/**
 * @file
 * Colour tokens for the Ágora Transparency Theme - task T-606.
 *
 * THIS IS THE ONLY FILE IN THE THEME THAT CONTAINS A COLOUR LITERAL.
 * Every other stylesheet reaches these values through var(). T-610 turns that
 * sentence into an invariant with a dirty case; until then it is a rule, and it
 * is the reason this is a file of its own rather than a block at the top of the
 * base stylesheet.
 *
 * HOW TO PARSE IT - the T-608 contract. The point of the shape below is that
 * the contrast checker needs a regular expression, never a CSS grammar. Two
 * line shapes carry everything, and nothing else in this file is data:
 *
 *   a token line is      two spaces, --agora-color-NAME, colon, space,
 *                        a hex literal, semicolon, end of line
 *   a pair line is       a whole-line CSS comment whose body is
 *                        @pair FOREGROUND-TOKEN BACKGROUND-TOKEN MIN-RATIO NOTE
 *
 * Three rules make that parse total rather than best-effort, and none of them
 * may be relaxed without amending T-608 first:
 *   1. ONE declaration per line. No two tokens ever share a line.
 *   2. EVERY value is a BARE HEX LITERAL - three digits or six, nothing else.
 *      No color-mix(), no rgb(), and no var() indirection inside a token's own
 *      value: a token whose value named another token would force the checker to
 *      resolve a reference graph, and that is the point where a regular
 *      expression stops being enough.
 *
 *      THE THREE-DIGIT FORM IS NOT A STYLE CHOICE AND MAY NOT BE "TIDIED" AWAY.
 *      This file was first written with six digits throughout, which is easier
 *      to parse. Drupal core's stylelint configuration extends
 *      stylelint-config-standard, whose color-hex-length rule is `short`, so
 *      `#ffffff` is an ERROR and the fixer rewrites it to `#fff`. Since the
 *      stylelint job is blocking, the linter wins and the parser adapts: expand
 *      a three-digit literal by doubling each digit before computing luminance.
 *      That is one line in the checker. Writing `#ffffff` to keep the parser
 *      simple would turn a blocking job red instead.
 *   3. Every foreground-on-background combination this theme is allowed to
 *      render is DECLARED as a pair line. The checker computes WCAG relative
 *      luminance for each declared pair; a pair nobody declared is a pair
 *      nobody checks, so adding a token without adding its pairs is a silent
 *      hole, and T-608's dirty case exists to make that hole loud.
 *
 * NO DARK MODE, AND THAT IS A DECISION RATHER THAN AN OMISSION.
 * There is no prefers-color-scheme block here and no second palette anywhere in
 * this theme. Two reasons, in order of weight. First, a dark palette re-declares
 * every token name inside a media query, so a flat name-to-value parse keeps
 * whichever declaration it read last and then reports a clean run having checked
 * one palette out of two - a false green of exactly the species this project
 * keeps catching. Supporting it honestly means keying T-608's token map by
 * scope, which is an amendment to T-608 rather than a drop-in. Second, dark mode
 * is not a WCAG 2.2 AA requirement, and this unit is already over its scope gate.
 * If it is wanted it is a task row of its own, with its own pairs and its own
 * computed ratios.
 *
 * THE PALETTE IS MEASURED FROM THE PROJECT'S OWN MARK, not chosen by taste.
 * Three values come out of the artwork and everything else is derived from
 * them: a deep warm navy for the ground the mark sits on, a warm cream for the
 * light, and the orange of the one column that stands open. The mark is a
 * portico with a door left open, which is the whole of what this product
 * claims, so the orange had to reach the page as something a reader can see.
 * Until this file was rewritten it could not: the only orange in the theme was
 * the focus ring, which a mouse user never sees at all.
 *
 * THE ACCENT EXISTS AT TWO STEPS, AND THE REASON IS ARITHMETIC RATHER THAN
 * TASTE. The brand orange reaches 3.41:1 on white and 3.05:1 on the cream
 * panel. That clears the non-text threshold of SC 1.4.11 and does not come
 * close to the 4.5:1 that SC 1.4.3 asks of body text, so the brand value is
 * declared for RULES, BARS AND EDGES only, and a darker step of the same hue
 * carries the few words that are allowed to be orange. Anyone tempted to set
 * body copy in the brand value will find no declared pair for it, which is the
 * point: the palette refuses the combination rather than trusting a reviewer to
 * notice it. The cream panel is the tightest pair in the file at 3.05 - lighten
 * that surface and the accent rule on it goes below threshold, so the two move
 * together or not at all.
 *
 * THE PALETTE IS STILL SOBER, AND THAT IS A CONSTRAINT AND NOT A HEDGE. Navy,
 * cream, one institutional blue, one accent used sparingly, four muted status
 * hues. A transparency portal for a town hall publishes salaries and contracts,
 * and the tables are the content; the colour exists to make the page look like
 * it was made on purpose, not to make it look busy.
 *
 * COLOUR IS NEVER THE SOLE CARRIER OF MEANING. The four status tokens are text
 * colours that accompany a written label; not one of them is a bare swatch. The
 * accent never carries meaning alone either - the current navigation item is
 * marked by weight and by a rule beneath it as well as by hue. The focus token
 * sits in a different hue family from the link token on purpose, so that a
 * focused control never reads as a link.
 *
 * Every ratio below was computed, not estimated, and is recorded in the task
 * report for T-606.
 */

:root {
  /* Surfaces. */
  --agora-color-surface: #fff;
  --agora-color-surface-subtle: #f7f2e8;
  --agora-color-surface-sunken: #efe7db;
  --agora-color-surface-inverse: #0d2b4d;
  --agora-color-surface-inverse-raised: #16395f;

  /* THE COOL SURFACE IS A TINT OF THE INVERSE ONE AND NOT A NEW HUE: the navy
     mixed 5.5% into white, so the palette stays navy, cream, one blue, one
     accent. It carries the front page's service-area band. The accent rule on
     it computes to 3.07:1 - as tight as the 3.05 on the cream panel - so the
     cool and the cream surfaces move together: lighten either and its accent
     pair drops below the non-text threshold. */
  --agora-color-surface-cool: #f2f3f5;

  /* Text. */
  --agora-color-text: #142943;
  --agora-color-text-muted: #55606d;
  --agora-color-text-inverse: #fff;
  --agora-color-text-inverse-muted: #d9d0c2;

  /* Links. */
  --agora-color-link: #14508c;
  --agora-color-link-hover: #0d3661;
  --agora-color-link-visited: #5b3a76;

  /* Accent - the open door of the mark. */
  --agora-color-accent: #c8781f;
  --agora-color-accent-strong: #94500e;
  --agora-color-accent-soft: #d98b33;

  /* The mark, drawn large. --agora-color-mark is the solid tone, and as of
     2026-09-05 the hero band no longer draws anything in it - the watermark
     below replaced it there. It is still rendered, and by an asset rather than
     by a rule: logo-inverse.svg writes this same value as a LITERAL, because an
     SVG loaded through an <img> is an isolated document that cannot read a
     custom property. That file's comment says the two move together, and this
     one is the other end of that sentence.

     --agora-color-mark-watermark is COMPUTED and not chosen. The note above its
     pair line carries the arithmetic and the reason it is a token at all. */
  --agora-color-mark: #f2ece1;
  --agora-color-mark-muted: #4a7ba8;
  --agora-color-mark-watermark: #697888;

  /* Focus. */
  --agora-color-focus: #b34700;
  --agora-color-focus-inverse: #ffb877;

  /* Borders. */
  --agora-color-border: #857c6e;
  --agora-color-border-strong: #5f584c;

  /* Status. */
  --agora-color-status-success: #1b6045;
  --agora-color-status-warning: #8a5a00;
  --agora-color-status-error: #a12a1e;
  --agora-color-status-info: #1c5375;
}

/* Body text, on every surface it may be rendered on. */
/* @pair --agora-color-text --agora-color-surface 4.5 body text on the page */
/* @pair --agora-color-text --agora-color-surface-subtle 4.5 body text on a panel */
/* @pair --agora-color-text --agora-color-surface-sunken 4.5 body text in a table header */
/* @pair --agora-color-text-muted --agora-color-surface 4.5 caption and metadata on the page */
/* @pair --agora-color-text-muted --agora-color-surface-subtle 4.5 caption on a panel */
/* @pair --agora-color-text-muted --agora-color-surface-sunken 4.5 caption in a table header */
/* @pair --agora-color-text-inverse --agora-color-surface-inverse 4.5 text in the footer band */
/* @pair --agora-color-text-inverse --agora-color-surface-inverse-raised 4.5 label on a solid button */
/* @pair --agora-color-text-inverse-muted --agora-color-surface-inverse 4.5 secondary text in the footer band */

/* The cool band - every token that sits on it, declared before any rule uses
   it there. The accent lands at 3.07, the tightest pair after the cream one. */
/* @pair --agora-color-text --agora-color-surface-cool 4.5 body text on the cool band */
/* @pair --agora-color-text-muted --agora-color-surface-cool 4.5 caption on the cool band */
/* @pair --agora-color-link --agora-color-surface-cool 4.5 link on the cool band */
/* @pair --agora-color-link-hover --agora-color-surface-cool 4.5 hovered link on the cool band */
/* @pair --agora-color-link-visited --agora-color-surface-cool 4.5 visited link on the cool band */
/* @pair --agora-color-accent --agora-color-surface-cool 3.0 accent rule on the cool band */
/* @pair --agora-color-border --agora-color-surface-cool 3.0 cell rule on the cool band */
/* @pair --agora-color-border-strong --agora-color-surface-cool 3.0 section rule on the cool band */
/* @pair --agora-color-surface-inverse --agora-color-surface-cool 3.0 a card mark on the cool band */
/* @pair --agora-color-focus --agora-color-surface-cool 3.0 focus ring on the cool band */

/* Links, in all three states, on every surface. */
/* @pair --agora-color-link --agora-color-surface 4.5 link on the page */
/* @pair --agora-color-link --agora-color-surface-subtle 4.5 link on a panel */
/* @pair --agora-color-link --agora-color-surface-sunken 4.5 link in a table header */
/* @pair --agora-color-link-hover --agora-color-surface 4.5 hovered link on the page */
/* @pair --agora-color-link-hover --agora-color-surface-subtle 4.5 hovered link on a panel */
/* @pair --agora-color-link-hover --agora-color-surface-sunken 4.5 hovered link in a table header */
/* @pair --agora-color-link-visited --agora-color-surface 4.5 visited link on the page */
/* @pair --agora-color-link-visited --agora-color-surface-subtle 4.5 visited link on a panel */
/* @pair --agora-color-link-visited --agora-color-surface-sunken 4.5 visited link in a table header */

/* The mark drawn at hero scale. It is decoration and carries no information, so
   WCAG exempts it - but the token file declares every combination this theme
   renders, and an unchecked exception is how the next one gets made. All three
   are held to the non-text threshold. */
/* ⚠️ THIS TOKEN IS RENDERED IN TWO PLACES AND THE NOTE NAMES BOTH, because it
   was briefly cut to one and the correction is worth leaving visible. It is
   the footer band's mark, written into logo-inverse.svg as a literal that file
   has to carry because an isolated document cannot read a token; and it is the
   hero portico AGAIN whenever that band carries a picture, which is the
   2026-09-05 amendment to D-048. Only the flat hero band takes the watermark
   below. Both renderings stand on this same surface, so one pair covers them. */
/* @pair --agora-color-mark --agora-color-surface-inverse 3.0 the footer mark and the hero portico over a picture */
/* ⚠️ THE PAIR BELOW IS NO LONGER RENDERED ANYWHERE, AND IT IS KEPT ON PURPOSE,
   the same way the soft accent's pair further down is kept. It is still a TRUE
   statement about these two tokens; the pair count is a floor that may rise and
   never fall; and this line is the only place --agora-color-mark-muted is named,
   so deleting it would leave a declared token unpaired and fail
   tests/bin/contrast-check on its own "every token is named at least once" rule
   as well as dropping the count.

   WHY IT STOPPED BEING RENDERED, and 3.19 is the whole answer: it clears the
   threshold by four hundredths. The hero band now carries a toned picture
   behind the mark, and a token with that little headroom tolerates a ground of
   0.0282 relative luminance when the band's own surface is already 0.0235.
   Measured on the rendered pixels, the record bars in this token fell to
   1.40:1, and a scan of every position and three sizes of the mark inside the
   band found nowhere they clear it. They are drawn in
   --agora-color-text-inverse-muted instead, which tolerates 0.1792 against a
   measured worst ground of 0.1275 and needed no new pair, because the footer
   band already declares it.

   ANYONE PUTTING THIS TOKEN BACK ON A SURFACE SHOULD READ THAT PARAGRAPH
   FIRST. It is not a bad value; it is a value with no margin, and margin is
   what a declared pair is for. */
/* @pair --agora-color-mark-muted --agora-color-surface-inverse 3.0 the record bars on the hero band */

/* THE WATERMARK, AND IT IS A COMPUTED VALUE RATHER THAN A CHOSEN ONE.
   Decision D-048, signed 2026-09-05.

   WHAT IT IS FOR. At hero scale the mark in the solid tone reads as a cut-out
   laid on the band rather than as part of it - the complaint that opened D-048,
   and it is a fair one at 12.15:1 against a ground whose whole job is to be the
   one quiet band on the page. The fix is to sink the mark into the band.

   ⚠️ WHY THIS IS A TOKEN AND NOT `opacity`, WHICH IS THE ENTIRE DECISION.
   Setting `opacity` on the mark would have been one declaration and it would
   have left this file describing a rendering that no longer happens.
   tests/bin/contrast-check parses TOKEN LITERALS: it would have gone on
   computing 12.15:1 for a portico painted at a third of that, reporting a pass
   about a colour nothing draws. Every check in this repository would have
   stayed green while the measurement underneath it stopped being about the
   page - a silenced invariant arriving through a mechanism nobody edited. So
   the composite is computed once, written down as a value, and checked like any
   other value.

   THE ARITHMETIC, so that this line can be re-derived rather than trusted.
   Source-over composite in non-linear sRGB, which is what `opacity` itself
   does, at alpha 0.40:

     channel = round(0.40 * mark + 0.60 * surface-inverse)

   applied to --agora-color-mark over --agora-color-surface-inverse. The result
   is the value declared above, and contrast-check computes 3.16:1 for it on
   that same ground.

   ⚠️ AND THE PAGE WAS READ BACK, BECAUSE COMPUTING A RATIO IS NOT THE SAME AS
   PAINTING ONE - which is the whole complaint D-048 makes against `opacity`,
   and it would be a poor decision that did not check itself the same way. The
   front page was rendered at a 1440 viewport with no masthead picture and every
   pixel of the mark compared against the pixel immediately behind IT: 14408
   pixels of structure and 3705 of record bars, median 3.16:1 - the declared
   value exactly - minimum 3.12:1, and 0 of either below the threshold. The
   minimum sits four hundredths under the median rather than on it because the
   band is not quite flat: the mark's own scrim in css/components.css composites
   this surface over itself, which rounds a few hundred pixels of the ground one
   step of 255 lighter. That rounding is predicted in that file's own comment.
   It is written down here as measured rather than left as a discrepancy for the
   next reader to find and have to explain.

   WHY 0.40 AND NOT LOWER, WHICH IS THE ONLY FREE PARAMETER HERE. The composite
   crosses 3.00:1 at alpha 0.3808, so 0.40 is the first hundredth above this
   theme's own floor with the ratio not sitting on it. Quieter than this is not
   available: assertion 6 of contrast-check allows a declared minimum of 4.5 or
   3 and nothing else, so 3 is as faint as a pair may claim, and a watermark
   whose pair was dropped instead would be a colour nobody checks. The mark is
   therefore as quiet as this theme's own gate permits and not one step quieter,
   which is the honest place for it.

   ⚠️ THIS TOKEN IS DRAWN ON ONE GROUND ONLY, AND THAT IS WHY THE PAIR IS TRUE
   RATHER THAN NEARLY TRUE. The hero band has two states: flat
   --agora-color-surface-inverse, which is what a site installing this theme
   alone gets and what screenshot.png shows, and a masthead picture a site sets
   at /admin/appearance/settings/agora_theme. THE WATERMARK IS PAINTED IN THE
   FIRST STATE AND IN NO OTHER - over a picture the mark goes back to the solid
   tone, which is what shipped before D-048 - so the ground this pair names is
   the only ground this token ever stands on, and 3.16:1 is a computed fact
   about the page rather than about one of its states.

   ⚠️ IT WAS NOT ALWAYS SO, AND THE FIRST ROUND IS WORTH A SENTENCE BECAUSE THE
   TOKEN SURVIVED IT UNCHANGED. The watermark was applied to both grounds for
   one commit. Measured over a picture, 52% of the structure and 89% of the
   record bars fell below 3:1, and - the finding that actually decided it - the
   composition stopped reading as a portico, because the recess is opaque and
   became the loudest thing in the mark once the columns stopped outshining it.
   Making the tone conditional fixed the appearance and, as a by-product, gave
   this pair a single honest ground. The measurement is kept in
   css/components.css beside the picture it was taken against.

   ⚠️ AND THE MARGIN IS THIN, WHICH MATTERS LESS THAN IT LOOKS. This token
   tolerates a ground of 0.0274 relative luminance before it drops below the
   threshold, against a band that sits at 0.0235. That is the same shape of
   headroom that took --agora-color-mark-muted off the page four paragraphs up,
   and it is accepted here for a reason that did not apply there: that token was
   being asked to survive a picture, and this one is never on one. The thinness
   is therefore against a FIXED value, not a drifting one - a ground that cannot
   move needs no headroom, and the day this token is painted over a picture is
   the day it needs a great deal. */
/* @pair --agora-color-mark-watermark --agora-color-surface-inverse 3.0 the watermark portico and record bars on the hero band */

/* The accent, as a graphic. Non-text, so the threshold is 3:1 - WCAG 2.2 SC 1.4.11. */
/* @pair --agora-color-accent --agora-color-surface 3.0 accent rule on the page */
/* @pair --agora-color-accent --agora-color-surface-subtle 3.0 accent rule on a panel */
/* @pair --agora-color-accent --agora-color-surface-inverse 3.0 accent rule in the footer band */
/* @pair --agora-color-accent --agora-color-surface-inverse-raised 3.0 accent rule on a solid button */

/* The accent, as a FILL WITH THE PAGE'S OWN TEXT ON TOP OF IT - the bar layer
   over the spend table. This is the third step of the same hue and it exists
   for one arithmetic reason, which is worth stating because "we already have an
   orange" is the obvious objection.

   The bar is drawn BEHIND a right-aligned money figure, so on the largest row -
   the one whose bar is full width - the figure sits ON the fill. That figure is
   TEXT, so SC 1.4.3 applies to it at 4.5:1, and the brand accent reaches only
   4.31:1 under the body text token. Four hundredths short is still short: the
   design moves, not the threshold. This step is light enough to carry the text
   at 5.39:1 and dark enough to still read as a bar.

   THE FILL AGAINST THE CELL BEHIND IT IS DELIBERATELY NOT DECLARED, and that is
   the one exemption in this file that is claimed rather than computed - so it is
   argued rather than asserted. SC 1.4.11 governs graphical objects that are
   REQUIRED TO UNDERSTAND THE CONTENT. The amount is written out in full, in
   text, inside the same cell the bar is drawn in; take the bar away and not one
   figure becomes unavailable. It is the same species as the zebra band in
   css/components.css, which conveys nothing either and is likewise undeclared,
   and it is why the bar carries no role, no aria-label and no title.

   For the record rather than for the gate, since a number nobody wrote down is a
   number nobody can check later: the fill measures 2.73:1 against the page
   surface and 2.45:1 against the banded row - visible on both, and neither
   figure is a conformance claim. Declaring either at 3.0 would be a claim this
   theme cannot meet AND a claim it does not need, and a pair that has to be
   argued down later is worse than a pair that was never written. */
/* MOOT SINCE THE BAR STOPPED RUNNING UNDER THE FIGURE, AND DELIBERATELY KEPT.
   The amount used to sit on the fill on the widest row, which is what this pair
   was declared for. It no longer can: css/components.css reserves a strip for
   the figure and scales the fill to stop short of it, so the amount always sits
   on the cell. The pair is retained rather than deleted for three reasons, and
   the third is the one that would have bitten. It is still a TRUE statement
   about these two tokens. The pair count is a floor that may rise and never
   fall. And the soft accent below is named by EXACTLY ONE of these
   declarations - this one - so removing the line would leave a declared colour
   unpaired and fail tests/bin/contrast-check on its "every token is named at
   least once" rule as well, trading a moot pair for a real finding.

   The prose here deliberately does NOT spell the marker or the token name on
   one line: the checker reads any such line as a declaration that misses the
   shape, and it said so when this note was first written. */
/* @pair --agora-color-text --agora-color-accent-soft 4.5 an amount on its own bar */

/* ⚠️ AND IT IS NOT MOOT ANY MORE. The block above argues at length for keeping a
   pair whose only job had gone, and gives three reasons. There is now a fourth
   and it is the ordinary one: the soft accent is the FILL OF THE HERO'S PRIMARY
   BUTTON, and the dark page text is the label printed on it, at the 5.39:1 that
   block had already computed for a use which no longer existed. The pair was
   retained on principle and the principle paid - tidied away when it went moot,
   this round would have had to derive it again from nothing.

   THE FILL AGAINST THE BAND IS A DIFFERENT QUESTION, DECLARED NEXT. A control's
   own edge is what identifies it as a control, which is the argument the light
   control on this band already makes two groups down; it arrives here at a
   different colour and has to be computed again rather than inherited.

   HELD TO 3:1 AND NOT 4.5. Nothing is read off the fill - it is the visual
   boundary of a user interface component, WCAG 2.2 SC 1.4.11. The label printed
   on top of it is the text, and that is the pair above.

   ⚠️ 5.23 IS THE FLAT-BAND FIGURE AND THE BAND IS NOT FLAT. READ THIS BEFORE
   MOVING THE BUTTONS. Above 60rem the masthead carries a photograph at four
   tenths opacity, so the ratio a reader actually gets depends on WHERE in the
   band the control stands. Two positions, measured in a real render at a 1440
   viewport by taking a screenshot of the page and decoding its pixels, rather
   than by simulating the composite:

     * at the button's own position, x 144 to 315, the brightest pixel adjacent
       to its edge is rgb 18 47 80, relative luminance 0.0274, and the fill
       measures 4.97:1 against it. The scrim over the type is opaque to about
       576px, so the picture barely arrives here - 0.0274 against the flat
       token's own 0.0235.
     * over the mark, at the band's inline end, the ground was measured at
       0.1275 at its brightest by the round that placed the mark, and this fill
       would reach 2.17:1 there. That is BELOW the threshold.

   So the declaration holds for a control in the band's type column and would
   NOT hold for one moved out past the scrim's ramp. Widening the row, adding a
   third control or moving either button outward owes this measurement again.
   The remedy would be a border in a declared token or a scrim - never a lowered
   minimum on the line below. */
/* @pair --agora-color-accent-soft --agora-color-surface-inverse 3.0 the primary control on the masthead band */

/* The accent, as text. A darker step of the same hue, so 4.5:1 applies. */
/* @pair --agora-color-accent-strong --agora-color-surface 4.5 current section in the navigation */
/* @pair --agora-color-accent-strong --agora-color-surface-subtle 4.5 required marker on a panel */
/* @pair --agora-color-accent-strong --agora-color-surface-sunken 4.5 sort indicator in a table header */

/* THE DARKEST STEP OF THE HUE AS A FILL, WHICH IS NEW, AND THE LABEL ON IT IS
   WHITE RATHER THAN DARK. That inversion is forced by arithmetic and not chosen,
   so it is written down here rather than left looking like a whim.

   The hero's primary button is filled with the soft accent and labelled in the
   dark page text. Every hovered control in this theme darkens its fill, so the
   hovered primary wants a darker step of the same hue - and there is no step of
   it that carries the dark text at 4.5:1. The brand accent reaches 4.31:1 under
   that text and this darkest step reaches 2.39:1; four hundredths short is
   still short, and the design moves rather than the threshold. White on this
   step is 6.15:1, so the label inverts with the fill.

   THE BUTTON'S OUTER EDGE DOES NOT MOVE WITH IT, and that is what keeps the
   hovered state inside SC 1.4.11 without a third pair. This step measures only
   2.32:1 against the masthead band, so it must never be the boundary; the
   stylesheet leaves the border in the soft accent on hover and darkens the
   background alone, so the edge a reader sees against the band is the same
   4.97:1 in both states. A hover rule that also repaints the border would put
   this token against the band and break that - see css/components.css. */
/* @pair --agora-color-text-inverse --agora-color-accent-strong 4.5 the primary control's label under the pointer */

/* Focus ring. Non-text contrast, so the threshold is 3:1 - WCAG 2.2 SC 1.4.11. */
/* @pair --agora-color-focus --agora-color-surface 3.0 focus ring on the page */
/* @pair --agora-color-focus --agora-color-surface-subtle 3.0 focus ring on a panel */
/* @pair --agora-color-focus --agora-color-surface-sunken 3.0 focus ring in a table header */
/* @pair --agora-color-focus-inverse --agora-color-surface-inverse 3.0 focus ring in the footer band */
/* @pair --agora-color-focus-inverse --agora-color-surface-inverse-raised 3.0 focus ring on a solid button */

/* Borders. A table rule separates data and is not decoration, so 3:1 applies. */
/* @pair --agora-color-border --agora-color-surface 3.0 cell rule on the page */
/* @pair --agora-color-border --agora-color-surface-subtle 3.0 cell rule on a panel */
/* @pair --agora-color-border --agora-color-surface-sunken 3.0 cell rule in a table header */
/* @pair --agora-color-border-strong --agora-color-surface 3.0 section rule on the page */
/* @pair --agora-color-border-strong --agora-color-surface-subtle 3.0 section rule on a panel */
/* @pair --agora-color-border-strong --agora-color-surface-sunken 3.0 section rule in a table header */

/* A solid control against what it sits on. Its own edge is what makes it a
   control rather than a coloured word, so it is checked at the non-text 3:1. */
/* @pair --agora-color-surface-inverse --agora-color-surface 3.0 solid button on the page */
/* @pair --agora-color-surface-inverse --agora-color-surface-subtle 3.0 solid button on a panel */

/* THE SAME TWO TOKENS THE OTHER WAY ROUND, and it is a different statement
   rather than a tidy duplicate. Above, a dark control stands on the light page
   and its own fill is what identifies it. Below, a LIGHT control stands on the
   dark masthead band - the primary button in the hero, and any field the site
   template places on that band - and there the pale fill is the boundary,
   because the border the element layer gives a control is a warm dark hairline
   that comes nowhere near 3:1 against navy. The stylesheet paints that border
   to match the fill for exactly this reason, so the pair that has to hold is
   the page surface against the inverse surface. The ratio is the same number
   in both directions; the combination being checked is not.
   WCAG 2.2 SC 1.4.11.

   ⚠️ AMENDED. "The primary button in the hero" IS NO LONGER ONE OF THESE, and
   the sentence above is kept rather than corrected in place so that the reason
   survives with it. That button is now filled with the soft accent - decided by
   [andres] on 2026-09-04, on the ground that the amber reads as the more
   inviting of the two - and its pair is declared with that token, several groups
   up. What still stands on this line is the hero's SECOND button on the days it
   is filled, and any field the site template places on the band: a search input,
   a select, anything core renders with the page surface as its ground. The pair
   is neither moot nor narrowed to nothing; it has lost one of its two named
   uses, which is worth saying out loud because a reader who greps for the
   primary button will land here first. */
/* @pair --agora-color-surface --agora-color-surface-inverse 3.0 a light control on the masthead band */

/* THE SAME INSTITUTIONAL TONE AS A DRAWING, on the one ground the two groups
   above do not already cover between them.

   The quick-access cards draw their marks in --agora-color-surface-inverse, and
   a card turns to --agora-color-surface-sunken under the pointer. The same
   token against --agora-color-surface and against --agora-color-surface-subtle
   is already declared two groups up, as a solid control - and a ratio is a
   property of two tokens rather than of what is drawn with them, so those two
   lines cover the resting card and a card standing on a tinted band without
   being restated. Only the sunken ground was missing, and a combination nobody
   declared is a combination nobody computed.

   IT IS HELD TO 3:1 AND NOT TO 4.5, by the argument the accent group makes: a
   mark that duplicates the label beside it is a graphical object under WCAG 2.2
   SC 1.4.11 rather than text. The threshold is not the interesting number here.
   The measured figure is 11.65 and the card's mark used to be drawn in the
   brand accent, which reaches 3.41 on the page surface - a value declared for
   rules, bars and edges, and six of them in a row is a default rather than an
   edge. The declaration is what makes the replacement checkable. */
/* @pair --agora-color-surface-inverse --agora-color-surface-sunken 3.0 a card's mark under the pointer */

/* Status text. Always rendered as text beside a written label, so 4.5:1. */
/* @pair --agora-color-status-success --agora-color-surface 4.5 success label on the page */
/* @pair --agora-color-status-success --agora-color-surface-subtle 4.5 success label on a panel */
/* @pair --agora-color-status-success --agora-color-surface-sunken 4.5 success label in a table header */
/* @pair --agora-color-status-warning --agora-color-surface 4.5 warning label on the page */
/* @pair --agora-color-status-warning --agora-color-surface-subtle 4.5 warning label on a panel */
/* @pair --agora-color-status-warning --agora-color-surface-sunken 4.5 warning label in a table header */
/* @pair --agora-color-status-error --agora-color-surface 4.5 error label on the page */
/* @pair --agora-color-status-error --agora-color-surface-subtle 4.5 error label on a panel */
/* @pair --agora-color-status-error --agora-color-surface-sunken 4.5 error label in a table header */
/* @pair --agora-color-status-info --agora-color-surface 4.5 information label on the page */
/* @pair --agora-color-status-info --agora-color-surface-subtle 4.5 information label on a panel */
/* @pair --agora-color-status-info --agora-color-surface-sunken 4.5 information label in a table header */
