/* ═══════════════════════════════════════════════════════════════════════
   modules.css — the locked module set
   JT, 2026-10-27: "It would be a lot easier if you just made this similar
   to a divi theme. Lock that in, then any updates arent taking massive
   time investments chasing small things."

   He is right, and the measurement proves it. The pages carry 125+ legacy
   class names, and eight of them are the same object:

       .card  .tile  .domain-card  .pub-card  .treatise-card
       .rgrid>*  .grid2>*  .flist>*

   Every fix had to be written eight times and I kept missing one. That is
   why `audit-real.py` reported 174 elements computing a radius under 8px
   while the radius rule "already existed": it named `.rgrid>*` and never
   `.domain-card`, of which there are 84.

   Deloitte proves the model. Their cmp-* library IS a module system, which
   is why 48 of their pages are internally consistent and 36 of ours are
   not.

   TWELVE MODULES. One canonical class each. One block of CSS each. A page
   is a sequence of modules, and fixing a module fixes every page at once.

   ⛔ APPLIED ADDITIVELY. harness/modularize.py adds m-* classes to the
   elements already present; it does not rewrite markup. No structural
   risk, no reindex breakage, and the legacy names simply stop mattering
   because nothing styles them any more.

   ⛔ COLOUR IS NOT SET HERE. surface.js marks every text element with its
   measured ground and register-v23-repair.css section 9 owns colour. A
   module that sets its own text colour reintroduces the exact fight this
   file exists to end.
   ═══════════════════════════════════════════════════════════════════════ */

:root{
  --m-gap:clamp(14px,1.6vw,20px);
  --m-pad:clamp(18px,2.1vw,28px);
  --m-band:clamp(34px,4.4vw,64px);
  --m-radius:16px;
}

/* ⛔ EVERY MODULE SELECTOR IS PREFIXED `html `, AND THAT IS LOAD-BEARING.
   Measured on /silos/: a .pub-card computed border-radius 0px with
   modules.css loaded and linked. The cascade showed why —

       modules.css  .m-cards__item   var(--m-radius)
       INLINE       .pub-card        var(--radius)     <- won

   The pages carry their own inline <style> blocks. `.pub-card` is
   specificity (0,1,0), identical to `.m-cards__item`, and an inline sheet
   comes last in document order, so it wins every tie. A module layer that
   per-page CSS can outrank is not canonical.

   `html .m-cards__item` is (0,1,1) and beats it without !important,
   which keeps the sheet debuggable. Same reason repair.css section 9
   prefixes the surface rules.
*/

/* ── the band every module sits in ─────────────────────────────────── */
html .m{padding:var(--m-band) 0}
html .m-in{max-width:var(--max);margin:0 auto;padding:0 var(--gut);width:100%}
html .m + .m{border-top:1px solid var(--rule-light)}
html .m--tint{background:var(--hue-soft)}
html .m--dark{background:var(--deep)}
html .m--flush{padding-top:0}

/* ── 1. HERO ───────────────────────────────────────────────────────── */
html .m-hero{position:relative;min-height:min(46vh,460px);display:flex;
  align-items:flex-end;overflow:hidden;
  background:linear-gradient(142deg,var(--deep-surface) 0%,
    var(--deep) 58%,#020d18 100%);
  border-bottom:3px solid var(--hue-ink)}
@supports (height:100svh){.m-hero{min-height:min(46svh,460px)}}
html .m-hero .m-in{padding-block:clamp(26px,3.6vw,52px);position:relative;z-index:2}
html .m-hero__eyebrow{font-family:var(--sans);font-size:12px;font-weight:700;
  letter-spacing:.085em;text-transform:uppercase;margin:0 0 12px}
html .m-hero h1{max-width:22ch;margin:0 0 14px}
html .m-hero__lede{font-size:clamp(16px,1.5vw,19px);line-height:1.55;
  max-width:56ch;margin:0 0 22px}

/* ── 2. NAV, the in-page jump strip ────────────────────────────────── */
html .m-nav{background:var(--deep-surface);padding:0;
  border-bottom:1px solid var(--rule-deep)}
html .m-nav .m-in{display:flex;align-items:center;min-height:56px}
html .m-nav__list{display:flex;flex-wrap:wrap;gap:2px 26px;align-items:center;
  width:100%;list-style:none;margin:0;padding:0}
html .m-nav__list a{font-family:var(--sans);font-size:14px;font-weight:600;
  text-decoration:none;min-height:44px;display:inline-flex;
  align-items:center;border-bottom:2px solid transparent;border-radius:0}
html .m-nav__list a:hover,.m-nav__list a:focus-visible{
  border-bottom-color:var(--hue-lift,var(--hue-ink))}

/* ── 3. PROSE ──────────────────────────────────────────────────────── */
html .m-prose .m-in{max-width:min(var(--max),880px)}
html .m-prose p{font-size:16.5px;line-height:1.62;max-width:68ch}
html .m-prose h2{margin-bottom:12px}

/* ── 4. CARDS — replaces EIGHT legacy class names ──────────────────── */
html .m-cards{display:grid;gap:var(--m-gap);
  grid-template-columns:repeat(auto-fit,minmax(min(258px,100%),1fr))}
html .m-cards__item{background:#fff;border:1px solid var(--rule-light);
  border-top:3px solid var(--hue-ink);border-radius:var(--m-radius);
  padding:var(--m-pad);min-width:0;display:flex;flex-direction:column}
html .m-cards__item h3{font-family:var(--sans);
  font-size:clamp(15px,1.4vw,17px);line-height:1.25;margin:0 0 9px}
html .m-cards__item p{font-size:15px;line-height:1.52;margin:0 0 10px;
  max-width:none}
html .m-cards__item > :last-child{margin-bottom:0}
html .m-cards__item a{font-size:14px;font-weight:600;min-height:44px;
  display:inline-flex;align-items:center;margin-top:auto}
@media(max-width:620px){.m-cards{grid-template-columns:1fr}}

/* ── 5. MEDIA + copy ───────────────────────────────────────────────── */
html .m-media .m-in{display:grid;gap:clamp(18px,2.4vw,34px);align-items:center;
  grid-template-columns:minmax(0,1fr)}
@media(min-width:900px){
  html .m-media .m-in{grid-template-columns:minmax(0,6fr) minmax(0,6fr)}
  html .m-media--flip .m-media__fig{order:-1}
}
html .m-media__fig{margin:0;min-width:0}
html .m-media__fig picture{display:block;width:100%}
html .m-media__fig img{display:block;width:100%;height:auto;max-width:100%;
  border:1px solid var(--rule-light);border-radius:var(--m-radius);
  background:var(--paper)}
html .m-media__fig figcaption{font-size:13px;line-height:1.5;margin-top:9px}

/* ── 6. TABLE ──────────────────────────────────────────────────────── */
html .m-table{overflow-x:auto;-webkit-overflow-scrolling:touch;
  border:1px solid var(--rule-light);border-radius:var(--m-radius);
  background:#fff}
html .m-table table{width:100%;border-collapse:collapse;min-width:620px}
html .m-table th,.m-table td{text-align:left;padding:12px 16px;
  border-bottom:1px solid var(--rule-light);font-size:15px;
  line-height:1.48;vertical-align:top}
html .m-table thead th{background:var(--deep);font-family:var(--sans);
  font-size:12.5px;font-weight:700;border-bottom:none}
html .m-table tbody tr:last-child td{border-bottom:none}
html .m-table td > a:only-child{min-height:44px;display:inline-flex;
  align-items:center}

/* ── 7. QUOTE ──────────────────────────────────────────────────────── */
html .m-quote{background:var(--deep);padding-block:clamp(38px,5vw,72px)}
html .m-quote blockquote{margin:0;max-width:30ch;font-family:var(--serif);
  font-size:clamp(22px,3.1vw,38px);line-height:1.22;letter-spacing:-.016em}
html .m-quote cite{display:block;margin-top:18px;font-family:var(--sans);
  font-size:12px;font-weight:700;letter-spacing:.08em;
  text-transform:uppercase;font-style:normal}

/* ── 8. STATS ──────────────────────────────────────────────────────── */
html .m-stats{display:grid;gap:18px;
  grid-template-columns:repeat(auto-fit,minmax(min(190px,100%),1fr))}
html .m-stats > div{border-top:3px solid var(--hue-ink);padding-top:14px;
  min-width:0}
html .m-stats b{display:block;font-family:var(--sans);
  font-size:clamp(28px,4vw,46px);font-weight:700;line-height:1;
  letter-spacing:-.02em;font-variant-numeric:tabular-nums}
html .m-stats span{display:block;margin-top:7px;font-size:13.5px;line-height:1.45}

/* ── 9. CTA ────────────────────────────────────────────────────────── */
html .m-cta{background:var(--deep-card)}
html .m-cta .m-in{display:grid;gap:clamp(22px,3vw,48px);align-items:center;
  grid-template-columns:minmax(0,1fr)}
@media(min-width:860px){
  html .m-cta--split .m-in{grid-template-columns:minmax(0,1fr) minmax(0,1fr)}
}
html .m-cta h2{max-width:22ch;margin:0 0 12px}
html .m-cta p{max-width:52ch;margin:0 0 20px}

/* ── 10. FAQ ───────────────────────────────────────────────────────── */
html .m-faq{display:grid;gap:0}
html .m-faq details{border-top:1px solid var(--rule-light);
  border-radius:var(--m-radius)}
html .m-faq summary{cursor:pointer;min-height:56px;display:flex;
  align-items:center;font-family:var(--sans);font-size:15.5px;
  font-weight:600;padding:10px 0;list-style:none}
html .m-faq summary::-webkit-details-marker{display:none}
html .m-faq p{padding:0 0 15px;max-width:68ch}

/* ── 11. FIGURE, the generated SVGs ────────────────────────────────── */
html .m-figure{margin:0}
html .m-figure img{display:block;width:100%;height:auto;max-width:100%;
  border:1px solid var(--rule-light);border-radius:var(--m-radius);
  background:#fff;padding:clamp(10px,1.4vw,20px)}
html .m-figure figcaption{margin-top:10px;font-size:13px;line-height:1.5;
  max-width:78ch}

/* ── 12. SLIDER ────────────────────────────────────────────────────── */
html .m-slider{display:grid;grid-auto-flow:column;
  grid-auto-columns:min(100%,520px);gap:var(--m-gap);overflow-x:auto;
  scroll-snap-type:x mandatory;scroll-behavior:smooth;padding-bottom:10px;
  overscroll-behavior-x:contain}
html .m-slider__item{scroll-snap-align:center;min-width:0;background:#fff;
  border:1px solid var(--rule-light);border-top:3px solid var(--hue-ink);
  border-radius:var(--m-radius);padding:var(--m-pad)}
@media (prefers-reduced-motion:reduce){.m-slider{scroll-behavior:auto}}

/* ── motion, applied per module not per page ───────────────────────── */
@media (prefers-reduced-motion:no-preference){
  html.js-motion .m[data-reveal]{opacity:0}
  html.js-motion .m[data-reveal].is-in{animation:mo-rise .5s
    cubic-bezier(.22,.61,.36,1) both}
  html.js-motion .m-cards__item[data-reveal]{opacity:0}
  html.js-motion .m-cards__item[data-reveal].is-in{
    animation:mo-rise .46s cubic-bezier(.22,.61,.36,1) both}
}

/* ── 13. WIDE MEDIA, for assets that carry information ───────────────
   audit-real.py: claim-teams-usmca.webp rendered at 610px of its 1,561px
   intrinsic width — 39%, and 23% at the worst measurement. A product
   screenshot at 39% is decoration we paid for and nobody can read, which
   is exactly what JT reported: "the teams images peppered throughout,
   are all unreadable, therefore useless, not impactful."

   A two-column media band gives the figure half the container. A CLAIM
   asset gets the whole of it instead, which takes 610px to ~1,200px:
   77% of intrinsic rather than 39%. The copy sits above it, not beside.

   Backdrops are exempt: a .stage-bg sits BEHIND a panel by design and is
   atmosphere, not content. */
html .m-media--wide .m-in{grid-template-columns:minmax(0,1fr) !important}
html .m-media--wide .m-media__fig{max-width:min(100%,1200px);
  margin-inline:auto}
html .m-media--wide .m-media__fig img{width:100%}
@media(min-width:900px){
  html .m-media--wide .m-in{grid-template-columns:minmax(0,1fr)}
}

/* .notice was the one genuinely square element left after the module
   layer took over: 12 of them, no radius anywhere. */
html .notice,html .callout,html .plot-card{border-radius:var(--m-radius)}

/* ── 14. THE LAST 225, all one shape ─────────────────────────────────
   Every remaining contrast failure was the on-deep family on a LIGHT
   ground, concentrated on the dossier pages:

       rgb(232,234,240) on rgb(254,242,242)   42   on-deep on hue-soft
       rgb(156,163,176) on rgb(254,242,242)   42
       rgb(232,234,240) on white              34

   Cause: dossier.css sets `color:var(--on-deep)` on the claim band at
   (0,3,2), and the fix that kept that band's background DARK was deleted
   when sections 9-12 were replaced wholesale. Light text, light ground.

   Two rules. The band is dark again, so the light text is correct; and
   no treatment may assert a text colour, because colour belongs to the
   measured surface and nowhere else. */
html [data-treatment="dossier"] main > section.arch-claim{
  background:var(--deep)}
html [data-treatment="dossier"] main > section.arch-claim,
html [data-treatment="dossier"] main > section.arch-claim *{
  color:inherit}

/* The catch-all sits above any treatment that still asserts a colour.
   html+body+attribute is (0,1,2) and beats a three-class treatment
   selector reaching a descendant by inheritance. */
/* ⛔ SCOPED TO main. The catch-all reached the CHROME and broke it: the
   skip link computed rgb(185,28,28) on rgb(245,158,11) because
   `html body a[data-on="light"]` overrode `.skip-link{color:var(--deep)}`.
   Red on amber, 2.4:1, caused by the rule meant to fix contrast.

   The masthead, skip link and colophon carry deliberate colours from
   register-v23-chrome.css and are not part of the surface contract.
   Colour by measured ground governs CONTENT. */
html body main [data-on="light"]{color:var(--ink-muted)}
html body main h1[data-on="light"],html body main h2[data-on="light"],
html body main h3[data-on="light"],html body main h4[data-on="light"],
html body main b[data-on="light"],html body main strong[data-on="light"],
html body main cite[data-on="light"]{color:var(--ink)}
html body main small[data-on="light"],html body main span[data-on="light"],
html body main figcaption[data-on="light"]{color:var(--ink-dim)}
html body main [data-on="dark"]{color:var(--on-deep-muted)}
html body main h1[data-on="dark"],html body main h2[data-on="dark"],
html body main h3[data-on="dark"],html body main h4[data-on="dark"],
html body main b[data-on="dark"],html body main strong[data-on="dark"],
html body main cite[data-on="dark"]{color:var(--on-deep)}

html body main a[data-on="light"],
html body main .band-eyebrow[data-on="light"],
html body main .cite-line[data-on="light"],
html body main .m-hero__eyebrow[data-on="light"]{color:var(--hue-ink)}
html body main a[data-on="dark"],
html body main .band-eyebrow[data-on="dark"],
html body main .cite-line[data-on="dark"],
html body main .m-hero__eyebrow[data-on="dark"]{color:var(--hue-lift,#5eead4)}

html body main a.btn,html body main a.btn *,
html body main button.btn,html body main button.btn *,
html body main .btn-primary,html body main .btn-primary *{
  color:#041424;background:var(--gold-rule);border-color:var(--gold-rule)}

/* ── 15. STEPPER — the click-through that replaces the slider ─────────
   JT: "get rid of the slider, make it a click through animation."

   ⛔ CSS-ONLY. Hidden radios plus :checked. The live site has already
   shown that script execution cannot be assumed: .htaccess sets
   X-Content-Type-Options: nosniff and declared no MIME type for .js,
   which kills every script silently while the CSS still renders. A
   click-through depending on JS would be the next casualty. Radios also
   give keyboard operation and screen-reader announcement for free. */
html .m-step{position:relative}
html .m-step__radio{position:absolute;opacity:0;width:1px;height:1px;
  pointer-events:none}
html .m-step__tabs{display:flex;flex-wrap:wrap;gap:8px;margin:0 0 18px}
html .m-step__tab{cursor:pointer;min-height:44px;display:inline-flex;
  align-items:center;padding:0 16px;border-radius:999px;
  border:1px solid var(--rule-light);background:#fff;
  font-family:var(--sans);font-size:13.5px;font-weight:600;
  color:var(--ink-muted);transition:background .25s ease,color .25s ease,
  border-color .25s ease}
html .m-step__tab:hover{border-color:var(--hue-ink)}
html .m-step__panels{display:grid}
html .m-step__panel{grid-area:1/1;background:#fff;
  border:1px solid var(--rule-light);border-top:3px solid var(--hue-ink);
  border-radius:var(--m-radius);padding:var(--m-pad);
  opacity:0;visibility:hidden;transform:translateY(10px);
  transition:opacity .34s ease,transform .34s ease,visibility .34s}
/* nth radio checked -> nth tab active and nth panel shown */
html .m-step__radio:nth-of-type(1):checked ~ .m-step__panels
  > .m-step__panel:nth-child(1),
html .m-step__radio:nth-of-type(2):checked ~ .m-step__panels
  > .m-step__panel:nth-child(2),
html .m-step__radio:nth-of-type(3):checked ~ .m-step__panels
  > .m-step__panel:nth-child(3),
html .m-step__radio:nth-of-type(4):checked ~ .m-step__panels
  > .m-step__panel:nth-child(4),
html .m-step__radio:nth-of-type(5):checked ~ .m-step__panels
  > .m-step__panel:nth-child(5),
html .m-step__radio:nth-of-type(6):checked ~ .m-step__panels
  > .m-step__panel:nth-child(6){
  opacity:1;visibility:visible;transform:none}
html .m-step__radio:nth-of-type(1):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(1),
html .m-step__radio:nth-of-type(2):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(2),
html .m-step__radio:nth-of-type(3):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(3),
html .m-step__radio:nth-of-type(4):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(4),
html .m-step__radio:nth-of-type(5):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(5),
html .m-step__radio:nth-of-type(6):checked ~ .m-step__tabs
  > .m-step__tab:nth-child(6){
  background:var(--hue-ink);border-color:var(--hue-ink);color:#fff}
html .m-step__radio:focus-visible ~ .m-step__tabs .m-step__tab{
  outline:2px solid var(--hue-ink);outline-offset:3px}
@media (prefers-reduced-motion:reduce){
  html .m-step__panel{transition:none}
}

/* -- 16. BACKDROP PANEL: the image behind, the words in front ---------
   JT: "i keep mentioning this massive wasted blue space in the
   background. it need an image here or something. too much navy, very
   boring having a gradiant only." And on the same asset elsewhere: "its
   still too small to read. get rid of it, or make it bigger by putting
   it in the background."

   One treatment answers both, and it is the one he approved on the home
   page: "i really like how the azure image got put in the back ground."

   The image runs full-bleed at its natural scale instead of being boxed
   at 39% of intrinsic width, and the copy sits on a plate over a
   gradient so contrast never depends on what is underneath it. */
html .m-backdrop{position:relative;overflow:hidden;
  background:var(--deep);min-height:min(62vh,620px);display:flex;
  align-items:center}
@supports (height:100svh){
  html .m-backdrop{min-height:min(62svh,620px)}
}
html .m-backdrop__bg{position:absolute;inset:0;z-index:0}
html .m-backdrop__bg img{width:100%;height:100%;object-fit:cover;
  object-position:center top;opacity:.55}
html .m-backdrop::after{content:"";position:absolute;inset:0;z-index:1;
  background:linear-gradient(100deg,rgba(4,20,36,.94) 0%,
    rgba(4,20,36,.82) 38%,rgba(4,20,36,.28) 72%,
    rgba(4,20,36,.18) 100%)}
html .m-backdrop .m-in{position:relative;z-index:2;
  padding-block:clamp(34px,5vw,72px)}
html .m-backdrop__plate{max-width:min(46ch,52%)}
html .m-backdrop__plate h2{margin:0 0 12px}
html .m-backdrop__plate p{max-width:46ch;margin:0 0 12px}
html .m-backdrop__plate > :last-child{margin-bottom:0}
@media(max-width:860px){
  html .m-backdrop__plate{max-width:none}
  html .m-backdrop__bg img{opacity:.3}
  html .m-backdrop::after{
    background:linear-gradient(180deg,rgba(4,20,36,.88) 0%,
      rgba(4,20,36,.93) 100%)}
}

/* -- 17. THE HERO GETS AN IMAGE, not only a gradient -----------------
   "too much navy, very boring having a gradiant only."
   The hero already carries a dark gradient. A backdrop behind it at low
   opacity gives the navy something to be, without ever competing with
   the headline: the gradient still owns the contrast. */
html .m-hero__bg{position:absolute;inset:0;z-index:0}
html .m-hero__bg img{width:100%;height:100%;object-fit:cover;
  object-position:center 28%;opacity:.26}
html .m-hero::after{content:"";position:absolute;inset:0;z-index:1;
  background:linear-gradient(115deg,rgba(4,20,36,.92) 0%,
    rgba(4,20,36,.74) 46%,rgba(4,20,36,.52) 100%)}

/* -- 18. CONVERGE: the animation that argues ------------------------
   JT: "You are animating static items that dont deliver any punch. If we
   have the animation, why arent we using it to bring disparate items
   viewers are searching for together. Instead of treatises."

   The mechanism was measured off deloitte.com's photo collage, ported,
   and verified live at 370px of travel. It was then pointed at a grid of
   treatise names, where it only made cards arrive. Motion that decorates
   costs load and buys nothing.

   Here the six things one legal question actually needs start scattered
   at the edges and resolve into a single chain as you scroll, driven by
   the same --p progress. The separation IS the problem; convergence IS
   the answer. The animation says the sentence.

   ⛔ DEGRADES TO A LIST. With no JS, --p never arrives, the fallback of 1
   applies, and the chips sit in a plain labelled grid that still reads
   correctly. Under prefers-reduced-motion the transforms never run at
   all. The argument survives the animation being absent. */
html .m-conv .m-in{max-width:min(var(--max),1100px)}
html .m-conv__lede{max-width:52ch;margin:0 0 clamp(20px,3vw,34px)}
html .m-conv__field{position:relative;display:grid;gap:12px;
  grid-template-columns:repeat(auto-fit,minmax(min(235px,100%),1fr))}
html .m-conv__chip{background:#fff;border:1px solid var(--rule-light);
  border-left:3px solid var(--hue-ink);border-radius:var(--m-radius);
  padding:14px 16px;min-width:0}
html .m-conv__chip b{display:block;font-family:var(--sans);
  font-size:14.5px;line-height:1.3;margin-bottom:4px}
html .m-conv__chip span{display:block;font-family:var(--sans);
  font-size:11.5px;font-weight:700;letter-spacing:.07em;
  text-transform:uppercase}
html .m-conv__core{grid-column:1/-1;background:var(--deep);
  border-radius:var(--m-radius);padding:clamp(18px,2.4vw,28px);
  border-top:3px solid var(--hue-lift,#5eead4);margin-top:6px}
html .m-conv__core b{display:block;font-family:var(--serif);
  font-size:clamp(19px,2.4vw,27px);line-height:1.2;margin-bottom:8px}
html .m-conv__core span{display:block;font-size:13.5px}

/* The travel. Odd chips enter from the left, even from the right, the
   outermost furthest — the measured spread on their collage was -199 to
   +100. (1 - p) runs 1 at entry to 0 at settle, so they arrive rather
   than snap. */
@media (prefers-reduced-motion:no-preference){
  html.js-motion .m-conv__chip{
    transform:translate3d(calc(var(--cx,0px) * (1 - var(--p,1))),
                          calc(var(--cy,0px) * (1 - var(--p,1))),0);
    opacity:calc(.35 + .65 * var(--p,1));
    will-change:transform,opacity}
  html.js-motion .m-conv__chip[data-i="1"]{--cx:-260px;--cy:-40px}
  html.js-motion .m-conv__chip[data-i="2"]{--cx:240px;--cy:-56px}
  html.js-motion .m-conv__chip[data-i="3"]{--cx:-180px;--cy:30px}
  html.js-motion .m-conv__chip[data-i="4"]{--cx:200px;--cy:46px}
  html.js-motion .m-conv__chip[data-i="5"]{--cx:-120px;--cy:64px}
  html.js-motion .m-conv__chip[data-i="6"]{--cx:150px;--cy:72px}
  html.js-motion .m-conv__core{
    opacity:calc(var(--p,1) * var(--p,1));
    transform:translate3d(0,calc(26px * (1 - var(--p,1))),0)}
}
html .m-conv{overflow-x:clip}
@media(max-width:620px){
  html .m-conv__field{grid-template-columns:1fr}
}

/* ⛔ THE TRAVEL IS IN FIXED PIXELS AND THE VIEWPORT IS NOT.
   JT, 2026-10-29, phone screenshot of the home page: content bleeding
   past both edges at the band junction.

   MEASURED at 390px: a .m-conv__chip is 350px wide and chip 1 carries
   --cx:-260px, so it sits at left=-240 and chip 2 at right=610. Six
   chips leave the viewport on both sides at once. The only thing
   stopping that being a horizontally scrolling page is `overflow-x:clip`
   on the band, which means the animation's whole first half plays
   outside the screen: the chips do not travel in, they appear from
   under a hard edge.

   The offsets were measured off a 1,208px desktop band, where -260px is
   21% of the width. Below 620px they become a function of the viewport
   instead, so the gesture is the same shape at every size and nothing
   relies on the clip to hide a mistake. The field is one column there,
   so a vertical-only arrival is the honest version of the move. */
@media (max-width:620px) and (prefers-reduced-motion:no-preference){
  html.js-motion .m-conv__chip[data-i]{--cx:0px}
  html.js-motion .m-conv__chip[data-i="1"]{--cy:-26px}
  html.js-motion .m-conv__chip[data-i="2"]{--cy:-18px}
  html.js-motion .m-conv__chip[data-i="3"]{--cy:22px}
  html.js-motion .m-conv__chip[data-i="4"]{--cy:30px}
  html.js-motion .m-conv__chip[data-i="5"]{--cy:38px}
  html.js-motion .m-conv__chip[data-i="6"]{--cy:46px}
}

/* -- 19. TYPE, SPACE AND THE HERO IMAGE -----------------------------
   JT, 2026-10-28: "the text is too small, too faint of contrast. the
   images with text are too small to read. Page 1, still has no hero
   image as the background. pages seem to need to scroll a lot. makes
   them unneccessarly long. if its a lot of mouse wheel rotations to get
   down a page on desktop, Mobile, will be awful."

   MEASURED, all four confirmed:
     /pricing/  body text 13px on 20 elements, 12px on 7
     every page hero image: NONE
     /           8,630px = 9.6 screens
     /westlaw-alternative/ 8,252px = 9.2 screens
     Deloitte's contact page is 2,741px, their median page ~5,100px.

   The 16.5px floor set earlier lost to per-page inline CSS, the same way
   the radius rules did before the module layer was raised. So the type
   scale is stated at html+body+main depth where nothing on a page can
   outrank it. */

/* TYPE: a real floor, and darker. --ink-muted was #374151 at 9.1:1;
   #1f2937 is 13.2:1 and reads as black rather than grey at small sizes. */
:root{--ink:#0b1220;--ink-muted:#1f2937;--ink-dim:#3f4a5a}
html body main p,html body main li,html body main dd,
html body main td,html body main th{font-size:17.5px;line-height:1.6}
html body main .m-cards__item p,html body main .m-conv__chip b,
html body main .m-step__panel p{font-size:16px;line-height:1.5}
html body main small,html body main figcaption,
html body main .auth,html body main .src,html body main .band-note,
html body main .updated,html body main .signin-note,
html body main .gate-stamp,html body main .tier-unit{font-size:14px}
html body main .m-cards__item h3,html body main h3{font-size:18px}
html body main .m-hero__lede,html body main .lede,
html body main .arch-lede{font-size:clamp(18px,1.7vw,21px)}
html body main .band-eyebrow,html body main .eyebrow,
html body main .cite-line,html body main .m-hero__eyebrow,
html body main .tier-name{font-size:13px;letter-spacing:.07em}

/* SPACE: the bands were clamp(34px,4.4vw,64px) top and bottom, so a
   12-band page spent ~1,500px on padding alone. Cutting to
   clamp(22px,2.6vw,40px) removes roughly a third of the scroll without
   touching a word. Deloitte's median section is 0.48 viewports; ours
   were running past that on padding, not content. */
html body main > .m,html body main > section{
  padding:clamp(22px,2.6vw,40px) 0}
html body main > .m.m-hero{padding:0}
html body main > .m + .m{border-top:1px solid var(--rule-light)}
html body main .m-cards{gap:12px}
html body main .m-in > *:last-child{margin-bottom:0}
html body main h2{margin-bottom:10px}
html body main p:last-of-type{margin-bottom:0}

/* IMAGES WITH TEXT: bigger. A media figure was capped by its column;
   these let it use the full measure, which takes a 1,561px capture from
   610px to ~1,060px. */
html body main .m-media__fig img,
html body main .m-figure img{max-width:none;width:100%}
html body main .m-media--wide .m-media__fig{max-width:min(100%,1180px)}

/* THE HERO IMAGE. The layer existed and no markup emitted it. */
html .m-hero__bg{position:absolute;inset:0;z-index:0;overflow:hidden}
html .m-hero__bg img{width:100%;height:100%;object-fit:cover;
  object-position:center 30%;opacity:.3}
html .m-hero{position:relative}
html .m-hero > *:not(.m-hero__bg){position:relative;z-index:2}

/* -- 20. THE FIELD UNDER THE HOME STAGE ------------------------------
   The home page keeps backdrop-1-word.webp, which is the real Word task
   pane and is the single most persuasive asset on the site. What was
   wrong is what sat AROUND it: 1440x900 of flat #041424. The citation
   field goes underneath, so the pane reads as a lit object on a ground
   instead of a card floating on a gradient. */
/* ⛔ DO NOT SET position ON .stage-bg. It was `position:relative` here
   from 2026-10-27 to 2026-10-28, added to give .m-hero__bg a containing
   block, and it silently killed the home page pinned scroll: this
   selector is (0,1,1) and beats register-v23.css:246 `.stage-bg`
   (0,1,0), which sets `position:sticky`. Measured: the backdrop held the
   viewport top on 0 of 7 scroll samples. `sticky` ALREADY establishes a
   containing block for absolutely positioned children, so the override
   bought nothing and cost the signature motion on page one.
   frame-gate.py check 1 fails if this regresses. */
html .stage-bg > .m-hero__bg{position:absolute;inset:0;z-index:0}
html .stage-bg > .m-hero__bg img{width:100%;height:100%;object-fit:cover;
  object-position:50% 38%;opacity:.55}
html .stage-bg > img{position:relative;z-index:1}
/* The vignette keeps the headline plate readable over the field. Without
   it the top-left rows sit at 0.30 opacity directly under the h1. */
html .stage-bg::after{content:"";position:absolute;inset:0;z-index:1;
  pointer-events:none;
  background:linear-gradient(100deg,rgba(3,15,29,.92) 0%,
             rgba(3,15,29,.72) 34%,rgba(3,15,29,.18) 62%,
             rgba(3,15,29,.42) 100%)}
html .stage > .over{position:relative;z-index:2}

/* Interior heroes: the same field, quieter, because those pages carry
   long copy over it rather than a plate. */
html .m-hero__bg img{opacity:.38}
html .m-hero::after{content:"";position:absolute;inset:0;z-index:1;
  pointer-events:none;
  background:linear-gradient(180deg,rgba(3,15,29,.78),rgba(3,15,29,.92))}
html .m-hero > *:not(.m-hero__bg){z-index:2}

/* -- 21. THE MEASURE, AND WHY THE PAGES WERE LONG --------------------
   JT, 2026-10-28: "pages seem to need to scroll a lot. makes them
   unneccessarly long. if its a lot of mouse wheel rotations to get down
   a page on desktop, Mobile, will be awful."

   MEASURED on /westlaw-alternative/ at 1440px, per band:
     band width   1440px
     inner width  1320px
     paragraphs    575px      <- 44% of the band, 745px left empty
     one band      1312px tall for 356 words

   Padding was only 9% of page height, so trimming padding was never
   going to fix this and the first attempt duly returned 3%. The height
   is a single narrow column of prose running down the middle of a wide
   band. 68ch is a correct measure for a line, and it is the wrong
   LAYOUT for 356 words on a 1440px screen.

   So long prose bands run in two columns on desktop. The measure per
   column stays inside the readable range, the empty 745px does the
   work, and the band halves. Below 1080px it is one column again,
   because two columns of 40ch on a phone is unreadable. */
@media (min-width: 1080px){
  html body main > .m.is-long > .m-in,
  html body main > section.is-long > .wrap,
  html body main > section.is-long > .m-in{
    column-count:2;column-gap:clamp(32px,3.4vw,64px);
    column-rule:1px solid var(--rule-light)}
  /* A heading must not sit at the foot of column one with its paragraph
     stranded at the head of column two. */
  html body main > .is-long h2,html body main > .is-long h3{
    column-span:all;break-after:avoid}
  html body main > .is-long p,html body main > .is-long li{
    break-inside:avoid;max-width:none}
  /* Cards, figures and the converge field are never columnised: they
     are already multi-column layouts of their own. */
  html body main > .is-long .m-cards,
  html body main > .is-long figure,
  html body main > .is-long .m-conv__field,
  html body main > .is-long table{column-span:all}
}

/* -- 22. THE TYPE FLOOR, STATED SO IT WINS ---------------------------
   Section 19 set the floor at html body main p, which is four element
   selectors, specificity (0,0,4). Every legacy per-page rule carries a
   class, so (0,1,1) beat it and /pricing/ still measured 13px on 20
   elements and 12px on 7 after the change landed.

   Raising the element count cannot win that race: one class outranks
   any number of elements. The floor is a design-system token, not a
   page style, so it is declared as one. The exceptions below are
   enumerated and carry the same weight, which keeps real hierarchy
   (ledes larger, captions smaller) instead of flattening everything. */
html body main p,html body main li,html body main dd,
html body main td,html body main th{font-size:17.5px !important;
  line-height:1.62 !important}
html body main .m-cards__item p,html body main .m-step__panel p,
html body main .m-conv__chip b{font-size:16px !important}
html body main small,html body main figcaption,html body main .auth,
html body main .src,html body main .band-note,html body main .updated,
html body main .signin-note,html body main .gate-stamp,
html body main .tier-unit,html body main .m-conv__chip span{
  font-size:14px !important}
html body main .band-eyebrow,html body main .eyebrow,
html body main .cite-line,html body main .m-hero__eyebrow,
html body main .tier-name,
/* The answer-card eyebrow belongs in THIS list, not in mechanisms.css.
   It is a <p>, so the 17.5px !important floor above claimed it and no
   rule in another sheet could win without !important of its own. The
   floor documents its exceptions here on purpose; adding a sixth is the
   intended way to extend it, and fighting it from a later sheet is not. */
html body main .m-answer__tag{font-size:12px !important;
  letter-spacing:.09em}
/* The index count is a display number, not body copy. It is a <p>, so
   the 17.5px floor above claimed it and "30 in the index" rendered at
   reading size, which is the opposite of the point: the count IS the
   impact. Declared here with the other floor exceptions rather than
   fought from a later sheet. */
html body main .m-index__count{
  font-size:clamp(28px,3.4vw,46px) !important;line-height:1.05 !important}
/* Ledes last, so they outrank the floor above at equal weight. */
html body main .m-hero__lede,html body main .lede,
html body main .arch-lede,html body main .m-conv__lede{
  font-size:clamp(18px,1.6vw,21px) !important;line-height:1.5 !important}
html body main p{max-width:72ch}

/* -- 23. THE MOBILE FOLD ---------------------------------------------
   Above 1080px the fold is invisible: the summary is gone and the body
   flows as ordinary prose inside the two-column band. Below it, the
   summary is a real 44px control naming the word count behind it. */
html body main details.m-fold{margin:0;padding:0;border:0}
html body main details.m-fold > summary{display:none}
html body main .m-fold__body{display:contents}
@media (max-width:1079px){
  html body main details.m-fold > summary{
    display:flex;align-items:center;gap:10px;
    min-height:44px;margin-top:14px;padding:10px 2px;
    font:600 15px/1.2 var(--ui,system-ui,sans-serif);
    color:var(--gold,#f59e0b);cursor:pointer;list-style:none;
    border-top:1px solid var(--rule-light)}
  html body main details.m-fold > summary::-webkit-details-marker{display:none}
  html body main details.m-fold > summary::after{
    content:"";width:8px;height:8px;margin-left:2px;
    border-right:2px solid currentColor;border-bottom:2px solid currentColor;
    transform:rotate(45deg) translateY(-2px);transition:transform .18s}
  html body main details.m-fold[open] > summary::after{
    transform:rotate(-135deg) translateY(0)}
  html body main details.m-fold[open] > summary{color:var(--ink-dim,#3f4a5a)}
  html body main .m-fold__body{display:block}
}

/* -- 24. THE PINNED PANELS WERE A SCREEN EACH ------------------------
   MEASURED on / at 1440x900, the five .over panels inside the two
   stages:

     .over.open        900px  min-height 900px  pad 90/90    81 words
     .over.statement   768px                    pad 120/150  88 words
     .over.trust       900px  min-height 900px  pad 120/120  64 words
     .over.trust       900px  min-height 900px  pad 120/120  97 words
     .over.trust       900px  min-height 900px  pad 120/120 111 words

   4,368px of scroll for 441 words: a full viewport per 64 to 111 words.
   Deloitte's median section, measured off their own pages, is 0.48
   viewports for 43 words. We were running at roughly twice their scroll
   cost per section, and that is the whole of JT's "mouse wheel
   rotations" note on the home page.

   The hero keeps its full screen, because an opening panel that does
   not fill the viewport reads as a mistake. Every panel after it drops
   to 58vh and loses half its padding. The sticky backdrop is untouched,
   so the pinned-scroll effect JT approved still works exactly as it
   did: the panels simply stop idling over it. */
html .stage .over:not(.open){min-height:58vh;
  padding-top:clamp(42px,5vh,68px);padding-bottom:clamp(42px,5vh,68px)}
html .stage .over.open{min-height:min(900px,96vh)}
@media (max-width:1039px){
  /* Unpinned on a phone, so min-height only adds dead space. */
  html .stage .over:not(.open){min-height:0;
    padding-top:clamp(28px,4vh,44px);padding-bottom:clamp(28px,4vh,44px)}
  html .stage .over.open{min-height:0}
}

/* -- 25. DENSE CARD DECKS ON A PHONE ---------------------------------
   MEASURED at 375x812 on /silos/: one band 3,135px tall holding 33
   cards and 219 words, another 2,696px holding 21 cards and 149 words.
   That is 54 stacked cards averaging SEVEN WORDS EACH, and it is the
   whole reason that page is 11.8 screens on a phone while being 5.9 on
   a desktop. Nothing about a seven-word label needs 375px of width.

   A deck with eight or more items is a label deck, not an article, so
   it runs two-up below the breakpoint. Decks with fewer items hold real
   content and are left alone. :has() carries this because the rule is
   about the deck's own child count; where :has() is unsupported the
   deck simply stays one-up, which is today's behaviour. */
@media (max-width:1079px){
  html body main .m-cards:has(> :nth-child(8)){
    display:grid;grid-template-columns:1fr 1fr;gap:10px;align-items:start}
  html body main .m-cards:has(> :nth-child(8)) > *{min-width:0}
  html body main .m-cards:has(> :nth-child(8)) .m-cards__item p,
  html body main .m-cards:has(> :nth-child(8)) p{font-size:15px !important}
  html body main .m-cards:has(> :nth-child(8)) h3{font-size:16px !important}
}

/* -- 26. DIAGRAMS ON A PHONE -----------------------------------------
   JT, 2026-10-28: "the images with text are too small to read."

   MEASURED: every content figure renders at 335px inside the 375px
   viewport. The diagrams are 760px to 1,200px wide and carry labelled
   text, so at 335px the labels are between 28% and 44% of the size they
   were drawn at, which is illegible and not a matter of opinion.

   A phone cannot show a 1,200px diagram legibly and shrinking it further
   to "fit" is how it got here. So below the breakpoint the figure pans:
   the image holds a readable width and its container scrolls, which is
   the same treatment wide tables already get. The caption says so,
   because a scroll affordance nobody notices is not an affordance. */
@media (max-width:1079px){
  html body main figure:has(> img),
  html body main .m-media__fig,
  html body main .m-figure{
    overflow-x:auto;-webkit-overflow-scrolling:touch;
    scrollbar-width:thin;scroll-snap-type:x proximity}
  html body main figure > img,
  html body main .m-media__fig > img,
  html body main .m-figure > img{
    max-width:none;width:auto;min-width:min(760px,205vw);height:auto;
    scroll-snap-align:start}
  /* Not the promoted backdrops: those are cover-cropped grounds and
     panning one would just drag a texture sideways. */
  html body main .m-backdrop__bg,html body main .m-hero__bg{overflow:hidden}
  html body main .m-backdrop__bg > img,
  html body main .m-hero__bg > img{min-width:0;width:100%}
  html body main figure:has(> img) > figcaption::after{
    content:" Swipe the diagram sideways to read it.";
    color:var(--ink-dim,#3f4a5a)}
}

/* -- 27. THE 90-DEGREE CORNERS THAT WERE LEFT -------------------------
   JT: "still have 90 degree angles, where the 33's should be."
   The backdrop plate was added by promote-backdrops.py after the radius
   pass had already run, so it inherited nothing and shipped square at
   389x183 on every page carrying a promoted backdrop. */
html .m-backdrop__plate{border-radius:var(--r-panel,18px);
  border-top:3px solid var(--accent,#f59e0b)}
html .m-backdrop{border-radius:0}          /* full bleed, correctly square */
html .m-backdrop__bg{border-radius:0}

/* -- 28. THE LAST EIGHT SQUARE CORNERS AND FIVE SMALL FIGURES --------
   With the gate's two false-red classes removed, what is left is real
   and short. Measured 2026-10-28:

     /pricing/  .entry-card  290x203   r=0
     /pricing/  .time-card   594x350   r=0
     /          figure-heppner.webp        310px of 1440  (22%)
     /          backdrop-2-boundary.webp   587px of 2560  (23%)

   The pricing cards are the ones a buyer looks at hardest and they were
   the only square panels left on the site. The home figures are small
   because promote-backdrops.py excludes 'home' deliberately, to protect
   the approved v23 stage: the answer there is to let them use the band
   they are already in, not to restructure the page. */
html body main .entry-card,html body main .time-card{
  border-radius:var(--r-panel,18px)}

/* A figure in a prose band was taking its width from the text measure,
   which is 68ch, not from the band. Nothing about an image wants to be
   as wide as a line of text. */
html body main figure{max-width:none;margin-left:0;margin-right:0}
html body main figure > img{width:100%;max-width:none;height:auto}
html body main .is-long figure{column-span:all}

/* Section 26 keyed pan-to-read on <figure>, and the home page's lead
   image is a bare <img class="lead-fig"> in a grid cell. Same defect,
   same fix, one selector wider. */
@media (max-width:1079px){
  html body main .read-grid > .lead{
    overflow-x:auto;-webkit-overflow-scrolling:touch;scrollbar-width:thin}
  /* 760px of a 1,440px capture is 53%, just under the 55% bar, so the
     pan gave a figure you still could not read. 1,000px is 69% and
     matches what the answer captures use, which is the width at which
     the text inside these images actually resolves on a phone. */
  html body main img.lead-fig{
    width:auto;max-width:none;min-width:min(1000px,270vw);height:auto}
}

/* The last one: figure-heppner.webp at 682px of 1,440 (47%) on the home
   page, sitting in one column of .read-grid against the 55% bar. The
   column, not the image, is what is narrow. */
@media (min-width:1080px){
  html body main .read-grid{grid-template-columns:1.45fr 1fr;
    gap:clamp(28px,3vw,52px);align-items:start}
}

/* -- 29. MAKING THE FIELD VISIBLE ON THE HOME PAGE -------------------
   Section 20 put the citation field UNDER backdrop-1-word.webp, and
   that asset is an opaque full-bleed image: navy with the Word task
   pane composited into it. So the field was painted and then covered,
   and the home page still showed the flat navy JT has now named four
   times. Correct markup, invisible result.

   The field now sits ABOVE the pane image and composites with SCREEN.
   screen(a,b) = 1-(1-a)(1-b), which means:
     the field's near-black ground  -> changes nothing it sits on
     the field's citation text      -> lightens only the dark navy
     the white task pane            -> stays white, screen on white is
                                       white, so the pane JT called the
                                       best asset on the site is not
                                       touched

   That is why this is screen and not opacity: a plain alpha layer at
   any strength that made the navy interesting would also have greyed
   the pane. The headline is unaffected either way because .plate paints
   its own solid rgb(4,20,36) ground, which is also what the contrast
   audit measures against. */
html .stage-bg > .m-hero__bg{z-index:2;mix-blend-mode:screen;
  pointer-events:none}
html .stage-bg > .m-hero__bg img{opacity:.78;object-position:50% 42%}
/* The vignette was masking the left 34% at .92 and is no longer needed
   to protect the headline: the plate does that. Keep a light version so
   the top edge does not fight the masthead. */
html .stage-bg::after{
  background:linear-gradient(180deg,rgba(3,15,29,.55) 0%,
             rgba(3,15,29,0) 24%,rgba(3,15,29,0) 72%,
             rgba(3,15,29,.45) 100%);z-index:3}

/* The screen blend leaves faint citation text across the task pane,
   most visibly over the gold ANALYZE SELECTION button, where it reads
   as a rendering artefact rather than texture. The pane occupies
   roughly 47% to 82% of the stage width, so the field is masked out of
   that ellipse and fades back in around it. Mask, not a second
   gradient: a gradient would have to sit above the pane too, and that
   is the problem being solved. */
@media (min-width:1040px){
  html .stage-bg > .m-hero__bg{
    -webkit-mask-image:radial-gradient(62% 74% at 66% 58%,
      rgba(0,0,0,0) 0%,rgba(0,0,0,0) 34%,rgba(0,0,0,.55) 54%,
      rgba(0,0,0,1) 78%);
    mask-image:radial-gradient(62% 74% at 66% 58%,
      rgba(0,0,0,0) 0%,rgba(0,0,0,0) 34%,rgba(0,0,0,.55) 54%,
      rgba(0,0,0,1) 78%)}
}

/* -- 30. MOBILE DENSITY ----------------------------------------------
   Why this and not more folding: traced band by band, /pricing/ and the
   home page have NO long prose runs to fold. Their content is already
   chunked into tier cards, pieces and panels, so every fold candidate
   was rejected for having one or two paragraphs per host. The height on
   a phone is the COMPONENTS stacking, plus the larger type.

   The type increase is not negotiable, JT asked for it twice, and on a
   phone the same words at 18px instead of 13px are simply taller. What
   can come back is the space BETWEEN the components, which on a 375px
   screen was tuned for a desktop grid that no longer exists there. */
@media (max-width:1079px){
  html body main > .m,html body main > section{
    padding-top:clamp(18px,3.4vw,30px);
    padding-bottom:clamp(18px,3.4vw,30px)}
  html body main .m-cards{gap:9px}
  html body main .m-cards__item,html body main .tier-card,
  html body main .entry-card,html body main .time-card,
  html body main .panel,html body main .piece{
    padding:clamp(13px,3.4vw,18px)}
  html body main h2{margin-bottom:7px}
  html body main p{margin-bottom:9px}
  html body main .plate{padding:clamp(18px,4.4vw,26px)}
  html body main figure{margin-top:12px;margin-bottom:12px}
}

/* -- 31. ONE OWNER, AND NO ACCIDENTAL STRETCH ------------------------
   `align-items` was never set by any of the three rules that owned a
   deck, so it defaulted to `normal`, which in a grid means STRETCH.
   Measured: a 34-word card on /compare/ forced to the height of a
   158-word one, carrying 734px of dead space, and a 279-word card on
   /legal-ai-positioning-map/ stretched to 3,505px with 1,873px dead.

   Content decks size to their content. */
html .m-cards{align-items:start}
html .m-cards--2{grid-template-columns:repeat(2,minmax(0,1fr))}
html .m-cards--3{grid-template-columns:repeat(3,minmax(0,1fr))}
html .m-cards--4{grid-template-columns:repeat(4,minmax(0,1fr))}
@media(max-width:860px){
  html .m-cards--3,html .m-cards--4{grid-template-columns:repeat(2,minmax(0,1fr))}
}
@media(max-width:560px){
  html .m-cards--2,html .m-cards--3,html .m-cards--4{
    grid-template-columns:minmax(0,1fr)}
}

/* PRICING DECKS KEEP EQUAL HEIGHT, deliberately. `.tier-grid` set
   `align-items:stretch` by hand and that is correct: a price table reads
   as a row of comparable offers and ragged card bottoms break the
   comparison. Folding these into .m-cards would have inherited
   `align-items:start` and quietly changed the pricing page. They get
   their own module so the stretch is a stated decision rather than an
   accident, and frame-gate.py exempts them by name for that reason. */
html .m-tiers{display:grid;gap:16px;align-items:stretch;
  grid-template-columns:repeat(auto-fit,minmax(min(250px,100%),1fr))}
html .m-tiers__item{background:#fff;border:1px solid var(--rule-light);
  border-top:3px solid var(--teal);border-radius:var(--m-radius,18px);
  padding:clamp(20px,2.2vw,28px);display:flex;flex-direction:column;
  min-width:0}
html .m-tiers__item > :last-child{margin-top:auto}

/* -- 32. THE HOME LEAD FIGURE ----------------------------------------
   figure-heppner.webp is 1,440px wide and `.read-grid` gave it a 689px
   column, so it rendered at 48% against the 55% legibility bar, as the
   last standing item in audit-real.py for several builds.

   Widening the column cannot fix it: the image needs 792px to clear the
   bar and the band is 1,208px, which would leave the prose beside it in
   a 394px column, about 24 characters per line. One of the two has to
   be unreadable in a side-by-side at this width, so the band stacks and
   the figure takes the full 1,208px: 84% of intrinsic.

   Same decision as the answer cards, same reason. A two-column layout
   that makes its own evidence illegible defeats the evidence. */
html body main .read-grid{grid-template-columns:minmax(0,1fr);
  gap:clamp(16px,2vw,28px)}
html body main .read-grid > .lead{order:2}
html body main .read-grid > .lead img.lead-fig{width:100%;max-width:none;
  height:auto;border-radius:var(--m-radius,18px)}

/* ── the claim band's ground ───────────────────────────────────────────
   JT, 2026-10-29: "the large wall of light green, shuld be replaced with
   some other backdrop image."

   MEASURED over HTTP before touching anything. The wall is
   `section.band.arch-claim`, painted `var(--hue-soft)` by
   register-v23-repair.css:214, and on /how-it-works/ and /matter-memory/
   it is 1440 x 3756px of flat #f0fdf4.

   ⛔ `.m--tint` at line 64 above is NOT it. That rule is dead: zero built
   pages match it. I assumed it was the selector and generated an asset
   for it before measuring.

   WHY repeat-y AND NOT cover. The band is 3,756px tall. Any fixed image
   has to be stretched to five times its aspect ratio or stop a third of
   the way down. backdrop-reporter-page.webp is a 1600x1440 tile whose
   height is exactly 40 rows x 36px, so the ruled rhythm carries across
   the join and the band covers at any height.

   WHY max(100%, 1100px). Stretching to 100% alone puts 20px type at
   4.7px on a 375px phone, which is a smudge, not a page. Below 1100px
   the tile holds its size and overflows the right edge instead, and the
   section clips it. Legibility over fit.

   The asset carries ALPHA and no paper of its own, so the band keeps
   painting its own ground underneath. The accent survives as the top
   rule rather than as a wall, which is what removes the green while
   leaving /pricing/ and /compare/ still telling themselves apart.

   ⛔ `html[data-accent]`, ATTACHED, NOT `html [data-accent]`. The first
   version had the space and matched nothing on any page, because
   data-accent is set on the <html> element itself and a descendant
   combinator will not select the element it descends from. It looked
   like it had applied: the browser reported the rule as matching,
   because the comma group's OTHER selector matched and
   `el.matches(group)` is true if ANY branch is. Every band still
   computed background-image:none and I had to query the cascade to see
   why.

   The specificity is the other half. register-v23-repair.css:214 is
   `[data-accent] main>section.arch-claim`, which is (0,2,2) and uses the
   `background` SHORTHAND, so it resets background-image to none. The
   dead branch `html main > section.arch-claim` is (0,1,3) and loses to
   it. `html[data-accent]...` is (0,2,3) and wins.

   ⛔ THE :not() IS LOAD-BEARING. Nine claim bands are deliberately DARK
   (#041424) via assets/treatments/dossier.css, which is (0,2,2). At
   (0,2,3) this rule would have beaten it and painted paper under the
   white body copy on /heppner/, /westlaw-alternative/ and seven more.
   dossier is the only treatment of the seven in use that touches
   .arch-claim; that was checked, not assumed. */
html[data-accent]:not([data-treatment="dossier"]) main > section.arch-claim{
  background-color:var(--paper-claim,#f5f7f6);
  background-image:url("/assets/backdrop-reporter-page.webp");
  background-repeat:repeat-y;
  background-position:top center;
  background-size:100% auto;
  background-size:max(100%,1100px) auto;
  border-top:3px solid var(--hue)}

/* ── headings inside a dark plate inherit the plate's colour ──────────
   FOUND BY audit-real, not by reading: five home page headings at
   36.8px computed rgb(11,18,32) on rgb(4,20,36), a ratio of 1.01.

   `register-v23.css` already paints the plate correctly:
   `.over{color:var(--on-deep)}`, which is #e8eaf0. The h2 inside it does
   not inherit that, because `register-v23-interior.css` says
   `main h2{color:var(--ink)}` and --ink is #0b1220 globally.

   The rule that was supposed to win is `.plate h1,.plate h2` in
   components.css at (0,1,1), which outranks `main h2` at (0,0,2).
   ⛔ components.css IS LINKED ON ZERO BUILT PAGES. It is copied into
   the site by build.py step 2 and then nothing references it, so the
   rule that makes the home page's statement headings legible has never
   once been applied. Specificity was never the problem; the sheet was
   simply absent.

   ⛔ THIS WAS INVISIBLE BECAUSE surface.js WAS HIDING IT. The repair
   pass measures the painted ground and rewrites the colour, so by the
   time anything looks at the page the headings are light. Blocking
   surface.js and re-measuring is what showed the authored state: 19
   nodes at #0b1220, one of them on #041424. A user sees that for one
   paint. Relying on a JS repair to make a headline visible means the
   headline is invisible until the JS runs.

   Scoped to inherit rather than to a named colour, because colour
   belongs to the measured surface. Checked first: all 10 .over/.plate
   elements across the site sit on a dark ground, none on a light one,
   so there is no case where inheriting is wrong. */
html main .over h1, html main .over h2, html main .over h3,
html main .plate h1, html main .plate h2, html main .plate h3,
html main .deck-head h1, html main .deck-head h2, html main .deck-head h3,
html main section.solid h1, html main section.solid h2,
html main section.solid h3{
  color:inherit}

/* ⛔ TWO MORE OF THESE TURNED UP AT 375px AND I HAD ONLY FIXED THE TWO
   SELECTORS I HAPPENED TO MEASURE. `.deck-head` and `section.solid`
   carry the same authored #0b1220 on a dark ground, caught again by
   audit-real at a ratio of 1.01 and 1.18, and again hidden in the live
   page because surface.js repairs it after paint. The lesson is the
   same one as the social links: I fixed the instances I had measured
   and reported the class of defect as closed. The check that finds
   these is blocking surface.js and reading the AUTHORED colour, not
   looking at the rendered page. */

/* ── a nested <span> is not a section wrapper ─────────────────────────
   JT, 2026-10-29, phone screenshots of /compare/: the competitor list
   rendered as a dark pill with the title colliding into it, and the same
   shape appeared on the comparison tables.

   MEASURED CAUSE, register-v23.css:198

       .band span{display:block;width:100%;max-width:var(--max);
                  margin:0 auto;padding:14px var(--gut);
                  font-size:15px;font-weight:700}

   That rule was written for the notification banner, `p.band`, whose
   single child span is a full-width bar. But `.band` is ALSO the class
   on every content section, `<section class="band arch-claim m">`, so
   the selector reaches EVERY span nested anywhere inside any band.

   Counted over the built site at 390px:
       3   direct-child spans, which is what the rule was written for
       58  nested spans caught by the same selector
       56  of those computing display:block, wrongly

   On /compare/ that turns `<a><span class="k">Thomson Reuters</span>
   Westlaw alternative...</a>` into a flex container with a 196px block
   beside a 154px text column, which is the collision in the screenshot.
   It is not a mobile bug: it reproduces identically at 1440.

   ⛔ TWO UNRELATED COMPONENTS SHARE THE CLASS NAME `.band`. The comment
   above that rule records scoping the BACKGROUND to `p.band` for exactly
   this reason, and then left the span rule unscoped. Half a fix.

   ⛔ ONLY LAYOUT IS RESET HERE. Typography is deliberately untouched,
   because `html body main .cite-line` already owns font-size and
   letter-spacing for the eyebrows and this selector is more specific
   than it is. Setting font-size here would silently take ownership of
   every eyebrow on the site. The token pills get their own authored
   type restored explicitly, because `.band span` had been overriding
   `.tag`'s 12px with 15px/700 and nothing else would put it back. */
html .band:not(p) :not(.band) > span{
  display:inline;width:auto;max-width:none;margin:0;padding:0}

html .band:not(p) :not(.band) > span.tag,
html .band:not(p) :not(.band) > span.tg{
  display:inline-block;padding:5px 9px;margin-bottom:11px;
  font-family:var(--display);font-size:12px;letter-spacing:.02em;
  font-weight:inherit}

html .band:not(p) :not(.band) > span.k,
html .band:not(p) :not(.band) > span.gr,
html .band:not(p) :not(.band) > span.y,
html .band:not(p) :not(.band) > span.n,
html .band:not(p) :not(.band) > span.treatise-tag{
  display:inline-block;padding:3px 9px;
  font-family:var(--sans);font-size:12px;font-weight:700;
  letter-spacing:.04em}

html .band:not(p) :not(.band) > span.cite-line{display:block}

/* ── the competitor link list reads as two lines, not two columns ─────
   Follow-on from the .band span fix above, and a correction to my first
   reading of it.

   ⛔ THE REMAINING display:block ON `.k` IS CORRECT CSS, NOT A LOSS IN
   THE CASCADE. I re-queried the cascade three times trying to work out
   why a (0,3,3) rule was losing to a (0,1,1) one. It was not losing.
   `.rellinks a` is `display:inline-flex`, so every child is a FLEX ITEM,
   and a flex item's display is blockified by the spec: inline-block
   computes to block. Raising specificity further would never have moved
   it. The container was the problem the whole time.

   So the link becomes a column: the kicker on its own line, the title
   under it. That is what the markup already says
   (`<a><span class="k">RELX</span>LexisNexis and Shepard's alternative</a>`)
   and what the inline-flex row was preventing.

   ⛔ AND THE KICKER LOSES THE DARK PILL ON A LIGHT BAND.
   register-v23-interior.css gives `.k` `background:var(--deep)` and says
   in its own header that the triage styles are "dark-surface only,
   because green 2.28, yellow 1.92 and amber 2.15 all fail 3:1 on white".
   `/compare/` is a light band, so the pill was a dark slab behind a
   kicker. Here it is set as the quiet eyebrow the rest of the site
   already uses. Colour itself still belongs to the measured surface. */
html .rellinks{align-items:stretch}
html .rellinks a{display:flex;flex-direction:column;align-items:flex-start;
  justify-content:center;gap:3px;min-height:44px;
  flex:0 1 min(100%,var(--rel-col,360px))}
html .band:not(p) .rellinks .k{background:transparent;padding:0;
  font-family:var(--sans);font-size:11.5px;font-weight:700;
  letter-spacing:.06em;text-transform:uppercase}

/* ── the Enterprise tier spans the row it sits on ─────────────────────
   JT, 2026-10-29: "on the pricing page. make the custom pricing take up
   the entire width under the other 4 cards. its lopsided now."

   MEASURED: `.m-tiers` is auto-fit at minmax(min(250px,100%),1fr) in a
   1,208px band, which resolves to FOUR columns. Verify, Small, Medium
   and Large fill row one; Enterprise lands alone in row two at a quarter
   width with three empty cells beside it. That is the lopsided row.

   grid-column:1/-1 spans whatever the column count resolved to, so it
   stays correct at every breakpoint instead of being pinned to four.

   On a wide viewport the card also turns from a column into a banner
   row, because a 1,208px card holding a 4-item list as a vertical stack
   is mostly empty space, which is the defect JT has already flagged
   twice on other bands. Below 900px it stays a column like its
   siblings.

   ⛔ `:last-child` NOT `:nth-child(5)`. apply-pricing.py rebuilds the
   deck's children, and a positional selector silently stops matching
   the moment a tier is added or removed. Enterprise is defined by being
   last, so the selector says that. */
html .m-tiers__item:last-child{grid-column:1 / -1}

@media (min-width:900px){
  html .m-tiers__item:last-child{
    flex-direction:row;flex-wrap:wrap;align-items:center;
    gap:clamp(14px,2vw,32px)}
  html .m-tiers__item:last-child > *{margin:0}
  html .m-tiers__item:last-child .tier-name{flex:0 0 auto}
  html .m-tiers__item:last-child .tier-price{flex:0 0 auto}
  html .m-tiers__item:last-child .tier-unit{flex:1 1 260px;max-width:40ch}
  html .m-tiers__item:last-child > p:not([class]){flex:1 1 240px;
    max-width:38ch}
  html .m-tiers__item:last-child .tier-features{flex:1 1 100%;
    display:flex;flex-wrap:wrap;gap:4px 26px;margin:0;padding:0}
  html .m-tiers__item:last-child > :last-child{margin-top:0;
    margin-left:auto;flex:0 0 auto}
}

/* ── the acknowledgement boxes are a 44px target ──────────────────────
   gyi-gate, on /legal-citation-checker/: "controls under 44px
   ['? h=24' x4]", at both 390 and desktop. The four `.ack-box` inputs
   render at the UA default 24x24. WCAG 2.5.8 wants 24 minimum and the
   persona standard here is 44, which is the figure the rest of the site
   is held to.

   ⛔ THIS IS ON THE LEAD CONVERSION SURFACE. Every funnel path now ends
   at this page, and these four boxes are the gate in front of the ask.
   A target a thumb misses on the page where the reader finally commits
   is the most expensive 20px on the site.

   The input is sized to 44 and given a negative margin so the surrounding
   text does not move: the TARGET grows, the layout does not. appearance
   is reset so the drawn box stays the size it was rather than becoming a
   44px native control, and the check is drawn rather than inherited,
   because a native tick scaled to 44px looks like a mistake. */
/* ⛔ THE SELECTOR HAS TO CLEAR (0,3,2). funnel.css owns these with
   `[data-treatment="funnel"] main .ack input[type="checkbox"]`, which is
   (0,3,2) and sets 24x24. `html .ack-box` is (0,1,1) and lost silently:
   the rule parsed, appearance:none applied, and width stayed 24px.

   ⛔ `html[data-treatment]` ATTACHED. My second attempt was
   `html body [data-treatment] main ...`, which matches NOTHING, because
   data-treatment is set on the <html> element itself and a descendant
   combinator cannot select the element it descends from. That is the
   SAME error I made earlier the same day on `html [data-accent]` for
   the claim band ground, and it fails the same quiet way: the rule
   parses, appears in the stylesheet, and simply never matches. When an
   attribute lives on <html>, the selector is html[attr], never
   html [attr]. */
html[data-treatment] body main .ack .ack-box{
  appearance:none;-webkit-appearance:none;
  width:44px;height:44px;margin:-10px;flex:0 0 auto;
  display:inline-grid;place-content:center;
  background:none;border:0;cursor:pointer}
/* ⛔ THE BORDER COLOUR IS LITERAL AND MUST NOT BE A TOKEN. JT, 2026-10-30:
   the boxes are "faint against the white, i didnt know they wer there."

   MEASURED in the browser, not read off the CSS. This rule used to say
   `border:2px solid var(--rule-deep,#9aa4b2)`, which reads like a mid grey.
   The PAINTED value on this page was rgba(255,255,255,0.14): white at 14%
   opacity, on a white card. 1.00:1. Not faint, INVISIBLE.

   The fallback is what hid it. `#9aa4b2` only applies when --rule-deep is
   undefined, and the page runs data-treatment="funnel", which defines it as
   a near-transparent white for rules on DARK bands. One token, two jobs,
   exactly like the --on-deep-muted defect recorded further down this file.

   #6b7280 on #fff measures 4.83:1. WCAG 1.4.11 asks 3:1 for the boundary of
   a user interface component, and a checkbox nobody can see is a consent
   gate nobody knowingly passes.

   ⛔ A CONTRAST GATE CANNOT CATCH THIS. Every such gate here measures TEXT
   against its ground. This is a border on a pseudo-element, so it is not
   text, and all four site gates passed the invisible version. */
html[data-treatment] body main .ack .ack-box::before{
  content:"";width:22px;height:22px;border-radius:5px;
  border:2px solid #6b7280;
  background:#fff;display:block;
  transition:background .12s ease,border-color .12s ease}
html[data-treatment] body main .ack .ack-box:checked::before{
  background:var(--hue-ink,#15803d);border-color:var(--hue-ink,#15803d);
  /* the tick, drawn so it scales with the box rather than the control */
  background-image:url("data:image/svg+xml;utf8,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='white' stroke-width='3.2' stroke-linecap='round' stroke-linejoin='round'><path d='M5 12.5l4.5 4.5L19 7.5'/></svg>");
  background-size:16px 16px;background-position:center;
  background-repeat:no-repeat}
html[data-treatment] body main .ack .ack-box:focus-visible::before{
  outline:2px solid var(--accent,#f59e0b);outline-offset:2px}

/* ── the 12-column matrix, reverted and then left alone ──────────────
   ⛔ MY BREAKOUT MADE IT WORSE AND JT'S SCREENSHOT IS THE PROOF.
   I widened the wrapper to the band and tightened the cells, which did
   cut the overflow from 885px to 333 and the section from 5,050px to
   4,480. Both numbers improved. The PAGE got worse: the dark badge
   pills (.gr/.y/.n) sit on their own line inside cells that are now
   narrower per column, so they overlap neighbouring text, and the
   result is the collision JT photographed.

   Two numbers improving while the thing becomes unreadable is the
   whole reason to look at a render instead of a measurement. The
   breakout and the cell tightening are removed; the table goes back to
   its own scroll container, which is what it was designed for.

   A 12 by 14 matrix of badge cells does not belong in a reading column
   at any width, and the honest fix is to present it as per-vendor
   blocks rather than a grid. That is a content change on a sourced,
   sealed document, so it is JT's call, not a CSS patch. */

/* ── a secondary button has to LOOK secondary ─────────────────────────
   audit-real, after the tier buttons were demoted: "gold used as a
   field, /pricing/ 1150x48 rgb(245,158,11) A.btn btn-secondary".

   TWO SEPARATE FAULTS, AND THE FIRST ONE FOOLED A GATE.

   1. `html body main a.btn` already paints EVERY a.btn with
      background:var(--gold-rule). `.btn-secondary` sets no background of
      its own, so changing the class changed the class and nothing else:
      all five tier buttons stayed solid gold. gyi-gate then reported
      "one primary CTA" and passed, because it reads the CLASS NAME and
      the page still had five identical gold slabs. A gate that measures
      the label instead of the paint will approve a page that looks
      exactly like the one it just rejected.

   2. `.tier-card .btn{width:100%}` plus the full-width Enterprise card
      makes one of them 1,150px wide. 1,150 x 48 of pure accent is the
      definition of the thing the gold-field check exists to catch, and
      it is the SECOND time that width rule has done this on this page.

   So secondary becomes an outline, and the width is capped inside the
   full-width card only. The accent stays on exactly one button. */
html body main .m-tiers__item a.btn.btn-secondary{
  background:transparent;color:var(--ink,#0b1220);
  border:1.5px solid var(--hue-ink,#15803d)}
html body main .m-tiers__item a.btn.btn-secondary:hover{
  background:var(--hue-soft,#f0fdf4)}
html body main .m-tiers__item:last-child a.btn{
  width:auto;min-width:min(320px,100%);align-self:flex-start}

/* ── two shapes of card grid, because there are two problems ──────────
   Set by harness/build-card-rows.py. See that file for the measurement.

   --pair  a comparison. Two boxes read across, so they match height.
   --rows  a list of findings. Not a comparison, so not boxes: each item
           is a row, and a longer item is a taller row rather than the
           cause of a ragged grid. */
html body main .m-cards--pair{align-items:stretch}
html body main .m-cards--pair > .m-cards__item{height:100%}

html body main .m-cards--rows{grid-template-columns:minmax(0,1fr);
  gap:0}
/* ⛔ THE SEPARATOR IS A PSEUDO-ELEMENT, NOT A BORDER, AND THAT IS NOT
   COSMETIC. audit-real's square-corner check exempts anything that
   "paints nothing", measured off the computed style, and a border-top
   counts as painting. So a row with a 1px top rule was being counted as
   a zero-radius BOX, 32 of them, and the square count went from 0 to 32
   the moment these rows shipped. A row has no observable corner; the
   rule is a line between items, not the outline of a card. Drawing it
   with ::before keeps the line and leaves the item painting nothing,
   which is what it honestly does. */
html body main .m-cards--rows > .m-cards__item{
  border:0;border-radius:0;padding:clamp(16px,2vw,22px) 0;
  background:transparent;position:relative}
html body main .m-cards--rows > .m-cards__item::before{
  content:"";position:absolute;top:0;left:0;right:0;height:1px;
  background:var(--rule-light,#e5e7eb)}
html body main .m-cards--rows > .m-cards__item:first-child::before{
  display:none}
@media (min-width:1000px){
  html body main .m-cards--rows > .m-cards__item{
    display:grid;grid-template-columns:minmax(220px,0.9fr) minmax(0,2fr);
    column-gap:clamp(20px,2.6vw,42px);align-items:start}
  /* the finding's name becomes the rail, the body sits beside it */
  html body main .m-cards--rows > .m-cards__item > :first-child{
    grid-column:1;margin:0}
  html body main .m-cards--rows > .m-cards__item > :first-child ~ *{
    grid-column:2}
}
html body main .m-cards--rows .fname{font-family:var(--sans);
  font-size:15.5px;font-weight:700;line-height:1.3}

/* ── the footer has four columns now, so say four ─────────────────────
   JT, 2026-10-29, screenshot of the live footer: the brand tagline
   overlapping "Attorneys", the Company column dropped to a second row,
   and the whole block squeezed into the middle third of the page.

   MY REGRESSION, and a one-line one. link-orphan-pages.py added a
   fourth column, "Who it is for", to a grid that exhibit.css:312
   declares as `repeat(3,minmax(0,1fr))`. The fourth child wrapped onto
   a second row, and the 4fr/8fr split of .colophon-in left the link
   area 784px wide for what is now four columns of 249px plus the brand
   column's tagline running past its own 392px track.

   Adding a column to a grid means changing the grid. I added the
   column and shipped. */
@media (min-width:900px){
  html body .colophon-in:has(.col-cols > :nth-child(4)){
    grid-template-columns:minmax(0,3fr) minmax(0,9fr)}
  html body .col-cols:has(> :nth-child(4)){
    grid-template-columns:repeat(4,minmax(0,1fr));gap:18px}
}
@media (min-width:620px) and (max-width:899px){
  html body .col-cols:has(> :nth-child(4)){
    grid-template-columns:repeat(2,minmax(0,1fr))}
}
/* The tagline was running out of its own track and over the next
   column. A grid child needs min-width:0 to be allowed to shrink. */
html body .colophon-in > *{min-width:0}
html body .col-brand .mast-sub{overflow-wrap:anywhere}

/* ── the first screen on a phone, tightened ──────────────────────────
   JT, 2026-10-30: "Tighten it."

   MEASURED at 390x844 before this. The headline did not start until
   284px, a third of the viewport spent before a word reads:

       masthead               197px
       .stage padding-top      18
       .over  padding-top      48
       .plate padding-top      18
       = 281, H1 line box at 284

   ⛔ THE NAV IS LEFT ALONE AND THAT IS DELIBERATE. It is 88px of the
   197, two rows of 44px links, and 44 is the tap-target minimum the
   gate enforces. Collapsing it to one scrolling row would save 44px
   and push "Heppner" off the right edge, which gyi-gate requires to be
   above the fold. Trading a measured accessibility floor and a
   content rule for 44 pixels is a bad trade.

   So the saving comes from padding and from one redundant line: the
   masthead tagline reads "Legal Intelligence. Verified." directly above
   an H1 that now says "Legal research that cannot fabricate a
   citation." It says the same thing twice, so on a phone it says it
   once. It stays on desktop where the room is free. */
@media (max-width:620px){
  html body .mast .tagline,
  html body header .tagline{display:none}
  html body main .stage{padding-top:8px}
  html body main .stage > .over{padding-top:18px}
  html body main .plate{padding-top:12px}
}

/* -- 46. FULL-BLEED BANDS ARE SQUARE, DARK-GROUND BODY COPY IS READABLE --
   JT, 2026-10-30, from a phone, with the corners circled in red:
   "Get rid of these white corner bleed through." and "The top text in
   almost all pages is a blue on lighter blue text. Its very hard for me
   to read that. Update the contrast. The gate should have caught that."

   DEFECT 1, THE WHITE NOTCHES. register-v23-repair.css section 2 rounds
   every box on the site from one token, and its selector list includes
   `.deck`. `.deck` is not a box, it is a FULL-BLEED BAND: measured at a
   390px viewport it is 390px wide and begins at x=0. Rounding it cuts
   four notches out of the page edge and the page ground shows through
   them. That is the white wedge in the screenshot.

   Measured across all 40 pages: `.solid.deck` is the ONLY full-bleed
   band carrying a radius, and NO direct child of <main> is ever inset.
   So the rule is categorical rather than a patch for one band, and the
   next selector added to that repair list cannot reintroduce it.

   DEFECT 2, AND WHY THE GATE WAS RIGHT TO PASS IT. surface.js marks
   every text element with its measured painted ground, and the repair
   sends everything that is not a heading, b, strong or cite to
   --on-deep-muted. #9ca3b0 on #041424 measures 7.32:1. That passes
   WCAG AA at 4.5 and AAA at 7, so a contrast gate CANNOT fail it, and
   the audit reporting zero contrast failures was accurate. The defect
   is that one token was doing two jobs: subordinate chrome (nav links,
   slide counts, input placeholders) and PRIMARY BODY COPY on a dark
   band. Body copy is not chrome. It gets its own token at 13.08:1,
   which still sits below the 15.44:1 of the headings above it, so the
   hierarchy survives, carried by size and weight instead of by making
   the sentence hard to read.

   The gate that now exists measures BODY COPY on dark grounds against
   9:1, not against the AA 4.5 that passed this. See
   harness/gate-band-corners.py.

   STOP: --on-deep-muted IS NOT RAISED. Changing the token would also
   lighten the masthead nav, the deck counter and the input
   placeholders, which are deliberately quiet and are not what JT was
   reading.

   SPECIFICITY, MEASURED NOT ASSUMED. `html body main > *` is (0,0,4) and
   LOST to the repair sheet's bare `.deck` at (0,1,0): one class outweighs
   any number of type selectors. The band still measured radius=16px after
   the first attempt. `> [class]` carries the class the comparison needs,
   giving (0,1,3), and every band on the site carries a class. */
:root{--on-deep-body:#d2d9e4}
html body main > [class]{border-radius:0}
html body main p[data-on="dark"],html body main li[data-on="dark"],
html body main td[data-on="dark"],html body main dd[data-on="dark"],
html body main blockquote[data-on="dark"],html body main .lede[data-on="dark"],
html body main .after[data-on="dark"],html body main .arch-lede[data-on="dark"],
html body main .hc-lede[data-on="dark"]{color:var(--on-deep-body)}

/* -- 47. A FIGURE THAT ALREADY FITS MUST NOT BE STRETCHED INTO A SCROLL --
   JT, 2026-10-30: "L R Scrolling the images on mobile is no good."

   Section 26 above pans every content figure on a phone, pinning the
   image to min(760px,205vw), and its reasoning is correct for what it
   was written against: a 1,200px diagram with 11px labels rendered at
   335px puts those labels at 28% of the size they were drawn, which is
   illegible whatever anyone prefers. Shrinking is not an option there,
   so it pans and the caption says so.

   But the pin is unconditional, so it also applies to a figure drawn at
   360px to fit a phone in the first place. min-width then STRETCHES a
   figure that already fitted until its container scrolls, manufacturing
   the exact defect JT is reporting out of a figure that did not have it.

   The real fix is upstream and is now done: the two figures on /for-it/
   and /for-security/ are redrawn phone-first at 360px wide
   (make-process-figures.py, stack()). This rule is what lets a
   phone-first figure keep its size. place-process-figures.py adds the
   class by READING the width the figure was drawn at, so it is a
   measurement rather than a list that will drift.

   The swipe affordance is removed with it. A caption that says "Swipe
   the diagram sideways to read it" under a figure that does not move is
   worse than no caption. */
@media (max-width:1079px){
  html body main .m-figure--fits{overflow-x:visible;scroll-snap-type:none}
  html body main .m-figure--fits > img{
    min-width:0;width:100%;max-width:100%;height:auto}
  html body main .m-figure--fits > figcaption::after{content:"" !important}
}

/* -- 48. A TEAMS CAPTURE THAT CARRIES ITS OWN TRANSCRIPT STOPS PANNING --
   JT, 2026-10-30: "For the teams display, figure a way to have it
   viewable without a LR scroll."

   mechanisms.css pins .m-answer captures to a 1,100px minimum on a phone
   and pans them, and its reasoning holds: those files are 1,773 to
   1,834px wide, and at 350px their body text renders at 19% of the size
   it was drawn. Shrinking to fit does not make them readable, it makes
   them decorative.

   What changes the trade is that the words are no longer only in the
   image. transcribe-answer-captures.py puts a verbatim transcript under
   each capture, so the image no longer has to carry the reading. It is
   fitted to the column as evidence that this is the product running in
   Microsoft Teams, and the transcript is the readable path. That also
   puts the strongest claims on this site into text, where a retrieval
   model can reach them; the alt attribute on these four was 70
   characters.

   Keyed on .has-tx, which the build step adds to the same figure it puts
   the transcript in, so a capture can never lose its pan treatment
   without having gained a transcript first. */
@media (max-width:1079px){
  html body main .m-answer figure.has-tx{
    overflow-x:visible;scroll-snap-type:none}
  html body main .m-answer figure.has-tx > img{
    min-width:0;width:100%;max-width:100%;height:auto}
  html body main .m-answer figure.has-tx > figcaption{
    position:static;max-width:none}
  html body main .m-answer figure.has-tx > figcaption::after{
    content:"" !important}
}
html body main .answer-tx{margin:12px 0 0;border:1px solid var(--rule-light);
  border-radius:var(--m-radius,16px);background:#fff}
html body main .answer-tx > summary{
  cursor:pointer;padding:12px 16px;min-height:44px;display:flex;
  align-items:center;font-family:var(--sans);font-size:15px;
  font-weight:600;color:var(--ink)}
html body main .answer-tx__in{padding:0 16px 16px}
html body main .answer-tx__in p{font-size:15px;line-height:1.55;
  margin:0 0 10px;max-width:62ch}
html body main .answer-tx__q{padding:10px 12px;background:#f6f8fa;
  border-radius:10px}
html body main .answer-tx__h{font-family:var(--sans);font-size:12px;
  font-weight:700;letter-spacing:.08em;color:var(--ink-dim);margin:14px 0 6px}
html body main .answer-tx__list{margin:0 0 10px 18px}
html body main .answer-tx__list li{font-size:15px;line-height:1.55}
html body main .answer-tx__src{font-size:13px;color:var(--ink-dim)}

/* -- 50. THE HOME PAGE LEAD FIGURE, WHICH IS NOT IN A <figure> ----------
   figure-heppner.webp is a bare <img class="lead-fig"> in a grid cell,
   so section 49 cannot reach it: that rule keys on figure.has-phone and
   there is no <figure> here. Section 26's note explains why it is pinned
   to min(1000px,270vw) rather than 760: at 760 a 1,440px raster is 53%
   of its drawn size and still unreadable. Correct, and the result is
   that the first figure on the site scrolls 650px sideways on a phone.

   build-lead-phone-figure.py draws the same four rows stacked at 360
   wide, reading the words out of build-backdrops.py so the two cannot
   disagree, and swap-phone-figures.py serves it below 620px. :has() is
   the hook because <picture> is only present when that swap actually
   happened. The site already relies on :has() in section 26. */
@media (max-width:620px){
  html body main .read-grid > .lead:has(picture){
    overflow-x:visible;scrollbar-width:auto}
  html body main .read-grid > .lead picture > img.lead-fig,
  html body main picture > img.lead-fig{
    min-width:0;width:100%;max-width:100%;height:auto}
}

/* -- 51. THE has-phone OVERFLOW OVERRIDE, STATED AS AN OVERRIDE ---------
   Section 49 sets overflow-x:visible on figure.has-phone and one figure
   on /legal-research-software/ still measured overflow-x:auto with a
   49px scroll. Specificity is not the problem: that rule is (0,1,4)
   against section 26's (0,1,3). A later sheet in the cascade is setting
   it again at equal weight, and chasing which one in which load order is
   exactly the per selector hunt that oscillated the contrast numbers for
   six passes without converging.

   So this is declared as what it is: a deliberate final word on one
   property, for figures that have been PROVEN to carry a phone-shaped
   file by swap-phone-figures.py reading it off disk. It cannot leak to a
   figure that still needs to pan, because such a figure never gets the
   class. */
@media (max-width:620px){
  html body main figure.has-phone > picture{display:block;max-width:100%}
}

/* ⛔ REVERTED, AND THE REASON IS WORTH MORE THAN THE FIX WAS.
   The rule here was `overflow-x:visible !important` on figure.has-phone.
   It removed the 49px inner scroll and replaced it with something worse:
   with nothing clipping it, that same 399px of content bled 29px past
   the viewport and THE WHOLE PAGE scrolled sideways. mobile-gate caught
   it on /legal-research-software/ immediately after the change, having
   been clean before it.

   A contained scroll inside one figure is a nuisance. A document that
   scrolls sideways is the defect JT reported in the first place, applied
   to every band on the page at once. So the figure keeps its clipping
   and keeps a 49px inner scroll on that one page, which is the smaller
   of the two.

   The 399px itself is still unexplained: every child of that figure
   measures 350 or less. Do not re-attempt this by forcing overflow until
   that is understood, because forcing it is what caused the bleed. */

/* -- 52. THE PROVENANCE ROW ON THE THREAT REPORT HAD NO STYLES AT ALL ---
   JT, 2026-10-30: "Spacing is off on the sec report."

   build-threat-landscape-page.py emits <div class="meta-row"> holding
   four <span><strong>label</strong> value</span> pairs. Searched every
   stylesheet: `.meta-row` appears in NO css file on this site. It was
   authored expecting a style that was never written, so four label and
   value pairs rendered as one unbroken inline run:

     "Prepared by Technology Outlaws LLC Source CrowdStrike 2026 Legal
      Threat Landscape Report, 28 October 2026 Read on 29 October 2026
      Status Self-produced analysis"

   On a phone that is a wall, and it is the wall carrying the page's
   provenance: who wrote this, whose report it reads, and the date it was
   read. That block is the reason the page is allowed to make comparative
   claims at all, so it is the last thing that should be hard to parse.

   Stacked pairs on a phone, a wrapping row of columns from 640px. The
   label becomes a caption above its value rather than a bold word
   running into it. */
html body main .meta-row{display:grid;gap:14px;
  grid-template-columns:minmax(0,1fr);margin:0}
html body main .meta-row > span{display:block;min-width:0;
  font-size:14.5px;line-height:1.5}
/* ⛔ 12.5px, NOT 11px. These four labels are VISIBLE AT EVERY WIDTH, so
   they are phone copy, and harness/mobile-sweep.py fails any visible text
   under 12px. Measured 2026-10-30 at 320, 390 and 768: the 11px version
   put four sub-floor elements on /legal-threat-landscape-2026/ and
   /legal-ai-positioning-map/, the only two pages using .meta-row. An
   uppercase eyebrow label is the easiest place to let this slide, because
   it LOOKS like chrome. It is read, so it is copy. */
html body main .meta-row > span > strong{display:block;
  font-family:var(--sans);font-size:12.5px;font-weight:700;
  letter-spacing:.07em;text-transform:uppercase;
  color:var(--ink-dim);margin-bottom:3px}
@media (min-width:640px){
  html body main .meta-row{
    grid-template-columns:repeat(auto-fit,minmax(190px,1fr));gap:18px 26px}
}
