pair measured 406.992 + 348.016 + 16px gap = 771px inside a 728px
   container and simply could not shrink. `minmax(0, 1fr)` removes that floor
   — which is why this is the fix rather than any padding change.

   Scoped at `max-width: 899px` as approved. Below 720 the container is
   `display: flex`, where `grid-template-columns` is inert, so the swipe
   carousel is untouched.

   3. THE TWO CONTRAST CASES, and why the selectors are heavier than they
   look. Neither element is styled from these stylesheets at all — both come
   from a page-level `<style>` block:

     404.html      `.nf-code        { color: var(--gs-ink-3) }`   (0,1,0)
     privacy.html  `.legal .updated { color: var(--gs-ink-3) }`   (0,2,0)

   A page `<style>` in the head loads after our external sheet, so an
   equal-specificity rule here would lose on order. `.gsh` is prepended to
   out-rank both regardless of order.

   `--gs-ink-2` rather than a literal: it already exists in both themes and
   tracks the palette. Measured on the page ground — light `#5d5348`
   **6.80:1**, dark `#9aa2b8` **7.85:1**, against the 3.72/4.23 that
   `--gs-ink-3` gives there. The global `--gs-ink-3` token is deliberately
   left untouched; it belongs to the shared brand system with acquret.com as
   parent, and moving it would ripple across both sites to fix two
   instances. */

.gsh .marquee { padding-block: 12px !important; }

@media (max-width: 899px) {
  .gsh .ac-consoles { grid-template-columns: minmax(0, 1fr); }
}

/* NOT `.gsh ...` — both elements sit OUTSIDE the `.gsh` wrapper. Measured:
   `.updated`'s chain is `.updated > .wrap > .legal > BODY`, and the `.gsh`
   element on that page is only the header/footer chrome. A `.gsh`-prefixed
   rule matched nothing and the colour never changed — caught by measuring
   the rendered page, not by reading the diff. `body` gives the one extra
   element needed to out-rank the page's own inline `<style>`, which loads
   after this sheet and so wins any equal-specificity tie. */
body .nf-code,
body .legal .updated { color: var(--gs-ink-2); }


/* ---- §105 the rack ports to homepage Act 03 (#tools) ----------------------
   §102, §103 and §104 are widened from `.gsh[data-lab].tk-toolwrap` to
   `:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools)` so /tools and the
   homepage toolkit section share one rack treatment. The homepage renders the
   SAME `.tl-card` / `.tl-screen` markup, which is why §102 was scoped so
   tightly in the first place — that scoping is now relaxed on purpose, and
   `#tools` is safe as the hook because it is the homepage anchor section and
   exists nowhere else.

   SPECIFICITY CHANGES, AND ALL THREE SECTIONS HAD TO MOVE TOGETHER. `:is()`
   takes the specificity of its most specific argument, so these selectors go
   from (0,4,0) to (1,3,0) — the id outranks any number of classes. That is
   still fine against the locked sheet, but it means §103 can only keep
   superseding §102 if BOTH are widened; widening one alone would have
   inverted the supersede order and silently handed the screen ground back to
   §102. Same for §104 against §102's replaced block.

   TWO THINGS ARE DELIBERATELY NOT PORTED, both measured on the homepage:

   1. `.tl-screen { height: 124px }` stays /tools-only. The homepage cards
      carry more text in the same box, so the fixed height CLIPS: measured
      card 1 / card 2 content cut by 9px / 22px at 375 and 390, and 9px / 9px
      at 720. It stops clipping at 880 and above. The homepage keeps its
      natural screen heights (136 desktop, 146-159 phone), so this port
      changes the screen's COLOUR and CHROME there, not its geometry.
      §104's `height: auto; min-height: 112px` relief stays /tools-only for
      the same reason — there is no fixed height on the homepage to relieve.

   2. `.tl-body h3` keeps its own size on each page. The brief lists the rack,
      the screen colours, the pips, the meters and the hover states; type is
      not in it, and porting §102's clamp plus §103's 26px would have taken
      the homepage heading from 27px to 26px on desktop and 22px to 26px on
      phones. Left alone; it is a one-line change if parity is wanted.

   What DOES port: the rack strip and pips, the card chassis (flex column,
   18px radius, 14px padding, raised ground, dark-mode variant), the screen
   ground and its full ink set, `.tl-eng`, the split/score meters, the live
   dot, the tags, the CTA pill and the carousel behaviour. */

/* ---- §102 /tools refresh — the rack, scoped --------------------------------
   EVERY rule here is prefixed `:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools)`. That is not tidiness: the
   homepage renders THE SAME TWO CARDS with the same `.tl-card` / `.tl-screen`
   / `.tl-grid` classes in its `.tl-sec` toolkit section, so an unscoped rule
   restyles the homepage too. `.tk-toolwrap` exists only on /tools, needs no
   new markup, and `#tools` is NOT usable as the hook — that id is a homepage
   anchor section.

   Most of the locked `.tl-*` geometry lives in gs-site-lab.css (10/30/19
   refs). Nothing there is edited; these out-specify at (0,3,0)+ and load
   later.

   THE CAROUSEL UNWIND IS THREE PROPERTIES, NOT ONE. Below 760 the locked
   sheet turns the grid into a scroll-snap rail — `overflow-x: auto`,
   `scroll-snap-type: x mandatory`, plus `grid-auto-flow: column` feeding
   `grid-auto-columns`. Setting `grid-template-columns: 1fr` alone leaves a
   row of auto-columns still scrolling sideways, so all four are reset.
   Measured before: at 390 the grid reported `301px 301px` with
   `overflow-x: auto`.

   SPECIFICITY CORRECTION, measured not assumed: the first pass scoped these
   at `.gsh.tk-toolwrap` (0,3,0) and the 2-col layout still won at 880px.
   The locked sheet carries `.gsh[data-lab] .tl-sec .tl-grid` at **(0,4,0)**
   inside `@media (width >= 720px) and (width <= 999px)`, which outranked it.
   Prefixing `[data-lab]` brings these to (0,4,0) and they win on load order,
   since gs-static-fix.css loads after gs-site-lab.css on this page. */

/* ---- hero -------------------------------------------------------------- */
.gsh[data-lab].tk-toolwrap .tk-lockpill,
.tk-hero .tk-lockpill {
  display: inline-flex; align-items: center; gap: .5ch;
  margin: 0 auto; padding: 7px 14px;
  border: 1px solid var(--gs-rule); border-radius: 999px;
  background: var(--gs-surface);
  font: 500 10.5px/1 var(--gs-mono);
  letter-spacing: .14em; text-transform: uppercase;
  color: var(--gs-ink-2);
}
.tk-hero .tk-lockpill i { color: var(--gs-green-active); font-style: normal; }
.tk-hero .tk-lockpill svg { flex: none; }

/* ---- the rack ---------------------------------------------------------- */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-grid { align-items: stretch; }

:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-card {
  display: flex; flex-direction: column; height: 100%;
}

/* Recessed display: dark ground in both themes, crisp 1px edge.

   THE FIXED HEIGHT IS /tools ONLY — see §105. On the homepage the same
   markup carries more text, so 124px clips it. */
.gsh[data-lab].tk-toolwrap .tl-screen { height: 124px; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen {
  box-sizing: border-box;
  /* `!important` is required, not stylistic: the locked sheet carries
     `.gsh .tl-sec .tl-screen { background: linear-gradient(...) !important }`,
     which no specificity can beat. Height and border applied fine without it,
     which is exactly why the rule looked live while the ground stayed cream
     in light and transparent in dark — caught by reading the computed value,
     not the rule list. */
  background: var(--v2-screen) !important;
  border: 1px solid var(--v2-bezel);
  border-radius: 10px;
  padding: 12px 14px;
  display: flex; flex-direction: column; justify-content: space-between;
}
/* THE SCREEN IS THEME-IMMUNE, SO ITS INK MUST BE TOO. Forcing a dark ground
   in light mode broke every ink the locked sheet had tuned for its original
   light/gradient screen: measured in LIGHT, the channel label fell to
   2.76:1 and the legend row to 2.40:1, while the same text read 8.05 and
   8.60 in dark. Several of those locked rules carry `!important`, so this
   blanket reset needs it too. `--v2-screen-ink` is the light ink the
   telemetry bay already uses on exactly this ground. */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen * { color: var(--v2-screen-ink) !important; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen .mono,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen small,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-legend,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-legend * {
  color: color-mix(in oklab, var(--v2-screen-ink) 78%, transparent) !important;
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen b { color: var(--v2-screen-ink) !important; }
/* The screen is a fixed dark ground in BOTH themes, so its ink must be too.
   `var(--v2-glow)` flips to a dark green for light backdrops and measured
   2.76:1 here against the dark screen — below AA. Pinned to the dark-mode
   glow, which is the value designed for this ground: 8.05:1 both themes. */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-eng { color: #5FD08A !important; letter-spacing: .12em; }

/* Body: Space Grotesk head with the Caslon accent, standardised tags. */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-body { display: flex; flex-direction: column; flex: 1 1 auto; }
.gsh[data-lab].tk-toolwrap .tl-body h3 {
  font: 700 clamp(19px, 1.7vw, 22px)/1.2 var(--gs-head);
  letter-spacing: -.01em; color: var(--gs-ink); margin: .55rem 0 .35rem;
}
.gsh[data-lab].tk-toolwrap .tl-body h3 em {
  font-family: var(--gs-serif); font-style: italic; font-weight: 400;
  color: var(--gs-ink);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-tags {
  display: inline-flex; flex-wrap: wrap; align-items: center; gap: 6px;
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-tags b,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-tags span {
  font: 500 10px/1 var(--gs-mono); letter-spacing: .12em; text-transform: uppercase;
  padding: 5px 9px; border-radius: 999px;
  border: 1px solid var(--gs-rule); color: var(--gs-ink-2);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-tags b { border-color: transparent; background: var(--gs-surface-2); color: var(--gs-ink); }

/* CTA anchors the card floor, full-width gradient pill. */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-go {
  margin-top: auto; align-self: stretch;
  display: inline-flex; align-items: center; justify-content: center;
  padding: 12px 18px; border-radius: 999px;
  background: linear-gradient(100deg, var(--gs-indigo), var(--gs-green-active));
  color: #fff; font: 600 14px/1 var(--gs-ui);
}

/* ---- §104 mobile carousel restored (<760), 2-col kept at tablet ----------
   Replaces §102's "below 900 unwound" block. §102 had unwound the locked
   rail into a single column at every width under 900; Kyle wants the swipe
   back on phones, with the tablet two-column row preserved.

   759, not 760, for the carousel edge — the same boundary §49 and §102 use.
   The locked sheet switches `.gsh[data-lab] .tl-sec .tl-grid` to a 2-col row
   grid inside `(width >= 720px) and (width <= 999px)`, so a `max-width: 760`
   rule would make 760 the one width where both sides claim the grid.

   THE 720-759 OVERLAP IS DELIBERATE AND NEEDS THE RESETS BELOW. Both queries
   match between 720 and 759. The tablet block wins the three properties it
   restates, but the carousel's OTHER declarations are `!important` and
   nothing in the tablet block turns them off — in particular the negative
   `margin-inline`, which on a non-scrolling row grid pushes the track
   `--gs-pad` past its container on each side. Measured at 759 with only the
   brief's three properties in the tablet block: the grid goes 727px -> 767px
   and sits from x=-4 to x=763, overhanging BOTH viewport edges, and document
   scrollWidth goes 759 -> 763 — 4px of sideways page scroll. With the resets
   below: grid 727px at x=16..743, document scrollWidth 759, overflow 0.
   The tablet block therefore restores margin, padding and the snap
   properties too.

   `.tl-screen`'s mobile height relaxation is KEPT, re-scoped from 899 to 759.
   It is not part of the carousel unwind: it exists so the screen can grow
   past the desktop 124px when the card narrows, and the carousel narrows it
   further than the old single column did — the card is now 86% of the track,
   so the screen needs 161px at 375 and 151px at 390 (it was 126px at 390
   under the one-column block). Dropping this rule with the rest of the block
   would have pinned those to 124px: measured 21px of content cut off at 375
   and 10px at 390. At 440 and up the natural height is back to 126 and the
   pin would not have clipped, which is why this only shows on small phones.
   At tablet the fixed 124px is wanted, which is where the 759 edge puts it. */
@media (max-width: 759px) {
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-grid {
    display: grid !important;
    grid-auto-flow: column !important;
    grid-template-columns: none !important;
    grid-auto-columns: 86% !important;
    gap: 12px !important;
    overflow-x: auto !important;
    scroll-snap-type: x mandatory !important;
    overscroll-behavior-x: contain !important;
    scrollbar-width: none !important;
    -webkit-overflow-scrolling: touch !important;
    margin-inline: calc(var(--gs-pad, 20px) * -1) !important;
    padding: 4px var(--gs-pad, 20px) 22px !important;
    scroll-padding-inline: var(--gs-pad, 20px) !important;
  }
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-grid::-webkit-scrollbar { display: none !important; }
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-card {
    scroll-snap-align: start !important;
    min-width: 0 !important;
    width: auto !important;
  }
  .gsh[data-lab].tk-toolwrap .tl-screen { height: auto; min-height: 112px; }
}
@media (min-width: 720px) and (max-width: 899px) {
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-grid {
    grid-template-columns: repeat(2, minmax(0, 1fr)) !important;
    grid-auto-flow: row !important;
    overflow-x: visible !important;
    /* undo the carousel declarations the three above do not cover */
    margin-inline: 0 !important;
    padding: 0 !important;
    scroll-snap-type: none !important;
    scroll-padding-inline: 0 !important;
    grid-auto-columns: auto !important;
  }
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-card { scroll-snap-align: none !important; }
}


/* ---- §102b the hero wave, which does not inherit ---------------------------
   The brief assumed `<em class="hl hl-fire">` would pick up the terracotta
   wave for free. On THIS page it does not. The locked rule is
   `.gsh[data-lab].gsh h1 em.hl-fire::after`, and the /tools hero sits
   OUTSIDE every `.gsh[data-lab]` wrapper — its chain is
   `em.hl-fire > .tk-h1 > div > section.hero.tk-hero > BODY`. Verified with
   `el.matches()`: **false**, and `::after` computed to `content: none`, so
   the wave simply did not draw.

   Two ways out. Adding `.gsh` + `data-lab` to the hero section would import
   the whole `.gsh` base cascade into a block that is currently
   inline-styled — too broad a change to make for one underline. So the
   declarations are re-stated here against `.tk-hero`, copied verbatim from
   the locked sheet (light and dark) so the artwork stays identical and keeps
   tracking any future change to the shared `h2` variants.

   `white-space: nowrap` is inherited from the locked rule and is safe here:
   it wraps the single word "message". */

.tk-hero h1 em.hl-fire { white-space: nowrap; position: relative; }
.tk-hero h1 em.hl-fire::after { content: ""; opacity: 1; pointer-events: none; background: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 200 14' preserveAspectRatio='none'%3E%3Cpath d='M2 9 C 40 3, 90 12, 130 6 S 185 4, 198 8' fill='none' stroke='%23C24A3A' stroke-width='3' stroke-linecap='round'/%3E%3C/svg%3E") center center / 100% 100% no-repeat; height: 0.22em; position: absolute; bottom: -0.14em; left: -2%; right: -2%; }
.gsh[data-theme="dark"] ~ * .tk-hero h1 em.hl-fire::after,
[data-theme="dark"] .tk-hero h1 em.hl-fire::after { background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 200 14' preserveAspectRatio='none'%3E%3Cpath d='M2 9 C 40 3, 90 12, 130 6 S 185 4, 198 8' fill='none' stroke='%23E05D52' stroke-width='3' stroke-linecap='round'/%3E%3C/svg%3E"); }


/* ---- §103 /tools rack chassis v2 ------------------------------------------
   Appended as supplied, strictly scoped. It deliberately supersedes parts of
   §102 rather than replacing them — later, same-or-higher specificity:

     - `.tl-screen` ground: §102's `var(--v2-screen)` is replaced by the
       warm-brown display below. Note the INVERSION — light theme gets the
       dark brown #292621, dark theme gets the warm off-white #F2ECE4. §102's
       `height: 124px` is NOT restated, so the pixel-matched screens and the
       desktop baseline alignment still come from §102.
     - the screen ink resets: §102's `--v2-screen-ink` version is superseded
       by the `--tl-display-*` set, which is the right call now the ground
       itself flips per theme.
     - `.tl-body h3` font-size overrides §102's clamp.

   Contrast measured on the new grounds before building, all passing:
     light #292621  — ink 12.84, muted 8.39, accent 7.80, indigo 5.54
     dark  #F2ECE4  — ink 14.26, muted 6.40, accent 5.58, indigo 6.29

   `background: ... !important` on `.tl-screen` is required, not stylistic:
   the locked sheet carries
   `.gsh .tl-sec .tl-screen { background: linear-gradient(...) !important }`.

   `.tk-lockpill` from §102 now matches nothing — the privacy badge reverted
   to the plain line. Left in place; CSS is frozen pre-launch. */

/* ===== /tools rack chassis v2 (scoped) ===== */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-card {
  border-radius: 18px;
  padding: 14px 14px 20px;
  background: var(--gs-raised);
  border: 1px solid var(--gs-rule);
  box-shadow: inset 0 1px 0 rgba(255,255,255,.8), 0 1px 0 var(--gs-rule), 0 18px 36px -26px rgba(26,29,43,.45);
}
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap, .gsh[data-lab][data-theme="dark"] #tools) .tl-card {
  background: #12141D;
  border-color: rgba(255,255,255,.08);
  box-shadow: inset 0 1px 0 rgba(255,255,255,.05), 0 18px 36px -24px rgba(0,0,0,.8);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-no { display: none; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-rack {
  display: flex;
  justify-content: space-between;
  align-items: center;
  padding: 4px 6px 12px;
  margin: 0 0 12px;
  border-bottom: 1px dashed var(--gs-rule);
  font-size: 11px;
  letter-spacing: .14em;
  text-transform: uppercase;
  color: var(--gs-ink-2);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-rack i { font-style: normal; color: var(--gs-green-active); }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-st {
  display: inline-flex;
  align-items: center;
  gap: 7px;
  font-size: 10px;
  font-weight: 600;
  color: var(--gs-ink);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .pip { width: 7px; height: 7px; border-radius: 50%; display: inline-block; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .pip.g { background: #1B6B38; box-shadow: 0 0 0 3px rgba(27,107,56,.15); }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .pip.a { background: #C27803; box-shadow: 0 0 0 3px rgba(194,120,3,.15); }
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap, .gsh[data-lab][data-theme="dark"] #tools) .pip.g { background: #35BD6B; box-shadow: 0 0 8px #35BD6B; }
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap, .gsh[data-lab][data-theme="dark"] #tools) .pip.a { background: #EDB04A; box-shadow: 0 0 8px #EDB04A; }

/* Display screen: Playlists warm brown (#292621) in light; warm off-white (#F2ECE4) in dark. */
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen {
  --tl-display: #292621;
  --tl-display-ink: #F2ECE4;
  --tl-display-muted: #C9C0B5;
  --tl-display-accent: #5FD08A;
  --tl-display-track: #514B43;
  --tl-display-indigo: #8498EF;
  background: var(--tl-display) !important;
  border-color: var(--tl-display);
  border-radius: 10px;
  box-shadow: inset 0 2px 10px rgba(0,0,0,.28), 0 0 0 4px var(--gs-surface), 0 0 0 5px var(--gs-rule);
}
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap, .gsh[data-lab][data-theme="dark"] #tools) .tl-screen {
  --tl-display: #F2ECE4;
  --tl-display-ink: #1A1D2B;
  --tl-display-muted: #5D5348;
  --tl-display-accent: #1B6B38;
  --tl-display-track: #D8CEC2;
  --tl-display-indigo: #485289;
  border-color: #D8CEC2;
  box-shadow: inset 0 2px 8px rgba(41,38,33,.08), 0 0 0 4px var(--gs-surface), 0 0 0 5px var(--gs-rule);
}
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen * { color: var(--tl-display-ink) !important; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen .mono,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen small,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-legend,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-legend * { color: var(--tl-display-muted) !important; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-screen b { color: var(--tl-display-ink) !important; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-eng { color: var(--tl-display-accent) !important; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-split,
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-score { background: var(--tl-display-track); }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-split i:first-child { background: var(--tl-display-accent); box-shadow: none; }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-split i:last-child { background: var(--tl-display-indigo); }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-score i { background: linear-gradient(90deg, var(--tl-display-indigo), var(--tl-display-accent)); }
:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-live { background: var(--tl-display-accent); box-shadow: 0 0 8px var(--tl-display-accent); }

:is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-tags b { background: var(--gs-ink); color: var(--gs-ground); border-color: var(--gs-ink); }
.gsh[data-lab].tk-toolwrap .tl-body h3 { font-size: 26px; }
@media (prefers-reduced-motion: reduce) {
  :is(.gsh[data-lab].tk-toolwrap, .gsh[data-lab] #tools) .tl-card { transition: none; }
}

/* ---- §106 homepage keeps its original card footprint ----------------------
   The §105 port carried /tools' outer padding and 18px radius onto the
   homepage card. That is a bigger box than the one Act 03 was laid out
   around, and it pushed the section's forward CTA down the page. Measured
   against the live build, candidate minus baseline:

     375   card 521 -> 589 (+68),  "Read the room" +77px down the page
     390   card 521 -> 578 (+57),  +67px
     768   card 458 -> 501 (+43),  +44px
     1280  card 435 -> 481 (+46),  +46px
     1440  card 435 -> 481 (+46),  +46px

   So the padding and radius go back to /tools-only and the homepage keeps
   `padding: 0` / `border-radius: 22px`. The rack strip then has to supply its
   own inset, because with no card padding it would otherwise sit flush to the
   card edge — `12px 16px 10px` with no bottom margin, which is close to the
   41px `.tl-chrome` strip it replaces.

   NOTE the guardrail figures in the brief, ~513 mobile / ~420 desktop, are
   not the live originals: measured on the deployed build they are 521 at
   375/390 and 435 at 1280/1440. Restoring the original footprint is the
   actual goal, so those measured numbers are the target. */
.gsh[data-lab] #tools .tl-card {
  padding: 0;
  border-radius: 22px;
}
.gsh[data-lab] #tools .tl-rack {
  padding: 12px 16px 10px;
  margin: 0;
  border-bottom: 1px dashed var(--gs-rule);
}

/* ---- §108 Homepage Fold 1: clean centered title (dots & rotate retired) ---
   RENUMBERED. The brief called this §107, but §107 is already taken in this
   sheet by the /about tab-animation revert earlier today. Numbers are how we
   refer to these in notes, so the duplicate would have been worse than the
   renumber: this is the brief's §107 and the desk block below is its §108. */
.gsh .hero .stage-card .stage-bar.lab-titlebar {
  display: flex !important;
  justify-content: center !important;
  align-items: center !important;
  padding: .5rem 0 !important;
  margin-bottom: .55rem !important;
}
.gsh .hero .stage-card .lab-titlebar .lab-seats-title {
  text-align: center !important;
  white-space: nowrap !important;
  font-size: clamp(12.5px, 2.2vw, 13.5px) !important;
  line-height: 1.2 !important;
}

/* ---- §109 Homepage Act 02: Sanity Desk top tabs & ticket contrast --------
   (the brief's §108.)

   `.dk-bar` is removed from the markup, so this `display: none` matches
   nothing. Kept as supplied — it keeps the bar hidden if that markup ever
   returns — and noted for the post-launch prune list.

   THE SEATS RADIUS IS BREAKPOINT-DEPENDENT. The brief specifies 21px, which
   is right against the 22px `.dk` chassis, but the locked sheet drops the
   chassis to 18px at `width <= 719px` (gs-site-lab.css:2047), where the
   matching inner radius is 17px. `.dk` is `overflow: hidden`, so 21px would
   be clipped rather than visibly wrong, but it would fight the clip on the
   corner pixels. Both values are set, each at its own width. */
.gsh .dk-bar { display: none !important; }

/* Persona tab container anchors top with matching chassis radius */
.gsh .dk-seats {
  background: var(--gs-surface);
  border-top-left-radius: 21px;
  border-top-right-radius: 21px;
}
@media (width <= 719px) {
  .gsh .dk-seats {
    border-top-left-radius: 17px;
    border-top-right-radius: 17px;
  }
}

/* Dispatch tickets: crisp raised white cards against the surface chassis.

   TWO CHANGES FROM THE BRIEF, BOTH TO PROTECT THE TICKET STATES.

   The brief used `!important` on background and border. Measured with it in
   place: a `.sel` ticket, a `.done` ticket and a `.done.sel` ticket all
   computed to EXACTLY the resting values — rgb(255,255,255) on
   rgb(232,225,216) — so the selected tint, the selected ink border and the
   whole done treatment were flattened. The locked states are
   `.gsh .dk-fire.sel` (tint `color-mix(--pc 5%, --gs-raised)` plus
   `border-color: var(--gs-ink)`, gs-site-lab.css:3160) and
   `.gsh .dk-fire.done` (`background: var(--gs-surface)`, :2016), both at
   (0,3,0) with no `!important`, so any `!important` here beats them. The
   brief says not to touch the ticket selection logic; the logic was intact
   but its only visual feedback was gone.

   1. `!important` dropped. It was never needed for the resting ground: the
      base locked rule `.gsh .dk-fire` is also (0,2,0) and this sheet loads
      later, so load order already wins it.
   2. `:not(.sel):not(.done)` added. Without it the dark rule still breaks
      the states — `.gsh[data-theme="dark"] .dk-fire` is (0,3,0), the same as
      `.gsh .dk-fire.sel`, and ours loads later. There are no dark-scoped
      ticket rules in the locked sheet at all, so dark relies on those same
      light-defined state selectors. Excluding the states is what makes this
      safe in both themes. */
.gsh .dk-fire:not(.sel):not(.done) {
  background: #FFFFFF;
  border: 1px solid var(--gs-rule);
  box-shadow: 0 1px 3px rgba(26, 29, 43, 0.04);
}
.gsh[data-theme="dark"] .dk-fire:not(.sel):not(.done) {
  background: #141722;
  border-color: rgba(255, 255, 255, 0.08);
  box-shadow: none;
}

/* ---- §111 Marquee: Fixed Dark Console Ribbon -----------------------------
   Third generation, replacing the Operator Telemetry Ticker shipped earlier
   today in `bac4e151` (which itself superseded §99/§100's hazard rail). The
   ribbon is now a dark console in BOTH themes, matching the telemetry
   screens in the tool cards, so it never sits muddy against a light ground.

   REPLACED IN PLACE rather than stacked as another supersede layer. That is
   a departure from how §99 -> §100 -> §111 were handled, and it is safe here
   for one specific reason: this block is mine from this afternoon, it is the
   last marquee block in the sheet, and nothing between it and EOF touches
   the marquee (§112 is the tools CTA only). Stacking a fourth generation
   would have meant four live marquee chassis rules fighting over the same
   properties. §99/§100/§101's hazard declarations are still above and still
   neutralised by the `!important` declarations here, exactly as before.

   PADDING IS 14.5px, NOT THE BRIEFED 14px, AND THAT IS THE WHOLE POINT OF
   THE 43px GUARDRAIL. The brief's arithmetic is "14px padding + 1px borders
   top and bottom = 43px exact", which omits the line box. The real sum:

     hazard rail (the 43px being matched)  12+12 padding + 13px line
                                           (font 13px/1) + 3+3 borders = 43
     brief as written                      14+14 + 12px line
                                           (font 12px/1) + 1+1          = 42
     shipped here                          14.5+14.5 + 12 + 1+1         = 43

   The ticker round dropped the track from 13px to 12px, which took 1px out
   of the line box, so 14px padding lands at 42 and the guardrail fails by a
   pixel. Measured 42px on all six marquee pages before this correction.
   14.5px is the only value that hits 43 exactly while keeping the briefed
   1px borders and 12px type; it is a half-pixel of padding, not a rendered
   half-pixel edge. Revert to 14px if a round number matters more than the
   exact height — that is a 42px rail and a 1px rise, instead of the 3px
   rise the ticker had.

   Still inherited from §99 and deliberately not restated — same three as
   before: `.marquee-track { gap: 1.5rem }`, `.marquee-track span
   { display: inline-flex; align-items: center }`, and `vertical-align:
   middle` on the separator. */
.gsh .marquee::before,
.gsh .marquee::after { content: none !important; }

.gsh .marquee {
  position: relative !important;
  overflow: hidden !important;
  background: #11131C !important;
  padding-block: 14.5px !important;
  border-top: 1px solid #1E2232 !important;
  border-bottom: 1px solid #1E2232 !important;
  border-image: none !important;
}
.gsh[data-theme="dark"] .marquee {
  background: #0E1018 !important;
  border-top: 1px solid rgba(255, 255, 255, 0.08) !important;
  border-bottom: 1px solid rgba(255, 255, 255, 0.08) !important;
}

.gsh .marquee-track,
.gsh .marquee-track span,
.gsh .marquee-track span:nth-child(odd),
.gsh .marquee-track span:nth-child(even),
.gsh[data-theme="dark"] .marquee-track,
.gsh[data-theme="dark"] .marquee-track span,
.gsh[data-theme="dark"] .marquee-track span:nth-child(odd),
.gsh[data-theme="dark"] .marquee-track span:nth-child(even) {
  font: 700 12px/1 var(--gs-ui) !important;
  text-transform: uppercase !important;
  letter-spacing: .08em !important;
  color: #D1D6E5 !important;
}

.gsh .marquee-track span::before,
.gsh[data-theme="dark"] .marquee-track span::before {
  content: "\25C6" !important;
  display: inline !important;
  width: auto !important;
  height: auto !important;
  border-radius: 0 !important;
  background: none !important;
  color: #35BD6B !important;
  font-size: .68em !important;
  line-height: 1 !important;
  margin-right: 1.5rem !important;
  opacity: 1 !important;
}

/* ---- §112 Tools CTA dark-mode ink consistency -----------------------------
   (the brief's §110.)

   THE BRIEF'S SELECTOR WAS CORRECTED, because as written it fixed only half
   the scope. It opened `.gsh[data-theme="dark"] :is(.gsh[data-lab]...)`,
   which needs a `.gsh` nested inside another `.gsh`. On the homepage that
   happens to work — there is one `.gsh` and it is an ancestor of `#tools`,
   so it satisfies both halves. On /tools it does not: that page has five
   `.gsh` wrappers (`gsh-chrome`, `tk-toolwrap`, `tk-closewrap`, `tk-mqwrap`,
   `gsh-chrome`) and they are SIBLINGS, so `.gsh.tk-toolwrap` has no `.gsh`
   ancestor. Measured: the brief's selector matched 0 elements on /tools and
   2 on the homepage; the compound form below matches 2 on both.

   WHY THIS IS WORTH SHIPPING, measured on the live dark gradient
   `linear-gradient(100deg, oklch(0.7 0.13 273), oklch(0.78 0.16 150))`
   = rgb(132,152,239) -> rgb(95,211,127):

     white (current live)  2.72:1 indigo end, 1.89:1 green end  — fails AA
     #0E1018 (briefed)     6.98:1              10.05:1
     #07080D               7.35:1              10.59:1

   So the current dark CTA is a real contrast failure, not a polish item.

   TARGET VALUE RESOLVED. This first shipped as #0E1018, which the brief had
   named as `.btn-solid`'s ink; measured, `.btn-solid` in dark is #07080D
   (`--gs-ground`), not #0E1018 (`--gs-surface`). Lovable confirmed the
   intent was a token-exact match, so it is #07080D here. Contrast on the
   gradient improves slightly with it: 7.35:1 indigo end and 10.59:1 green,
   against 6.98 and 10.05. Light mode is untouched and stays #fff. */
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap,
    .gsh[data-lab][data-theme="dark"] #tools) .tl-go {
  color: #07080D !important;
}
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap,
    .gsh[data-lab][data-theme="dark"] #tools) .tl-go svg {
  stroke: #07080D !important;
}

/* ---- §113 Dark-Mode Hardware Choice Tiles & Linen In-Card CTAs -------------
   GROUND UPDATED 21 Nov to warm linen #F2ECE4 (see §114). Cold #FFFFFF fought
   the linen CRT tone of `.tl-screen`. Rules 3 and 4 below — the #picks seat
   kickers — are UNCHANGED and must stay: the §114 brief omitted them, and
   dropping them would let the dark pastel tints (#5FD08A/#FF7A70/#818CF8/
   #FBBF24) back onto the tiles.
   Establishes clear hierarchy in dark mode:
   - Interactive choices & in-card triggers = crisp white hardware elements.
   - Section progression CTAs = signature brand gradient.
   - Rich discipline inks preserved on white grounds for WCAG AA compliance.

   STRUCTURE VERIFIED BEFORE BUILDING, because two of these selectors depend
   on it:
   - `.ac-consoles` holds exactly two `.ac-console` children and nothing
     else, games first ("01 // 60-SECOND GAMES") and playlists second, so
     `:nth-child(2)` is the playlists console. It is a positional selector
     against author markup, so it breaks silently if a third console or a
     wrapper is ever added between them.
   - Only the four games tiles carry `data-seat`; the playlists tiles have
     none, so rules 3 and 4 cannot collide.
   - The two Act 05 in-card CTAs are `a.btn.btn-solid.ac-cta`, while the
     section advance "Shape the roadmap" is a bare `a.btn.btn-solid` sibling.
     `.ac-cta` therefore hits the two in-card triggers and leaves the advance
     CTA on its gradient. "Read the room" in `#tools` is likewise a bare
     `.btn-solid` and is not `.tl-go`.

   THE TINTS BEING OVERRIDDEN ARE NOT §82's. The brief points at §82, but the
   dark pastel seat inks are in THIS sheet at the `.ac-tile[data-seat=...]`
   rules (#5FD08A / #FF7A70 / #818CF8 / #FBBF24), which also carry an
   `html[data-theme="dark"] .gsh ...` twin at (0,5,1). §113 wins on the
   `#picks` id — (1,5,0) — and on `!important`, which those rules lack.
   §82 itself is the Acts 04/05 vertical budget and is untouched.

   CONTRAST, computed on #FFFFFF before building — all pass AA:
     .ac-title #0E1018   18.99      marketing  #1B6B38   6.55
     .ac-time  #5D5348    7.51      sales      #B23A31   5.93
     CTA ink   #07080D   20.01      product    #3B5CCC   5.85
                                    leadership #8A4F0C   6.55

   TWO THINGS WHERE THE CODE AND THE PROSE DISAGREE, shipped as written:

   1. `.ac-time` carries `opacity: .8` as well as the colour. Composited over
      white that is #7D756D = **4.53:1** — it passes, but by 0.03. Dropping
      the opacity line would put it back to 7.51. Left as briefed.
   2. The brief's prose says the tiles get a "crisp dark border/shadow", but
      the value given is `border-color: rgba(255,255,255,.18)` on a #FFFFFF
      tile, which renders invisible — the separation comes entirely from the
      box-shadow against the dark ground. Same for the white CTAs' 1px
      `rgba(255,255,255,.2)`. Left as briefed; if a visible edge was meant,
      it needs a dark value.

   §112's `.tl-go` colour/stroke is now restated here alongside the white
   ground. Same value, so the duplication is inert; §112 is left in place
   rather than folded in, since CSS is frozen pre-launch. */

/* 1. Act 05 Tiles: solid white chassis on dark theme */
.gsh[data-theme="dark"] #picks .ac-tile {
  background: #F2ECE4 !important;
  border-color: #D8CEC2 !important;
  box-shadow: 0 4px 14px rgba(0, 0, 0, 0.28) !important;
}

/* 2. Text ink on white tiles */
.gsh[data-theme="dark"] #picks .ac-tile .ac-title,
.gsh[data-theme="dark"] #picks .ac-tile .ac-title em {
  color: #0E1018 !important;
}
.gsh[data-theme="dark"] #picks .ac-tile .ac-time {
  color: #5D5348 !important;
  opacity: 1 !important;   /* was .8, which composited to 4.53:1 */
}

/* 3. Seat kickers: canonical light-mode discipline inks on white tile */
.gsh[data-theme="dark"] #picks .ac-tile[data-seat="marketing"] .ac-seat { color: #1B6B38 !important; }
.gsh[data-theme="dark"] #picks .ac-tile[data-seat="sales"]      .ac-seat { color: #B23A31 !important; }
.gsh[data-theme="dark"] #picks .ac-tile[data-seat="product"]    .ac-seat { color: #3B5CCC !important; }
.gsh[data-theme="dark"] #picks .ac-tile[data-seat="leadership"] .ac-seat { color: #8A4F0C !important; }

/* 4. Playlists unit kicker (Card 02) on white tile */
.gsh[data-theme="dark"] #picks .ac-console:nth-child(2) .ac-tile .ac-seat {
  color: #5D5348 !important;
}

/* 5. In-Card CTAs: solid white pills with carbon ink in dark mode */
/* Act 05 (Enter the games / Open the playlists) */
.gsh[data-theme="dark"] #picks .ac-cta {
  background: #F2ECE4 !important;
  color: #07080D !important;
  border: 1px solid #D8CEC2 !important;
  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.24) !important;
}
.gsh[data-theme="dark"] #picks .ac-cta svg {
  stroke: #07080D !important;
}

/* Tool cards (Open the calculator / Audit a post) on /tools and homepage */
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap,
    .gsh[data-lab][data-theme="dark"] #tools) .tl-go {
  background: #F2ECE4 !important;
  color: #07080D !important;
  border: 1px solid #D8CEC2 !important;
  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.24) !important;
}
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap,
    .gsh[data-lab][data-theme="dark"] #tools) .tl-go svg {
  stroke: #07080D !important;
}

/* ---- §114 Warm-linen unification, and the /games CTA pill ----------------
   Companion to §113, whose grounds were updated in place rather than
   restated here — the brief's replacement block omitted §113's `#picks`
   seat-kicker rules, and pasting it over §113 would have restored the dark
   pastel tints on the homepage tiles. Only the genuinely new rules live
   here: the tools tag pill, and the whole /games treatment.

   SUPERSEDED 21 Nov: the /games seat-ink question below is moot. That card
   is no longer linen, so its kickers keep this sheet's dark `--c` values —
   see the block at the end of this section. Kept for the record because the
   measurement still stands: **#8C6D37 on #F2ECE4 is 4.11:1 and fails AA**,
   so if a linen games card ever returns, #8A4F0C is the leadership ink.

   CONTRAST ON #F2ECE4, measured before building:
     title    #0E1018  16.18      marketing  #1B6B38  5.58
     meta     #5D5348   6.40      sales      #B23A31  5.05
     CTA ink  #07080D  17.05      product    #3B5CCC  4.99
     .gh-go   #F2ECE4 on #07080D = 17.05   leadership #8A4F0C  5.58

   All pass AA. Two sit under the brief's ">5.5" line: sales 5.05 and product
   4.99. Those are the canonical discipline inks and the same values the
   homepage tiles use, so moving them is a brand-token decision, not a page
   fix. Rejected alternatives for the record: #8E2F2A would give sales 6.91
   but a different red from the homepage, defeating the parity this round is
   for.

   `--c` has exactly one consumer in this sheet (`.gh-meta .seat`), so these
   direct colours fully replace it — no element keeps a pastel accent.

   `.gh-go` inverts rather than matching the homepage pill: a linen pill on a
   linen tile would disappear, so it becomes carbon with linen ink. */

/* 1. Tools: primary tag pill joins the linen family */
:is(.gsh[data-lab][data-theme="dark"].tk-toolwrap,
    .gsh[data-lab][data-theme="dark"] #tools) .tl-tags b {
  background: #F2ECE4 !important;
  color: #07080D !important;
  border-color: #F2ECE4 !important;
}

/* 2. /games: dark chassis kept, warm linen CTA pill only
   -------------------------------------------------------------------------
   REVISED 21 Nov after Lovable's audit of `be8c4694`: the full linen card
   read as "four oversized light slabs with an empty black bar across the
   bottom". The chassis, headings, line and meta all revert to their native
   dark values; only the CTA pill and the active filter tab carry linen, which
   is what gives /games parity with `.tl-go` on /tools.

   THE SEAT-KICKER OVERRIDES ARE DROPPED, NOT KEPT, AND THAT IS A DEPARTURE.
   The brief says to keep the canonical discipline inks. Those were chosen for
   the linen ground; with the card reverted they sit on `--gs-raised`
   (#1A1D2B), where all four fail AA:

     marketing  #1B6B38  2.56      product    #3B5CCC  2.86
     sales      #B23A31  2.82      leadership #8A4F0C  2.55

   Removing the overrides lets this sheet's existing dark `--c` values apply
   again — #5FBF84 / #E07C74 / #8FA6C4 / #D2AE6A — which measure 7.39 / 5.82 /
   6.71 / 7.97 on the same ground. Shipping four AA failures to honour a
   colour choice made for a surface that no longer exists was not a trade
   worth making silently; the canonical inks cannot be made legal on a dark
   card, only on a light one.

   Reverting `.gh-card` also restores its native hover border, which the
   previous `border-color: … !important` had been outranking. */

/* Warm linen CTA pill (parity with .tl-go) */
.gsh[data-theme="dark"] .gh-go {
  background: #F2ECE4 !important;
  color: #07080D !important;
  border: 1px solid #D8CEC2 !important;
  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.24) !important;
}
.gsh[data-theme="dark"] .gh-go .gh-ar {
  stroke: #07080D !important;
}

/* Active filter tab */
.gsh[data-theme="dark"] .gh-in:checked + .gh-tab {
  background: #F2ECE4 !important;
  border-color: #F2ECE4 !important;
  color: #07080D !important;
}
