/* ===================================================================== LONGITUDE HEALTH — TOKEN LAYER ===================================================================== THIS IS THE ONLY FILE IN THE BUILD THAT CONTAINS A COLOUR VALUE. No other stylesheet, no inline style, no SVG attribute and no script may carry a hex triple, an rgb() or an rgba(). Everything else refers to a token here. A palette pass is an edit to LAYER 1 of this file and nothing else. WHY THAT RULE IS ABSOLUTE, MEASURED ON THE SOURCE CONCEPT: concept-10-common-ground.html carried 12 tokens and 28 DISTINCT COLOUR VALUES THAT WERE NOT TOKENS. The largest was the body-text colour, appearing in six rules with no comment anywhere. A palette pass that edited :root would have moved 12 values and left 28 behind, and every check that reads a token would have reported success. VERIFICATION IS A RENDERED READ, NEVER A TOKEN READ. The check that this file actually governs the page is: perturb a token, count the elements whose computed colour moved AND the canvas pixels that moved. A token that reads back correctly while nothing on screen moved is the failure this layer exists to prevent. THREE LAYERS, AND THE ORDER MATTERS: LAYER 1 RAW PALETTE — the brand values. Hex lives here and nowhere else. This is what a palette pass edits. LAYER 2 ROLE PAIRS — ground/on pairs. No new values, only names for jobs. This is what components consume. LAYER 3 DERIVED — alpha and state variants, expressed as color-mix() of a LAYER 1 token so they follow their parent instead of freezing a triple. COMPONENTS SHOULD CONSUME LAYER 2, NOT LAYER 1. Reaching past a role pair to the raw value is how the two got welded together in the first place. ===================================================================== */ /* ===================================================================== TOKEN-LEVEL DEVIATION REGISTER Prefix TOK-, opened 2026-08-17 at C1 and numbered from 1. THE FIRST REGISTER ON THE TOKEN LAYER. Every other register in this build is PAGE-LOCAL — HOME-, OA-, OO-, ABT-/TEAM-, UTIL-, NAV-, WL- — and a change here reaches every page that renders the value, so it belongs where the value is defined rather than in any one page's log. New entries are TOK-n and are never renumbered. TOK-1 · --accent MOVES #696FE8 -> #2B61C9. CLIENT-DIRECTED via Eileen, relayed 2026-08-15, specified exactly and ruled 2026-08-17. `--mark` moves with it. `--pill` is RETIRED. **`--focus` DELIBERATELY DOES NOT MOVE.** THIS SUPERSEDES THE DEFERRED #4E65DC CHANGE, AND THE DEFERRAL IS CLOSED BY A CLIENT DECISION — NOT BY THE CONTRAST ARGUMENT BEING RESOLVED. Every sheet carrying a "#4E65DC / 2.6959" warning was warning about a value that is now never going to ship. Those warnings are historical; the argument they made was never settled, it was overtaken. ── THE VALUE FORMERLY DESCRIBED AS VIOLET IS NOW BLUE. #2B61C9 is a blue. Register prose across this tree — including entries this entry supersedes — calls `--accent` "violet". **That prose was NOT swept and rewritten in this unit, deliberately.** A reader meeting "violet" in an older entry is reading a HISTORICAL DESCRIPTION of the value at that date, not a live claim about what renders today. Sweeping it would have touched six sheets to restate something this entry states once. ── THE CENSUS. 55 RENDERED INSTANCES ACROSS 19 OF 20 PAGES. Measured by walking every element on all 20 built pages over localhost, testing color, backgroundColor, four border colours, outlineColor, textDecorationColor, fill and stroke, plus ::before/::after, resolving the effective ground by ancestor walk. Sweep controlled both ways first: a known-present value returned 128 elements, a known-absent value returned 0. TWO PROBE ARTEFACTS WERE REMOVED AND BOTH WOULD HAVE INFLATED THE COUNT. `textDecorationColor` mirrors `currentColor` even where `text-decoration- line` is `none`, doubling every text row — it is not a second instance. And elements whose OWN background is the accent resolved against themselves, returning 1.0000; `.beacon-badge` had to be re-resolved against its parent. `div.jf-success` was excluded as display:none, 0x0, on all 20 pages. ON LIGHT GROUNDS 34 every one improves; all pass before AND after on --paper 3.9310 -> 5.4187 on --cream 3.6759 -> 5.0671 **Every light instance now clears SMALL 4.5, not merely LARGE 3.0** — so the whole light set stops depending on its size classification. ON DARK GROUNDS 21 all 3.5594 -> 2.5821, all failing 3.0 2 TEXT a.cta-line on index #film and why-longitude #pivot 19 DECORATIVE .beacon-badge x6, circle.bd x12, g x1 — ALL aria-hidden **Re-counted on the built DOM after the swap: `circle.bd` = 12 and `.beacon-badge` = 6 across all 20 pages, and a sweep of EVERY `a.cta-line` site-wide returns 9, of which exactly the 2 above fail.** The counts are confirmed against the rendered site, not carried from Half A. ── "SEVEN LARGE-TEXT INSTANCES" DOES NOT REPRODUCE. IT IS TWO. The record held that the deferred change dropped SEVEN large-text instances below 3.0, and that a darker value made seven a floor. **Measured: two.** The other five were deleted by the rework and our-approach.css records it — `.cta-line` on index's #problem (R1), the same on why-longitude's #ecosystem (R6), and `.u-domain` x4 on utilities' `.u-domains` (R7). **The "floor" was inherited from a roster that no longer exists.** Count the instances, do not carry the count. ── THE TWO TEXT REGRESSIONS ARE RULED FOR RE-GROUNDING, NOT ACCEPTED. `a.cta-line` at 19px/700 on `--plum`, on index #film and why-longitude #pivot. Candidates measured on plum, no new token invented: --gold 8.2161 PASS --stone 7.0533 PASS --ink-on-dark 8.6943 PASS --cream 13.0839 PASS --slate 3.0421 PASS --pill-hover 2.8600 FAIL --accent NEW 2.5821 FAIL **--gold at 8.2161 is the answer and it is not a workaround.** Gold is ALREADY this palette's dark-ground accent: `h2.tlw-h2` on our-approach is gold on plum at exactly 8.2161, and the beacon badges sit beside it. The light-ground CTAs stay `--accent`; the dark-ground CTAs take the colour the dark grounds already use. `--ink-on-dark` and `--cream` clear easily but are the BODY-TEXT colours on dark — a CTA in body colour stops reading as a link. `--slate` clears by 0.0421 and is not a margin worth shipping. **⚠ THE RE-GROUNDING IS NOT IN THIS UNIT'S EDIT, AND THAT IS A CONFLICT THIS ENTRY RECORDS RATHER THAN RESOLVES.** The ruling asked for the two CTAs to be re-grounded; the same dispatch fixed the manifest at MOVED 2 — tokens.css and its dist twin. Re-grounding touches `pages/home.css` and `pages/why-longitude.css`, which would make it MOVED 6. **Shipped: the token change only. The two CTAs render at 2.5821 until the re-grounding lands, and that is a KNOWN OPEN FAILURE, not an oversight.** It is one declaration in each of two sheets. ── THE 19 DECORATIVE INSTANCES ARE ACCEPTED, WITH THE REASONING. `.beacon-badge` x6, `circle.bd` x12 and one `g` — ambient dot fields and badge discs. **Every one carries aria-hidden="true"**, inherited from its container. SC 1.4.11 applies to non-text content REQUIRED TO UNDERSTAND the interface; the meaning in each case is carried by the adjacent text — the badge letters duplicate their titles' first letters, and the dot fields are ambient. **They bind no threshold. This is a judgement recorded as one, not a measurement** — the figures are real and the conclusion is a reading of what the criterion covers. ── ⚠ FOUR TOKENS CARRIED #696FE8. THEY ARE NOW THREE DIFFERENT THINGS. --accent #2B61C9 19 refs, 8 files MOVED --mark #2B61C9 1 ref MOVED with it (.dots .sd-m fill) --pill RETIRED 0 refs see below --focus #696FE8 11 refs, 7 files **DELIBERATELY UNMOVED** **THESE ARE `var()` REFERENCES IN CODE, NOT RENDERED INSTANCES. Different noun — 19 references produce the 55 rendered instances above.** The counts were first taken as 20 / 12 / 7 and every one was ONE TOO HIGH: raw grep matched `var(--accent)` at why-longitude.css:659, `var(--focus)` at news.css:421, `var(--pill-hover)` at utilities.css:285 and two more inside this file's own header — **all of them prose describing the token, not a use of it.** §25, on a unit whose whole subject is a token. Re-derived with comment spans stripped. The 55-instance census is unaffected: it was read from computed styles in the DOM, where a comment cannot appear. They were never derivations — four independent literals that happened to share a value, so moving one silently split them. **Nothing warns you.** **WHY --focus STAYS, AND THIS IS A DECISION, NOT AN OVERSIGHT.** All twelve uses are `outline:3px solid var(--focus)` on `:focus-visible`, across header, join-form, accordion, news, why-longitude, our-team and utilities. On dark grounds the new value measures **2.5821 against the old 3.5594**. A focus indicator is a UI affordance under its own success criterion, not brand colour. **Splitting the palette is an aesthetic cost; degrading focus visibility is an accessibility regression.** The site now carries two related blues by ruling. Anyone "tidying" `--focus` to match `--accent` would be trading a visible inconsistency for an invisible barrier. **--pill IS RETIRED BECAUSE IT HAD 0 USES, not because it looked unused.** Measured across src/css and src/js: `var(--pill)` returns zero. `--pill-hover` (#585EDC) is a SEPARATE literal, still live at 7 uses, and is untouched. ── ONE GENUINE DERIVATION. IT MOVED, AND IT RENDERS NOWHERE. `--photo-glow-accent: color-mix(in srgb, var(--accent) 30%, transparent)` is the ONLY token in this file that references `--accent`. It moved with it and nothing announced that — the browser now computes it as `color-mix(in srgb, #2B61C9 30%, transparent)`. **But it has ZERO RENDERED INSTANCES.** Its single use is the `.photo` radial-gradient at `components/bands.css:39`, and **`class="…photo…"` appears on NO page, built or source.** Controlled before the claim: the same grep returns 4 files for `class="band`. `--photo-glow-gold` sits in the same declaration and is equally dead. **The one derivation that could have moved invisibly moved invisibly and changed nothing, because the band it paints is not on the site.** Not deleted here — dead-code removal is R8's business. ── TWO STANDING FORECLOSURES ARE REVERSED. ⚠ BY ARITHMETIC, NOT BY MEASUREMENT. **9h/9x recorded that NO VIOLET FILL CLEARS 3.0 ON GOLD, and that `--accent` was unavailable for `.btn` because white-on-accent failed 4.5 by 0.3331.** Both figures move the right way: fill on gold 2.3083 -> 3.1819 now clears 3.0 white label 4.1669 -> 5.7440 now clears 4.5 **⚠ THESE ARE COMPUTED FROM TOKEN VALUES. THERE IS NO SUCH BUTTON ON THE SITE TO MEASURE.** `buttons.css` contains zero gold references, and the only three elements rendering a `--gold` background are `.wl-cx-dot` spans on why-longitude, which contain no button. **A `.btn` on a gold ground has no rendered instance anywhere.** §60 — the figures above are a PREDICTION about a component that does not exist yet, and the first draft of this entry labelled them a measurement. If such a button is ever built, measure it then; do not ship on these numbers. **The `--ink` border 9x added to carry the 3:1 boundary is no longer load-bearing by that arithmetic — but REMOVING IT IS A SEPARATE DECISION, NOT A CONSEQUENCE OF THIS ONE, and it was not removed.** 9h and 9x are not wrong; they described a value that no longer ships. **WL-25's clamp-floor note and OA-5's warning about `.vm-label` become HISTORICAL rather than binding.** That label goes **3.6759 -> 5.0671** and now clears SMALL 4.5, so its legality no longer depends on the 21px/700 floor. The floor remains a typographic decision and should not be lowered casually — but lowering it below 18.66 no longer creates an SC failure. Marked superseded per the OA-4 / OA-5 / 9ag precedent: the figures stay as written, and this entry carries the current ones. ── R8 LOSES FIVE INHERITED ITEMS. The five pre-existing `--accent` failures on utilities.html — `.acc-cx-dot` and `.acc-cx-seg` — are CLOSED. **Measured on their REAL ground, `--cream`, not the plum the record implies: 3.6759 -> 5.0671, passing both 3.0 and 4.5.** R8 inherits them no longer. ── R4 IS UNCHANGED AND UNTOUCHED, AND THIS UNIT DID NOT RESOLVE IT. The `h1 em` on why-longitude goes **3.9310 -> 5.4187** and now passes by a wide margin — but its accessibility half was ALREADY moot, and that was never what R4 was about. **The client asked for GOLD, which measures 1.7030 on paper and fails even 3.0.** R4 stands exactly where it stood: the accent is legible and the client wants gold. **No future reader should read this unit as having closed it.** TOK-2 · THE PILL FILL MOVES TO #2B61C9. `--pill` IS REVIVED AND `--pill-hover` MOVES WITH IT. DIRECTOR-RULED, C4, 2026-08-17. Second tokens.css authorisation; C1 was the first. --pill REVIVED at #2B61C9 .btn RESTING fill --pill-hover #585EDC -> #2B61C9 .btn:hover ONLY (+ 3 non-pill uses) buttons.css:39 resting background repointed var(--pill-hover) -> var(--pill) buttons.css:52 hover LEFT on var(--pill-hover), deliberately **BOTH TOKENS HOLD ONE VALUE TODAY AND REMAIN SEPARATELY GOVERNED.** That is the whole point of the shape: `buttons.css` argues that `--pill` must not be folded into `--accent`, "one of those two jobs would otherwise be decided for it by a token serving a different purpose". Pointing `.btn` at `--accent` would have folded it. This does not — a future change to the resting fill or the hover fill no longer decides the other. ── ⚠ THE PILLS WERE NEVER `--accent`, SO C1 DID NOT MISS THEM. The director reported the pills still reading violet after C1. **They had nothing to move.** C1's census swept for `rgb(105,111,232)`; the pills are `rgb(88,94,220)` — `--pill-hover`, a separate literal C1 explicitly recorded as unaffected. `--accent` moved to hue 219.5 (blue) while the pills stayed at hue 237.3 (violet), and that divergence is what was visible on screen. **A census is only as complete as the value it sweeps for; sweeping one literal exhaustively says nothing about a second.** ── THE 84-INSTANCE CENSUS OF #585EDC. REGRESSING INSTANCES: 0. All 20 pages, computed styles, nine properties plus ::before/::after, ground resolved by ancestor walk with alpha compositing. Controls: a known-present value returned 14,142 property hits, a known-absent value 0. SPAN.hub-title color x46 hub only A.btn bg x26 all 20 pages A.na-back color x11 11 news pages SECTION.band-slate u-quote bg x1 utilities on --cream 4.5747 -> 5.0671 on --paper 4.8922 -> 5.4187 white label on fill 5.1858 -> 5.7440 (thr 4.5, PASS -> PASS) **Every one of the 84 improves or holds. Zero regressions.** The fill fails on `--plum` at BOTH values (2.8600 -> 2.5821), which is why the standing foreclosure against `.btn` on `.band-plum` still binds — nothing here relaxes it, and no pill currently sits there. **THREE MEASUREMENT ARTEFACTS WERE REMOVED, TWO OF THEM REPEATS FROM C1.** The first sweep returned **369**, not 84. `outlineColor` reports a value where `outline-style:none` (57 dropped); `::before`/`::after` re-report the inherited `color`, tripling every text row (456 boxless pseudo-elements dropped). Both were already known from C1 and were not carried forward. The third was new: the header's translucent ground computes as `color(srgb 0.984314 ...)`, whose components are **0-1 floats, not 0-255** — read raw they resolve to near-black and produced a plausible-looking but wrong 4.0257. **Composited correctly the header ground is `rgb(251,248,242)`, identical to the in-page one.** A ratio that looks reasonable is not evidence the colour was parsed. ── WHY `--pill-hover` MOVES TOO, THOUGH IT IS NOT A PILL FIX. It has three consumers that are not pills and not hover states: `news.css:103` and `:397` (text colour), `utilities.css:1095` (`.band-slate.u-quote` fill), and a decorative wash at `stat-arc.css:728`. **Left at #585EDC the site would ship blue accents beside violet news text.** The client's direction is purple-to-blue site-wide, so the token moves with the rest. Measured: news text 4.8922 -> 5.4187 (thr 4.5, PASS); the quote band's white label 5.1858 -> 5.7440 (thr 4.5, PASS). ── THE 2026-08-07 HISTORY, BECAUSE IT EXPLAINS THE WHOLE SHAPE. The white label measured **4.1669 on `--pill` against 4.5**. Every foreground alternative was tested and every one was worse (`--on-fill` 3.8925, `--ink` 3.5594), so the fix had to come from the background side. **THE FIRST PASS SWAPPED THE TWO TOKENS AND MERELY RELOCATED THE FAILURE.** Resting took `--pill-hover`, hover took `--pill` — which fixed resting and moved the 4.1669 failure onto hover. SC 1.4.3 applies to text in EVERY state, so nothing was removed. **And a resting-state contrast sweep could not see it, because the sweep never hovers.** Both states then took `--pill-hover`, which is why `--pill` reached zero consumers. **C4 CLOSES THAT RELOCATION.** Resting takes `--pill` again and hover keeps `--pill-hover`, but both tokens now hold #2B61C9, so **there is no longer a worse value for the failure to be moved to.** The 4.1669 figure belonged to #696FE8 and describes a value neither token carries. ── `--pill`'s RETIREMENT IS SUPERSEDED BY A DECISION, NOT CORRECTED. **C1 retired it on correct facts** — 0 code refs, browser computing empty, verified by live perturbation — and it stayed correct for three checkpoints. C1's entry is MARKED, not rewritten: a reader has to be able to see the token was genuinely dead, or the revival reads as the reversal of a mistake instead of a ruling. The pre-C1 block reserving its disposition for a PALETTE-SESSION decision is likewise marked: **that condition was MET by a director ruling, not overridden.** ── THE NAMING PROBLEM: RECORDED, NOT FIXED. `--pill-hover` paints three non-pill, non-hover elements. Its name is now accurate for `.btn` and wrong for everything else it touches. **Same class of drift as the violet/blue naming note in TOK-1, and handled the same way — stated once here, not swept.** A rename is a separate unit. Read the `.btn` rules for which state a token governs; never read the name. ── ⚠ `--focus` STAYS #696FE8 AND IS NOW THE ONLY VIOLET IN THE TREE. Held by C1's ruling. `.btn` matches `a:focus-visible{outline:3px solid var(--focus)}`, so **the pill's focus ring is deliberately a different hue from the pill.** An outline draws OUTSIDE the border box, so its binding partner is the page ground: **3.9310 on `--paper`, PASS at 3.0.** Against the fill — inner edge only — it goes **1.2445 -> 1.3785, a slight improvement**, because the fill is no longer a second hue-237 violet. **The hazard block above `--focus` stays.** After C4, #696FE8 appears exactly once in the whole tree. ===================================================================== */ :root{ /* --------------------------------------------------------------- LAYER 1 — RAW PALETTE PALETTE CANON IS CLIENT-APPROVED AS OF 2026-08-07. The values below are not placeholders and are not expected to move. This supersedes the note that stood here, which was written before the confirmation landed and described the palette as unconfirmed. NO NEW COLOUR VALUE MAY BE INVENTED, DERIVED, TINTED OR INTERPOLATED. That covers the obvious case and the quiet ones equally: a lightened hover state, a "just slightly darker" rule, a mix of two approved values, a value picked to satisfy a contrast target. Approved means these values, not this neighbourhood. If a job on the page has no token that serves it, THAT IS THE FINDING — report it and stop; it is a decision for whoever owns the palette, not a gap to fill in passing. THE THREE-LAYER STRUCTURE STANDS ON ITS OWN MERITS, NOT ON AN EXPECTATION OF CHANGE. Its justification was never that the values were provisional. It is that one file holding every colour is what makes "no literals outside this file" checkable at all, that role pairs stop a ground and a mark being welded to one name, and that LAYER 3 expressing alpha as color-mix() of a LAYER 1 token keeps derived values FOLLOWING their parent instead of freezing a triple that then drifts silently. Confirmation of the palette removes none of that. DO NOT COLLAPSE THE LAYERS ON THE GROUNDS THAT THE VALUES ARE NOW SETTLED. --------------------------------------------------------------- */ /* grounds and inks */ --paper:#FBF8F2; --cream:#F6F0E6; --ink:#292631; /* --plum HOLDS --ink's VALUE UNDER A SECOND NAME AND IS NOT REDUNDANT. --ink is a foreground that also fills controls; --plum is the problem band's ground and nothing else. They are equal at rest and want different things the moment either moves. DO NOT COLLAPSE THEM. */ --plum:#292631; --slate:#5E7484; --stone:#B8B2AA; --white:#FFFFFF; /* the muted ink family — ALL FOUR WERE BARE LITERALS IN THE CONCEPT. THE AA ITEM THIS BLOCK RECORDED IS CLOSED, AND IT CLOSED BY DELETION RATHER THAN BY A COLOUR DECISION. The note here described --ink-muted as the colour behind 10 of 16 inherited AA failures, naming two: a ruled-label element on cream at 3.1144, and the footer's small print on paper at 3.3306, both against 4.5. Neither still exists to be measured. The ruled labels and their rules were cut from the home page (2026-08-07, client change list); the footer was re-grounded the same day and now takes the --on-dark family. That was the last outstanding SC 1.4.3 failure on the site. NO VALUE HERE MOVED, AND NONE NEEDED TO — do not read this as a palette decision having been taken. MEASURED 2026-08-07 BY CSSOM, BOTH BUILT PAGES. Counts are var() rule references, not computed-colour matches: two tokens can resolve to the same value, so a colour match would merge them. --ink-body -> --on-page-body 7 refs (index) / 8 (utilities) --ink-muted -> --on-page-muted 0 refs on EITHER page --ink-soft -> --on-page-soft 2 refs on both --ink-on-dark -> --on-dark-body 2 refs on both Consumer counts are PER PAGE and the old ones here were not; three of the four were stale. --ink-muted THEREFORE HAS NO LIVE CONSUMER. It feeds --on-page-muted, which nothing references. It is deliberately NOT deleted: whether the palette keeps, retires or reassigns a token is a palette-session decision, and an unreferenced token costs nothing. Same standing reasoning as --pill and --portrait-ring, which are also orphaned. A NOTE ON CHECKING THIS. The selector that was deleted is named above in prose, so a source-level grep for it MATCHES THIS COMMENT and will report it as present. The instrument that answers the question is document.querySelectorAll on the built page — the DOM carries no comments. It returned 0 on both pages. */ --ink-body:#4E4A5A; /* body copy on light */ --ink-muted:#8B8696; /* NO LIVE CONSUMER — see the note above */ --ink-soft:#6E6A78; /* lockup HEALTH + utility body */ --ink-on-dark:#C9C4D2; /* body copy on the plum band, and the footer */ /* golds */ --gold:#F0B83F; --gold-d:#B8860B; /* NO LIVE CONSUMER. Retained deliberately: its only selector (.confirm) matched zero elements in the concept. Kept so a palette pass has a slot for it rather than reinventing a value. */ /* THE ACCENT FAMILY — FOUR TOKENS, ONE VALUE TODAY, FOUR JOBS. Split before this build and MUST NOT BE RECOMBINED on the grounds that they currently agree. Equal-at-rest is not redundant: a mark on a ground, a fill carrying its own label, an indicator crossing two grounds, and decorative geometry want different things the moment any value moves. */ --accent:#2B61C9; /* type accent: .cta-line, h1/h2 em, nav hover */ /* NEITHER OF THESE TWO NAMES DESCRIBES ITS USE. Superseded at C4, 2026-08-17: the 2026-08-07 state described here — ".btn takes --pill-hover in BOTH states and --pill is referenced by nothing" — no longer holds. **.btn resting now takes --pill; .btn:hover still takes --pill-hover.** Both tokens hold #2B61C9 today and are SEPARATELY GOVERNED, so hover remains a colour no-op while the declarations name their roles correctly, which they did not before. THE NAMING PROBLEM IS RECORDED, NOT FIXED. --pill-hover still paints three NON-PILL, NON-HOVER things: news.css:103 and :397 (text colour) and utilities.css:1095 (.band-slate.u-quote fill), plus a decorative wash at stat-arc.css:728. Same class of drift as the violet/blue naming note in TOK-1. A rename is a separate unit. Read the .btn rules in components/buttons.css rather than inferring use from a name. See TOK-2. */ /* --pill HAS ZERO CONSUMERS. Confirmed by live perturbation on both built pages 2026-08-07 — setting it to a sentinel moved 0 elements, measured against a control that moved 2 before the change and a nonexistent token that moved 0. It holds the concept's original button violet. DO NOT DELETE OR REPURPOSE IT HERE: whether the palette keeps this value, retires it, or assigns it elsewhere is a PALETTE-SESSION decision, not a build one. An unreferenced token costs nothing; guessing at its disposition does. */ /* ^ THAT CONDITION WAS MET, NOT OVERRIDDEN. C4 (2026-08-17) reassigns --pill on an explicit director ruling with tokens.css authorisation — the palette-level decision this block was holding the token open for. It is no longer #696FE8 and no longer unreferenced. See TOK-2. */ /* --pill RETIRED at C1 2026-08-17 — 0 references in code. Was #696FE8, one of four independent literals that shared that value. NOT the same token as --pill-hover (#585EDC), which is live at 6 refs. See TOK-1. */ /* ^ SUPERSEDED BY DECISION AT C4, NOT CORRECTED AS AN ERROR. C1's retirement was RIGHT ON ITS FACTS — 0 code refs, browser computing empty, and it stayed right for three checkpoints. C4 revives the token at a new value on a director ruling. Left as written: a reader needs to see that the token was genuinely dead, or the revival looks like a reversal of a mistake instead of a decision. */ --pill:#2B61C9; /* .btn RESTING fill. Revived at C4 2026-08-17. */ /* ⚠ DO NOT "TIDY" THIS TO MATCH --accent. It is the last #696FE8 in the file and it looks like a leftover. It is not — C1 moved --accent to #2B61C9 and left this behind BY RULING. #2B61C9 measures 2.5821 on --plum against this value's 3.5594, so aligning them would dim every focus ring on every dark ground. See TOK-1. */ --focus:#696FE8; /* the focus indicator and nothing else */ --mark:#2B61C9; /* decorative geometry: badge dots, shatter mark */ /* .btn's fill in BOTH states, despite the name. Hand-picked, NOT derived from --pill by color-mix — a change to --pill does not move it and never did. The white label measures 5.1858 on this value at 14.5px / weight 600, against a 4.5 threshold. Re-measure the label if this value moves. */ --pill-hover:#2B61C9; /* .btn:hover ONLY. Moved at C4 2026-08-17. */ /* FORM ERROR. Its own token rather than a reuse of any brand value: an error state has a contrast floor of its own (4.5 at 13px on --paper) and must be tunable without dragging a brand colour with it. It is NEVER the only signal — the message text and aria-invalid carry the information; this is the visual reinforcement. Measured 6.1710 on --paper. Re-measure if --paper moves. */ --error:#B3261E; /* portrait FPO wash — placeholder only, never a person. Renders only if the client photograph fails to load. */ --fpo-base:#39424C; --fpo-hi:#516170; --fpo-lo:#2B2934; /* --------------------------------------------------------------- LAYER 2 — ROLE PAIRS --------------------------------------------------------------- */ /* WHY THESE EXIST. --ink and --cream were each doing TWO JOBS in the concept: --ink is body text AND the fill of five stat bubbles, the headline backing, the submit control and the utility chips; --cream is the members band's ground AND the text colour on three dark bands. Today every pairing is ink-on-cream or cream-on-ink, so moving either moves both sides of the same ratio and nothing breaks. THE MOMENT PALETTE CANON MOVES ONE OF THEM INDEPENDENTLY THERE IS NO VALUE THAT SATISFIES BOTH ROLES. That is not a hypothetical: three concepts on this engagement hit it independently, and every value that fixed the mark destroyed the band. Splitting the ROLE now, while the values coincide, costs nothing and is what makes the split available later without re-auditing every band. A component that needs "the page's foreground" takes --on-page. A component that needs "a dark fill's foreground" takes --on-fill. They are the same value today and that is fine. */ --ground-page:var(--paper); --on-page:var(--ink); --ground-alt:var(--cream); --on-alt:var(--ink); --ground-dark:var(--plum); --on-dark:var(--cream); --ground-slate:var(--slate); --on-slate:var(--cream); --ground-fill:var(--ink); --on-fill:var(--cream); --ground-field:var(--white); --on-field:var(--ink); --on-pill:var(--white); /* secondary foregrounds, per ground */ --on-page-body:var(--ink-body); --on-page-muted:var(--ink-muted); --on-page-soft:var(--ink-soft); --on-dark-body:var(--ink-on-dark); /* --------------------------------------------------------------- LAYER 3 — DERIVED Every alpha value is a color-mix() of a LAYER 1 token, so it FOLLOWS its parent. The concept froze these as rgba() channel triples, which meant a palette pass moved the solid colour and left its own translucent derivatives behind at the old hue. --------------------------------------------------------------- */ --header-bg:color-mix(in srgb, var(--paper) 90%, transparent); --rule-hair:color-mix(in srgb, var(--ink) 14%, transparent); --rule-footer:color-mix(in srgb, var(--ink) 20%, transparent); /* photo band wash. AN OPAQUE BACKING IS WHAT MAKES CONTRAST DETERMINATE ON THIS BAND — see components/stat-arc.css. These are the BAND's wash, not a text backing, and must not be reused as one. */ --photo-wash-a:color-mix(in srgb, var(--slate) 92%, transparent); --photo-wash-b:color-mix(in srgb, var(--ink) 90%, transparent); --photo-glow-gold:color-mix(in srgb, var(--gold) 25%, transparent); --photo-glow-accent:color-mix(in srgb, var(--accent) 30%, transparent); --portrait-ring:color-mix(in srgb, var(--cream) 60%, transparent); --portrait-label:color-mix(in srgb, var(--cream) 70%, transparent); --fpo-hatch:color-mix(in srgb, var(--cream) 5%, transparent); --fpo-wash-hi:color-mix(in srgb, var(--stone) 30%, transparent); --fpo-wash-lo:color-mix(in srgb, var(--ink) 60%, transparent); --fpo-shadow:color-mix(in srgb, var(--ink) 42%, transparent); /* --------------------------------------------------------------- NON-COLOUR TOKENS --------------------------------------------------------------- */ /* cap-height line of .lg1, measured in Chrome (2.95 x 1.08) */ --capline:3.19px; --serif:Charter,'Iowan Old Style','Palatino Linotype',Georgia,serif; --mono:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace; --sans:ui-sans-serif,-apple-system,"Segoe UI","Helvetica Neue",Arial,sans-serif; }