/* ===========================================================================
   ANNEAL — polish layer

   Enqueued LAST, after every engine and vertical sheet, by the ms_enqueue_chain
   filter in inc/vertical/anneal.php.

   WHAT IS *NOT* HERE, AND WHY THAT MATTERS

   This file used to be twice this size. On the live site it also carried the
   whole token layer — every primitive, every semantic role, the ramp and the
   floods — because it was an OVERLAY bolted on top of another brand's tokens.
   In this repo ANNEAL is a brand, so all of that is generated from
   brands/anneal/tokens/*.tokens.json into tokens.css and theme.json, and the
   overlay half simply does not exist. Nothing below re-declares a token value.

   What remains is only what a token cannot express: rules that override a
   CLASS the engine or the vertical already styles. Each one is a documented
   consequence of brand law, not a preference.
   =========================================================================== */

/* -- SHAPE AND ELEVATION · the rules that live outside the token names --------
   A handful of components draw their own radius or shadow rather than taking a
   token. Elevation in this system is a hairline plus a surface change, and the
   only shadow in the brand is the contact shadow baked into a photograph — the
   object grounding itself, which is a different thing. -------------------- */
:root {
  --radius: 0;
  --radius-sm: 0;
  --radius-md: 0;
  --radius-lg: 0;
  --radius-xl: 0;
  --radius-2xl: 0;
  --shadow-brand: none;
  --shadow-glass-top: none;

  /* THE SHADOWS COME BACK IF THIS IS NOT HERE. tokens.css nulls the elevation
     roles for this brand, and then verticals/skincare/assets/components.css
     re-declares --elevation-lift and --elevation-pop as real drop shadows,
     unconditionally, on :root. It loads after the tokens and before this file,
     so on every ANNEAL page the ramp cards, the flood cards and the set card
     were lifting on hover with a shadow underneath — against §1.4, "No drop
     shadows: elevation is a value change plus a 1px border", on a brand whose
     only legitimate shadow is the contact shadow inside a photograph. Nulled
     again here, last, where nothing can re-declare them. */
  --elevation-lift: none;
  --elevation-pop: none;
  --elevation-card: none;
  --elevation-whisper: none;
  --shadow-sm: none;
  --shadow-lg: none;
}

/* NOT EVERY box-shadow IS A DROP SHADOW, and this list learned that the hard
   way. `.flood` was added here with .rampcard and .sk-set, and its shadow is an
   INSET RING — the card's only edge, since it carries no border. Killing it
   left the starter cards with no boundary at all: measured on the deployed
   page, box-shadow `none` and border-width `0px`. §1.4 bans the drop shadow
   and prescribes the hairline; an inset ring IS the hairline, so it stays.
   .rampcard, .sk-set and .shopcard genuinely do draw outer shadows on hover,
   which is why those three belong. */
.card, .ms-card, .shopcard, .panel, .ms-panel, .sk-bento > *,
.ms-topbar, .ms-header, .site-header, .ms-drawer, .ms-modal,
.rampcard, .sk-set {
  box-shadow: none;
}

/* -- THE SURFACE LADDER · why the page looked grey --------------------------
   Measured on the deployed page: #DEDFE4 covered 6,473,000 px² — twenty-two
   times more than the paper ground beneath it — while #FDFDFE, the brightest
   step the brand has, appeared NOWHERE. Every content tile, the FAQ, the hero
   and the closing frame were all painted in the same single mid-grey. The
   neutral scale has twelve steps and the page was using three.

   That is the whole "dirty" reading, and it is not a palette that needs
   widening — adding a hue is forbidden outright (§1.6, no second hue family)
   and would not have fixed it. It is a palette that was being used at one
   step. The photography law is explicit about the register the page is
   supposed to sit in: "ONE clean LIGHT ground... Bright, airy, high-key,
   even." A storefront flooded with brand-100 contradicts every frame mounted
   on it.

   So the ladder gets used as a ladder. Paper stays the page. Content tiles
   move up to SURFACE, the near-white, and become objects sitting on it. Tint
   stops being the page's colour and goes back to its actual job — standing in
   for a photograph, where it belongs because it is close to the frames' own
   field.

   AND EVERY TILE GETS ITS HAIRLINE. This is the other half, and the more
   important half: not one tile on this page carried a border. §1.4 defines
   elevation for this brand as "a value change plus a 1px border" — the page
   had neither, so panels dissolved into each other with nothing to say where
   one stopped. A 7-point value step alone is too quiet to separate anything;
   the value step plus the hairline is the system as specified, and it is what
   makes a surface read as a made object rather than a patch of colour.
   ------------------------------------------------------------------------ */
.sk-tile {
  background: var(--surface);
  border: 1px solid var(--border);
}

/* Ink grounds RE-ASSERT their fill, and take the hairline meant for a dark
   stage — --border is blue-derived, tuned for a white ground, and vanishes
   here.

   The re-assertion is not belt-and-braces. `.sk-tile--flood` and
   `.sk-tile--deep` are single-class selectors declared in the vertical, and
   `.sk-tile` above is also a single class declared in THIS file, which loads
   later — so on equal specificity the plain `.sk-tile` background won and both
   ink panels turned white. The page lost its dark anchor entirely: the
   manifesto and the statement panel, measured at 777,000 px² of ink before,
   fell to 25,000. Caught by looking at the rendered page rather than the diff,
   which is the second time on this page that a change read correctly as source
   and wrong as pixels. */
.sk-tile--flood {
  background: var(--brand);
  border-color: var(--on-dark-line);
}
.sk-tile--deep {
  background: var(--surface-ink);
  border-color: var(--on-dark-line);
}

/* A tile that IS a photograph keeps tint: it is the colour showing while the
   frame decodes, matched to the frames' own field so a slow image never
   flashes a second colour into the grid. Declared after .sk-tile so it wins. */
/* The other two tint variants, re-asserted for the same cascade reason.
   --wash is a deliberately recessed panel and --photo is the colour showing
   while a frame decodes; both are tint ON PURPOSE and neither should be lifted
   to surface. All five .sk-tile--* variants that set a background are now
   accounted for here, so a later one cannot be silently overwritten again. */
.sk-tile--wash,
.sk-tile--photo,
.sk-close__tile {
  background: var(--surface-2);
}

/* The FAQ is drawn as two tiles by the vertical but does not carry .sk-tile,
   so it needed saying twice. Same treatment, or the one chapter between the
   manifesto and the close stays grey while everything around it lifts. */
.ms-faq-section .ms-faq__head,
.ms-faq-section .ms-faq__list {
  background: var(--surface);
  border: 1px solid var(--border);
  /* And the same padding as every other chapter. These two tiles carry their
     own clamp — 46.08px at 1440 against the inset's 48.96 — which put the FAQ
     heading at 63 while the whole rest of the page sat at 65. Two pixels, and
     the only block on the page still off the axis. */
  padding: calc(var(--sk-inset) * 1.4) calc(var(--sk-inset) - 1px);
}

/* The closing panel sits ON the photograph, so it is the one surface that is
   already separated by what is behind it. It still takes the hairline: the
   frame under it is pale, and a white panel on a pale field with no edge is
   the same dissolve as everywhere else. */
/* -- THE CLOSING PANEL, SIZED TO CARRY THE SECTION --------------------------
   Measured: 469px of a 1408px tile — a third of the width, 246px tall against
   704 — with the range photographed across the whole frame behind it. At that
   size it reads as a sticker applied to someone else's picture rather than as
   the page's last word, which is the one thing a closing section has to be.

   Wider, and padded on the page's own inset rather than a clamp of its own, so
   the panel's inner edge belongs to the same rhythm as every other block of
   text. It stays well under half the frame: the photograph is still the
   section, and the panel is still mounted on it. */
.sk-close__panel {
  background: var(--surface);
  border: 1px solid var(--border);
  /* The panel is display:inline-block, so it shrink-wraps its longest line and
     a max-width alone changes nothing — raising it from 30rem to 34rem moved
     the rendered width by 18px, because the heading never asked for more.
     width:100% makes the panel claim its column and lets the max-width do the
     limiting, which is the job it was written for. */
  width: 100%;
  max-width: 34rem;
  padding: var(--sk-inset);
}

/* -- THE CLOSING PANEL, ON THE PAGE'S LEFT EDGE ------------------------------
   Its inset was a separate clamp — 40px flat at 1280 — so the panel landed at
   56px while every other block of text on the page started at 60. Nine pixels
   is not visible as nine pixels; it is visible as the last section of the page
   not lining up with the rest of it. It now takes the same inset token as
   everything else, which is what makes the edge hold all the way down. */
.sk-close__copy {
  padding: var(--sk-inset);
}

/* No blur. A frosted bar is a different material from a page whose product is
   photographed flat on an even field. */
.ms-topbar, .ms-header, .site-header, .ms-drawer, .ms-sticky, .is-glass {
  backdrop-filter: none;
  -webkit-backdrop-filter: none;
}

/* -- TYPE · weight -----------------------------------------------------------
   Display runs at 500. 600 is reserved for the mark, the wordmark and small
   emphatic labels; 700 ships in the font for print and identity and is absent
   from the web ladder entirely. A heavy weight on a light page is the loudest
   thing on it, and this brand is quiet. -------------------------------- */
h1, h2, h3, h4, h5, h6,
.display-xl, .display-hero, .display-lg, .display-md,
.title-2xl, .title-xl, .title-lg, .title-md, .title-sm {
  font-weight: 500;
}

/* -- TYPE · tracking ---------------------------------------------------------
   The engine runs ONE rung across every display step. ANNEAL has a ladder: the
   tighten grows with the size, from -0.014em at a small heading to -0.035em at
   a hero. This cannot be done by re-pointing a token, because a single rung
   serves several steps — so each class is mapped onto its own ANNEAL rung. */
.display-xl, .display-hero { letter-spacing: -0.035em; }
.display-lg                { letter-spacing: -0.032em; }
.display-md                { letter-spacing: -0.03em; }
.title-2xl                 { letter-spacing: -0.026em; }
.title-xl                  { letter-spacing: -0.022em; }
.title-lg                  { letter-spacing: -0.018em; }
.title-md, .title-sm       { letter-spacing: -0.014em; }

/* -- THE TYPOGRAPHIC REGISTER ------------------------------------------------
   Controls and nav are CAPS. A button label is .t-button — uppercase at
   +0.06em. This is the one place the brand changes how a word is SHAPED rather
   than coloured: lowercase-sans controls are the register of a friendly
   consumer app, and this system is built the other way. */
.btn, .ms-btn, .pill,
button.button, input.button, a.button,
.wc-block-components-button {
  text-transform: uppercase;
  letter-spacing: var(--p-tracking-6);
}

/* Brand law reads "Nav items are letterspaced caps at data size." SIZE — the
   sentence picks a rung on the type scale and says nothing about the family.
   An earlier pass in this file read "data size" as "the data face" and put the
   entire navigation into Martian Mono. That was a misreading, and it is
   corrected here; the size and the tracking it got right are kept. Weight goes
   to 500 because Switzer at 12px caps is lighter on the page than the mono was
   and the nav should not thin out as it changes face. */
.ms-nav a, .ms-nav__link, .ms-header nav a {
  font-family: var(--p-font-sans);
  font-size: var(--p-text-12);
  font-weight: 500;
  text-transform: uppercase;
  letter-spacing: var(--p-tracking-14);
}

/* -- THE LABEL THAT CAPTIONS ITSELF -----------------------------------------
   The FAQ chapter opens with an eyebrow reading "FAQ" directly above a heading
   reading "Questions, answered." Two labels for one thing, and the small one
   says nothing the large one does not. An eyebrow earns its place by adding a
   register the heading cannot carry — a temperature, a position, a count. This
   one is a category tag on a page with nine sections and no other category
   tags, which is the cheap device the review flagged everywhere else and then
   walked past here. Dropped, not restyled. */
.ms-faq-section .ms-faq__head .ms-eyebrow {
  display: none;
}

/* An eyebrow opens further than a code does — small type needs air, not
   compression. */
.ms-eyebrow, .sk-hero__eyebrow, [class*="__eyebrow"] {
  letter-spacing: var(--p-tracking-12);
}

/* -- THE THIRD ROLE HAD EATEN THE OTHER TWO ---------------------------------
   Measured on the deployed homepage before this change: 84 elements set in
   Martian Mono against 62 in Switzer. The DATA face was the most-used face on
   the page. Of those 84, twenty-five were data — a price, a temperature, a
   volume, an index. The other fifty-nine were language wearing a data face:
   the navigation, the footer column titles, the hero's opening line, the six
   step names in the routine, the ingredient in all eight scale rows, a support
   address, "30 days, sealed".

   This is the largest single reason the page reads cheap, and it is not a
   detail — it is the page's entire texture. Martian Mono is even-width and
   mechanical by design, which is exactly what makes it good at columns of
   figures. Set it at 12px caps across sixty elements and every section becomes
   a terminal readout. It also fights the photography: those frames are soft,
   high-key and expensive, and the type mounted beside them was a printout.

   The engine is not at fault and is not touched. It was built for
   cofactorsupply.com — a supply company, where mono-first is correct and stays
   correct. ANNEAL inherited that system wholesale and never asserted its own.
   This is where it does, in the brand's own last-loaded sheet.

   THE RULE, and it is one line: MONO RENDERS QUANTITIES. A number and its
   unit — $165.00, 1000 °C, 30 ml, 01. Anything a person reads as language is
   Switzer, whatever case or tracking it carries. Letterspaced caps stay
   letterspaced caps; they are simply set in the brand's own face now, which is
   tighter and quieter at the same size. Nothing below touches a price, an
   index, a temperature, a volume, a count or a PDP fact value — those are the
   twenty-five that were always right.

   Weight 500 throughout: Switzer at 12px caps sits lighter than the mono it
   replaces, and a label should not thin out as it changes face. ---------- */

/* Labels and small caps — language, so Switzer; still caps, still tracked. */
.ms-eyebrow,
.sk-hero__eyebrow.u-mono,
.ms-footer__coltitle,
.ms-footer__receipt dt,
.sk-proto__step.u-mono,
.sk-statement__row.u-mono,
.ms-crumbs,
.ms-marquee__item,
.ms-topbar__text,
.pdp-facts dt,
.pdp-ing__role,
.shopcard__cat {
  font-family: var(--p-font-sans);
  font-weight: 500;
}

/* -- THE VARIANT THAT LOST TO ITS OWN BASE CLASS -----------------------------
   .ms-eyebrow--onink is the on-dark eyebrow. The engine sets it correctly:

     engine/assets/sections.css:85
       .ms-eyebrow--onink { color: color-mix(in oklab, var(--brand-foreground) 62%, transparent); }

   which resolves to ~#A3A3A6 on #0F1014 — 7.56:1, comfortably clear. But the
   vertical then declares the BASE class:

     verticals/skincare/assets/components.css:54
       .ms-eyebrow { ...; color: var(--ink-3); }

   Both selectors are one class — specificity (0,1,0) — and components.css
   loads after sections.css, so the base wins and the modifier never applies.
   Every on-ink eyebrow on the site therefore renders #61626A on #0F1014:
   3.14:1, under the 4.5 floor. "BEFORE YOU BUY" on the PDP is one of them.

   A modifier must outrank the class it modifies, so the fix is specificity,
   not order — .ms-eyebrow.ms-eyebrow--onink is (0,2,0) and beats the base
   wherever either sheet ends up in the chain. The colour is the engine's own
   value, restored rather than reinvented.

   NOTE: the collision is in the SHARED vertical, so cofactorsupply.com has the
   identical defect. It is not fixed here — that sheet is shared and cofactor's
   generated output has to stay byte-identical in this change. Fixing it at the
   source, for both brands, is a separate one-line change to components.css.
   ------------------------------------------------------------------------ */
.ms-eyebrow.ms-eyebrow--onink {
  color: color-mix(in oklab, var(--brand-foreground) 62%, transparent);
}

/* Mixed-case prose that was tracked as though it were a code. Tracking that
   flatters caps pulls lowercase apart, so it comes off here. */
.sk-scale__active,
.ms-footer__receipt dd {
  font-family: var(--p-font-sans);
  font-weight: 400;
  letter-spacing: normal;
}

/* "AM" and "PM" head the two routine columns at 24px. A two-letter heading is
   still a heading — it names a half of the day, it does not measure one. */
.sk-proto__h.u-mono {
  font-family: var(--p-font-sans);
  font-weight: 500;
  letter-spacing: var(--p-tracking-6);
}

/* -- FOCUS -------------------------------------------------------------------
   The engine removes the outline on .btn and draws a soft 3px ring instead. A
   visible focus ring is 2px solid at 3px offset and the outline is never
   removed — a blurred ring on a system with no other blur reads as a mistake. */
.btn:focus-visible,
.pill:focus-visible,
.control:focus-visible,
a:focus-visible, button:focus-visible, summary:focus-visible,
input:focus-visible, select:focus-visible, textarea:focus-visible,
[tabindex]:focus-visible {
  outline: 2px solid var(--ring);
  outline-offset: 3px;
  box-shadow: none;
}

/* -- KEPT COMMERCIAL DEVICES, SET IN THE BRAND'S OWN BADGE -------------------
   The spend ladder, the compare-at price and the saving chips stay — that was
   a commercial decision and this file does not argue with it. What changes is
   that they stop being FILLED chips: a badge here is transparent with a
   hairline, and ink fill belongs to the primary button, which is meant to be
   the only filled control in view.

   NOTE FOR WHOEVER OWNS THE BRAND: these devices are named as hard NOs in
   ANNEAL's brand law, and the home page's own dark band reads "There is no
   sale. There never is." Kept on an explicit instruction; the page currently
   contradicts its own copy. */
.pdp-price__save,
.pack__ribbon,
.pack__save,
[data-ms-price-save] {
  background: transparent;
  color: var(--ink-3);
  border: 1px solid var(--border);
  border-radius: 0;
}

.pdp-price__was,
.pack__was,
del {
  color: var(--price-compare);
  text-decoration-thickness: 1px;
}

/* Stars read as a measured value, never a filled gold rating. */
.stars, .stars__i, .ratingmini, .proofstars {
  color: var(--ink-2);
}

.btn--onflood {
  background: var(--flood-100);
  color: var(--flood-100-on);
}

/* -- THE INSET · how far the words sit from the glass -------------------------
   WHAT IS NOT WRONG HERE, because it looked wrong and is not: --sk-gap. The
   bento header in the vertical's sections.css argues at length that the outer
   gutter and every seam between tiles must be ONE number, and it is right —
   the system it replaced had three different left edges on one screen and a
   vertical rhythm that was the sum of three values nobody chose together.
   Nothing below touches --sk-gap. Tile geometry and every seam are unchanged.

   WHAT IS WRONG is a gap in that system rather than the system. A tile states
   its own inner padding with .sk-tile--pad, and every padded tile on the page
   gets 24 -> 48px of it. The hero's copy tile is .sk-hero__copy.sk-tile--5 —
   a tile that was never given --pad — so the first words on the site sat at
   the bento GUTTER, 16px from the edge of the glass, while every other block
   of text on the page sat at 48. That is an omission, not a decision, and it
   is the single loudest cheap-looking thing above the fold.

   Two changes, both conservative. The inset becomes a named measure instead of
   an inline clamp repeated per component, and its CEILING rises: 3.2vw capped
   at 48px stops growing at 1500px, so on a wide monitor a full-bleed tile
   holds its text at the same 48px it uses on a laptop and the page reads as a
   phone layout that was stretched. At 1280 the new value computes to ~44px,
   within a hair of today's 41 — this is not a whitespace grab, it is the same
   measure taught to keep scaling.

   --sk-inset is a LAYOUT measure, not a design token: it is composed from the
   space primitives rather than naming a new brand value, which is why it is
   declared here and not in the token pipeline. ------------------------- */
:root {
  --sk-inset: clamp(var(--p-space-6), 3.4vw, var(--p-space-18));
}

/* More down the page than across it. A tile padded equally on all four sides
   is correct for a card and too tight for a CHAPTER: at 1440 that was 49px of
   air above a heading that opens a section, which is why the page read dense
   even after the surfaces were fixed. Editorial rhythm gives the block axis
   more, and only the block axis — the inline value is the page's left edge and
   every text block on the page is aligned to it. */
.sk-tile--pad {
  /* The inline value is one pixel short on purpose. Every content tile now
     carries a 1px hairline, and a border sits OUTSIDE padding — so a tile
     padded by the full inset puts its text at gutter + 1 + inset while the
     hero, which has no border, puts its text at gutter + inset. One pixel,
     invisible alone, and the reason the page had two body edges instead of
     one. Taking the hairline out of the inset makes every text block on the
     page start at the same number. */
  padding: calc(var(--sk-inset) * 1.4) calc(var(--sk-inset) - 1px);
}

/* -- THE CHROME JOINS THE PAGE ----------------------------------------------
   Measured at 1440: the wordmark and every footer column start at 80, the body
   starts at 65. Header, body and footer on three different left edges — the
   exact fault the bento header in the vertical was written to end, surviving
   in the one place that header does not reach, because .ms-header__bar and the
   footer's .ms-wrap are a CENTRED 80rem container while the page beneath them
   is full-bleed. Nobody reads 15px as 15px; they read a masthead that does not
   line up with the page hanging off it.

   The chrome takes the body's axis: no max-width, and the same gutter + inset
   every text block on the page uses. On a very wide screen the footer columns
   now spread with the page instead of stopping short of it, which is what
   full-bleed means and what the eight chapters above them already do. ---- */
.ms-header__bar,
.ms-footer > .ms-wrap {
  max-width: none;
  padding-inline: calc(var(--sk-gap) + var(--sk-inset));
}

/* -- THE HERO KEEPS THE PHOTOGRAPH'S GROUND, AND ONLY THE HERO ---------------
   The vertical sets `.sk-hero { background: var(--brand-muted) }`, and
   --brand-muted is now the light UI wash — which is right for the button
   hovers it also feeds and wrong here.

   Below 1280 the hero's stage is a BAND (1/1 on a phone, 16/9 on a tablet) and
   the copy panel sits directly under it, so the photograph's bottom edge butts
   straight into CSS. That floor measures about 1.24:1 against paper; the new
   wash is 1.11:1. Painting one against the other draws a horizontal step across
   the hero exactly where the frame ends — the seam this value was measured to
   prevent.

   So the hero takes --stage-glass, which still carries brand.100, the measured
   frosted glass. At and above 1280 the stage is inset:0 and the photograph
   covers the whole section, so this changes nothing on a desktop; it exists
   entirely to keep the joint clean on the two smaller layouts.

   This is the same split as tint vs brand.100 one tier up: a photographic
   measurement and a UI wash were sharing one token, and they have different
   jobs. ------------------------------------------------------------------ */
.sk-hero { background: var(--stage-glass); }

/* -- THE MASTHEAD, BACK TO A MARK --------------------------------------------
   The mark is drawn 3843 x 512 — 7.506 : 1 — so its HEIGHT sets a width seven
   and a half times larger, and every judgement here has to be made in width.

   This rule previously read clamp(20px, 4.6vw, 32px), and the comment beneath
   it argued for the ceiling on the grounds that "a height that reads correctly
   on a desktop bar is 240px of width, which is 17% of a 1440 viewport." That
   was the error. 240px is not a masthead, it is a banner: for comparison the
   houses this brand is positioned against set theirs at roughly 90–110px. At
   32px tall the wordmark also stood 2.7x the 12px navigation beside it, so the
   first thing the eye met on the site was the loudest thing on it — on a page
   whose entire thesis is restraint.

   Reset to a 20px ceiling, which is 150px of width on any desktop, and a 16px
   floor, which is 120px — both comfortably inside §4's 110px minimum, so the
   mark is never set below the size it is licensed at. Against 12px nav the
   ratio falls to 1.67, which reads as a masthead leading a row rather than a
   headline sitting on top of one.

   The middle term is 1.6vw rather than 4.6vw: it reaches the ceiling at about
   1250px and eases to the floor at 1000, so the mark shrinks with the bar
   instead of tracking the viewport at a rate that made it a phone banner. -- */
.ms-header__bar .ms-wordmark__svg {
  height: clamp(16px, 1.6vw, 20px);
}

/* The hero copy tile carries no --pad, so it gets the inset a padded tile would
   have had — ON THE TILE, not on the container around it.

   The first attempt put it on .sk-hero__inner instead, which replaced the bento
   gutter and left the hero's words at 44px while every other tile's words sat
   at gutter + inset = 60. That produced THREE left edges on one screen — 24,
   44, 60 — which is precisely the failure the bento header was written to end.
   Measured on the deployed page, which is the only reason it was caught.

   Start only, not both sides. A symmetric inset would take 87px out of a
   five-column measure that is already wrapping the headline one line further
   than its <br> intends; the right-hand side of this tile opens onto the
   photograph, so there is nothing there for a right inset to protect. */
.sk-hero__copy { padding-inline-start: var(--sk-inset); }

/* -- THE LADDER · size as rank ------------------------------------------------
   Six section headings, six independently authored clamps, and no relationship
   between them. Measured live at 1280 before this change:

     close 70.4  >  manifesto 60.8  >  hero 56.3  =  statement 56.3  >  proto 46.1

   The page's closing panel was the largest type on the site and the hero was
   fourth. Cover the words and the page tells you the goodbye matters most,
   which is exactly the "nothing leads" flatness that reads as unfinished.

   Three rungs, and the hero is deliberately UNCHANGED. Raising it is the one
   move that cannot be made from a stylesheet: its copy tile spans five of
   twelve columns, so a larger size wraps the headline into four lines instead
   of the two the <br> was written for. That needs a wider measure, which needs
   the photograph re-cropped, which is the hero recomposition — a separate
   piece of work. Everything below the hero steps down to sit under it, which
   achieves the ranking now without shipping a worse headline.

     rung 1  hero              up to 61  — one per page, and it leads
     rung 2  chapter statement up to 48  — statement, manifesto, close
     rung 3  chapter head      up to 36  — the scale, the protocol, the FAQ
   ------------------------------------------------------------------------ */
.sk-statement__h,
.sk-manifesto__serif,
.sk-close__h {
  font-size: clamp(var(--p-text-28), 3.4vw, var(--p-text-48));
}

.sk-h2,
.ms-faq-section .title-2xl {
  font-size: clamp(var(--p-text-24), 2.8vw, var(--p-text-36));
}

/* -- THE HERO · a measure that fits the sentence, and a crop that keeps the
      product ------------------------------------------------------------
   No new photography. The master turned out to be a 2400x1310 frame that is
   already art-directed for exactly this layout — flat pale field across the
   left, the vessel standing at 69% of the width, its printed index rule and
   `03 · 900` legible on the glass. (The review claimed a 1280 master upscaled;
   that was a misread of naturalWidth, which reports density-corrected inside a
   srcset. The asset was never the problem.) Two things in CSS were.

   1 · THE MEASURE WAS TOO NARROW FOR ITS OWN SENTENCE. The headline is written
   with a hard break — "Resilience is a process,<br>not a promise." — so it is
   meant to set as two lines. Its tile spans five of twelve columns, about
   495px at 1280, and the first line needs roughly 672px at this size. It
   therefore wrapped to three, and the authored break became a third line
   nobody chose. Widening the tile to eight columns gives the line the room it
   was written for. The tile gets wider; the TEXT does not move right, because
   it is left-aligned and the longest line still ends near 730px — well clear
   of the vessel at 806. The empty column between them is the composition, not
   a gap to be filled.

   2 · THE CROP SACRIFICED THE PRODUCT INSTEAD OF THE EMPTY FIELD.
   `object-position: 0% 50%` anchors the frame LEFT, so every pixel of
   horizontal overflow is taken off the right — the side the vessel is on.
   Simulated across eight viewport shapes, that clips the vessel outright at
   1280x1120, a perfectly ordinary tall desktop window. Anchoring at 72% takes
   the overflow off the left instead, where the frame is flat empty field that
   the section's own background already continues, and the vessel survives
   every shape tested with its right edge at worst 1027 of 1280.

   This is not re-cropping a product shot to fit a layout, which law §5 forbids.
   The frame is already being cropped by `cover`; this only chooses WHICH edge
   pays. Law §5 also settles which: "The temperature numeral is never cropped
   out", and that numeral is printed on the glass.

   KNOWN LIMIT, not fixed here: past about 2400px of viewport width the frame
   runs out of height and the crop starts taking the vessel's BASE, which takes
   its contact shadow with it — the one shadow the brand has, and the thing
   that grounds the object. No object-position solves it; the bottle needs
   992px of a 960px window. That wants a taller master, which is photography.
   ------------------------------------------------------------------------ */
@media (min-width: 1280px) {
  .sk-bento__row > .sk-hero__copy { grid-column: span 8; }
  .sk-hero__img { object-position: 72% 50%; }
}

/* The floor comes up with the measure. The ceiling stays at 61: the widened
   tile carries a 672px first line at that size with room to spare, and the
   next step up the ladder needs 912px, which this frame does not have. */
.sk-hero__h1 {
  font-size: clamp(var(--p-text-36), 4.6vw, var(--p-text-61));
}

/* -- THE ACCENT DECISION, AND THE THREE PLACES IT WAS BEING IGNORED ----------
   THE DECISION: there is no chromatic UI accent, and none is being added. The
   brand law settles this outright — "Primary is INK #101014. It used to be
   #AC471E; a saturated primary contradicts a product whose only colour is a
   hairline." The tokens already agree: --brand is ink, and --accent is a
   SURFACE role (the hover tint), not a hue. So the engine devices that render
   "the accent" as ink are not broken; they are correct, and the review that
   called this brand "accentless, which is why accents vanish" and proposed
   promoting ramp amber to a UI accent was proposing the exact change the brand
   made and then reversed. It is not made here.

   What the law does grant is two — only two — legal appearances for ramp
   colour: the whole continuum, or a single per-item index rule. Everything
   else is "a panel, button, large field or background", and an instrument that
   decorates has stopped being an instrument. Three places were outside that:

     1. The footer strip renders the ramp as nine discrete adjacent blocks.
        Law: "The ramp always renders as a continuum. Separated swatches turn
        it into a data-viz heat map." That is precisely what it looked like.
     2. The footer wordmark is stroked in --ramp-3 at 45%. That is a ramp stop
        used as a large decorative field, on the one element §1.10 protects
        most ("No device may cut the wordmark"), and it landed in the muddy
        middle because a warm mid-value on a cold near-black has nowhere to go.
     3. The hero carries an 8px strip of flood NEUTRALS — not a ramp violation,
        but not anything else either: a grey-on-grey sliver that says nothing
        and reads as a rendering artefact.
   ------------------------------------------------------------------------ */

/* 1 · The strip becomes the continuum. The spans stay in the markup — they are
   the engine's, and the engine still draws them for cofactor — but here they
   carry no fill and the band itself holds the gradient. */
.ms-footer__ramp {
  background: var(--ramp-gradient);
}
.ms-footer__ramp span {
  background: none;
}

/* 2 · The wordmark takes the ground's own light at low strength: an ink-on-ink
   ghost, which is what a watermark is. No ramp stop, no second hue, and it
   finally sits at one clear value instead of halfway between two. */
.ms-footer__mark .ms-wordmark__svg path {
  stroke: color-mix(in oklab, var(--brand-foreground) 14%, transparent);
}

/* 3 · The sliver goes. The scale is stated by the exhibit further up the page
   and by the continuum in the footer; a third, meaningless strip is noise. */
.sk-hero__ramp {
  display: none;
}

/* -- HOVER · the cards stopped floating -------------------------------------
   A leftover from nulling the shadows, and only visible once they were gone.
   .flood and .rampcard both answer a hover with `translateY(-2px)` plus
   --elevation-lift. With the lift now `none`, the card rises two pixels with
   nothing underneath it and, on .flood, no change to its edge either. A
   surface that moves without casting anything does not read as raised; it
   reads as a rendering glitch.

   §1.4 is explicit about what elevation is for this brand — "a value change
   plus a 1px border" — and about what it is not: there are no drop shadows,
   and the only shadow in the system is the contact shadow inside a
   photograph. So the answer to a hover is the EDGE, not altitude. The inset
   ring on a flood card roughly doubles; the ramp card already changed its
   border and now does that alone.

   The image still scales inside its frame. That is not the card floating —
   it is content moving within a fixed opening, which stays grounded and is
   the one motion the vertical already uses everywhere. ---------------- */
.flood:hover, .flood.is-hover, .flood:active,
.rampcard:hover, .rampcard.is-hover {
  transform: none;
}

.flood:hover, .flood.is-hover {
  box-shadow: inset 0 0 0 1px
    color-mix(in oklab, var(--band-on, var(--foreground)) 26%, transparent);
}

/* -- THE ITALIC THAT DOES NOT EXIST ------------------------------------------
   The engine rations one display-italic per view (`.wit`, primitives.css) and
   the skincare PDP restates the same device on an authored `<em>`. Both resolve
   to `font-style: italic` on --font-display — and this brand's display face is
   Switzer, which ships here in four ROMAN cuts and no italic at all. The
   browser therefore synthesised an oblique by shearing the roman. At the
   manifesto's 60px that is the most legible "template site" tell on the page,
   and it is a drawing this foundry never made.

   Emphasis moves to the axis the brand already owns. Type law: the ladder is
   carried by size and tracking, not by a change of drawing — so the display
   line takes the display weight, and the inline em takes one step of weight
   instead of a sheared letterform. Licensing and shipping a real italic cut is
   a brand decision, not a CSS one; until it is made, a decided roman beats a
   synthetic oblique. ---------------------------------------------------- */
.wit {
  font-style: normal;
  font-weight: 500;
}

/* Same device, restated on the PDP for an authored `<em>` that cannot carry a
   class. Its accent colour is --brand, which on this brand IS --foreground — so
   the colour half was already painting ink on ink and the emphasis rested
   entirely on the shear. Weight now carries it alone. */
.pdp-inside__sub em {
  font-style: normal;
  font-weight: 500;
}

/* -- THE SECONDARY CONTROL NEEDS AN EDGE -------------------------------------
   .btn--ghost is declared `border-color: transparent` on the assumption that a
   filled hover will announce it — and its hover fill is --brand-muted, which is
   the exact colour of the hero it sits on. So the home page's second call to
   action, "Read the scale", rendered as a bare word: no edge at rest, no change
   under the cursor, nothing to say it could be pressed at all.

   A hairline is this brand's entire vocabulary for elevation, so the secondary
   control is drawn the way every other bounded surface here is drawn — a
   hairline at rest that resolves to full strength on hover. currentColor rather
   than a token, so one rule holds on the ink bands and on paper alike, and the
   inherited fill is left exactly as it was. ---------------------------- */
.btn--ghost {
  border-color: color-mix(in oklab, currentColor 30%, transparent);
}
.btn--ghost:hover {
  border-color: currentColor;
}

/* -- THE TICKER · separators, and edges that are not cuts --------------------
   Two defects in one band. The separator dot is drawn in --brand, and on this
   brand --brand IS the ink: #101014 dots on the #040406 band, invisible, so
   five distinct claims ran together as a single unbroken sentence. And the band
   clips its own loop with nothing but `overflow: hidden`, so the first and last
   phrase are permanently sawn through mid-glyph at the viewport edge.

   The dot is re-pointed at the band's own text colour, one step below the
   phrases so it separates without competing. The edges take a mask, which is
   what makes a marquee read as continuous motion seen through a window rather
   than a line of text with its ends cut off.

   BOTH BANDS, because there are two. The home page runs .ms-marquee and every
   product page runs .pdp-marquee — separate components carrying the identical
   pair of defects on the identical ink ground, which is exactly how a fix ships
   looking finished on the page someone screenshotted and stays broken on the
   eight pages that actually sell. ------------------------------------- */
.ms-marquee__dot,
.pdp-marquee__dot {
  background: color-mix(in oklab, var(--brand-foreground) 38%, transparent);
}

.ms-marquee,
.pdp-marquee {
  -webkit-mask-image: linear-gradient(90deg,
    transparent 0, #000 var(--p-space-16),
    #000 calc(100% - var(--p-space-16)), transparent 100%);
  mask-image: linear-gradient(90deg,
    transparent 0, #000 var(--p-space-16),
    #000 calc(100% - var(--p-space-16)), transparent 100%);
}

/* -- THE BAND AT THE TOP OF THE PAGE COMES OFF THE INK -----------------------
   Brand law, §2: "LIGHT-FIRST. :root is light. [data-theme=dark] / .on-dark is
   punctuation — one section, a footer, a statement. Never the page default."

   .ms-marquee sits ABOVE the header, which makes it the first band on every
   page and the darkest value in the document — #040406 across the full width,
   directly over a #F5F6FB page and a high-key photograph. That is not
   punctuation; that is chrome, and a black full-bleed strip of moving caps at
   the top of a shop is the exact silhouette of a sale bar. It was doing more
   to make this page look cheap than any single element below it.

   The content stays — five brand facts, none of them promotional, one of them
   the refusal to discount. Only the delivery changes: the lightest surface in
   the ladder (#FDFDFF, which nothing else at the top of the page was using, so
   the band still separates from the #F5F6FB body), a real hairline under it
   instead of the 8%-white line that was invisible, and the phrases in --ink-3
   at 5.97:1. The dot goes back to ink — the engine's own choice, which was
   right and which the on-dark override had to invert.

   Motion halves. A 30s cycle at this width reads as a ticker demanding to be
   read; 60s reads as something moving slowly enough to ignore, which is what a
   standing statement should do.

   .pdp-marquee is deliberately NOT included. It runs mid-page on a product
   page, where a dark band IS the punctuation the law describes — one statement
   against a light document. Same component, different job, different answer.
   ------------------------------------------------------------------------ */
.ms-marquee {
  background: var(--surface);
  border-block: 1px solid var(--border);
}
.ms-marquee__item { color: var(--ink-3); }
.ms-marquee__dot  { background: color-mix(in oklab, var(--foreground) 30%, transparent); }
.ms-marquee__track { animation-duration: 60s; }

/* -- THE PRINTED LINE, KEPT WHOLE -------------------------------------------
   The macro on "Printed on the glass. Not on a label." was slicing the very
   words it points at. I called that unfixable earlier — the master cropped
   through its own type, needs a re-shoot. Measuring it says otherwise.

   Sampled from the loaded frame: the type runs from 66% to 95% of the master's
   height, and the master keeps 3% clear beneath it. The tile is a wider ratio
   than the frame, so `cover` takes 9% off the height — CENTRED, which is 4.5%
   off the bottom, which is more than the 3% margin the photographer left. The
   crop was eating the clearance, not the photograph.

   Anchoring the bottom moves all 9% to the TOP, where the frame is plain
   glass, and the printed line arrives intact with its own margin under it. No
   re-export needed. On the one section whose entire argument is that the type
   is printed into the glass, the type is now never cut.

   CROP LAW for this brand, stated once here: if a photograph contains printed
   type, the crop includes the whole line or excludes it entirely. ------- */
.sk-statement .sk-tile__img {
  object-position: 50% 100%;
}

/* -- THE PRODUCT PLATE IS 4 : 5, AND STAYS 4 : 5 -----------------------------
   §5: "Product plates are 4:5 — the production frame ratio exactly. Never
   re-crop a product shot to fit a layout." The flood card's stage is 1 : 1 and
   the plates are 4 : 5, so every one of these cards was cutting 19% off a
   production frame — the same violation already fixed on .shopcard__stage, in
   a second component that was missed at the time. The vessel loses its cap or
   its base depending on where the subject sits, and its contact shadow with
   the base.

   Matching the stage to the source ratio ends the crop entirely: the plate is
   shown as shot, full-bleed, no letterbox. Brand-only — cofactor's square
   1024x1024 packshots are untouched by this.

   BOTH SELECTORS, because the row variant states its own ratio at
   `.flood--row .flood__media` — two classes, which outranks the base and kept
   the 1 : 1 on exactly the cards the home page uses. The row card grows taller
   as a result. That is the layout adapting to the plate, which is the
   direction §5 requires; the alternative is the plate adapting to the layout,
   which is the sentence it forbids. ------------------------------------ */
.flood__media,
.flood--row .flood__media {
  aspect-ratio: 4 / 5;
}

/* -- PRODUCT-CARD IMAGE · portrait stage for a portrait bottle ---------------
   The engine's shop card stage is 1 : 1, which is right for cofactor's square
   1024×1024 packshots. ANNEAL's photography is shot 4 : 5 with the vessel
   centred, so a square cover-crop sliced the dropper cap off the top and the
   base off the bottom — the cards looked cut off. Match the stage to the source
   ratio and the whole bottle shows, still full-bleed, no letterbox. Brand-only:
   cofactor's square stage is untouched. */
.shopcard__stage { aspect-ratio: 4 / 5; }
