Lane G — palette engineering: are the six roles sound, distinguishable, and built the way the best systems build them?

Status: Reference · Date: 2026-10-09 · Evidence appendix for colour-system-research-2026-10-09.md. Scan scripts referenced by path live in the session scratch directory (not committed); every number comes from the shared helper whose formulas are reproduced in §3.

Date: 2026-10-09 · Author: lane G · Evidence for docs/design/research/visual-benchmark-2026-10-08.md §3 and §5 · English only · repo files were read, never modified. How to read the numbers. Every colour number (ΔE2000, Lc, WCAG ratio, L*/C*/h, OKLCH, CVD-simulated hex) comes from bench/colour_lab.py (functions apca, wcag, lab/lch/oklch, de2000, cvd = Machado 2009 severity 1.0, cvd_table). Searches and scans are scripts in bench/lane_g/ that import the helper (lg.py adds only OKLCH→hex conversion and a gamut check). Three computations need a model the helper does not have and are marked: CAT16 adaptation (adapt.py, matrix verified against colour-science 0.4.7), Helmholtz–Kohlrausch Γ (colour-science 0.4.7, Nayatani 1997), and the peer hexes (read from published npm packages). APCA font-size minima use the helper's published-LUT version (coordinator note 2026-10-09; Lc values are unchanged). Vocabulary. Lc = APCA lightness contrast (|Lc|; positive = dark text on light). "n/p/d/t" = ΔE2000 under normal / protan / deutan / tritan vision. "OI parity" = the worst pairwise ΔE2000 of Okabe–Ito's own palette under the same simulation (21.7 / 12.3 / 11.5 / 11.1, table D).

---

0. The six verdicts that matter

1. Brand #006D37 is sound and stays; success #14804A is not. Brand vs success is ΔE2000 6.7 / 7.0 / 6.4 / 7.0 under all four visions — the same colour at a glance — and #14804A is also under the APCA 75 line on white (Lc 74.0). The ΔE ≥ 20 target is not reachable while staying green: inside the text-contrast window (on-white Lc ≥ 75, white-on Lc ≥ 60) the best min(n,p,d) against brand is 11.3 at OKLCH hue 170, 14.1 at 180, and only crosses 20 at hue ~195 (blue-teal) — table P. Final rule: meet Okabe–Ito parity (≥ 12 under protan/deutan, ≥ 20 normal) for every pair that can share a component, and carry meaning with glyph + label. Final values — two tiers, as Polaris and Atlassian split text / icon / fill: $success #00796A for icons, fills and large figures (dark #75D1C6): ΔE vs brand 13.4 / 14.7 / 17.2, vs error 55.5 / 16.2 / 24.2 (was 61.9 / 13.2 / 9.5), on-white Lc 75.9, white-on Lc 81.3; and $successText #014B40 for small text and badge text (coincides with Polaris' text-success, which is where the candidate came from; scan-confirmed; L* 27.9, Lc 93.0 on white, 85.7 on the tint): ΔE vs brand 16.3 / 17.6 / 18.1, vs the neutral badge text 23.6 / 13.6 / 13.1, vs error 51.5 / 12.9 / 28.3. The two-tier split is forced by a second finding: a plain teal (#00796A) as text turns into the same mid-grey as the neutral badge under deutan (ΔE 3.0, bg 0.7), whereas the deep teal does not (13.1). #00796A sits at the hue of the shipped "open/recruiting" accent badge, so success and accent become one teal and brand is the only grass-green.
2. The luminance-only separation the brief asks for is impossible inside the text window; the fill tier supplies it. Text-capable mid-tones must sit at L* ≲ 47 (Lc 75 on white falls at L* 46.7); brand 39.9, error 42.4, success fill 45.2 differ by ≤ 5.3 L*. A ΔL* of 19.5–21.9 gives only 2.0–2.2:1, not the 3:1 luminance difference SC 1.4.1 accepts; 3:1 against error needs L* ≥ 75. Only warning as a bright fill with ink text (#F5B301, L* 77.1) does it: vs error WCAG 3.19:1, ΔE 46.4 / 40.4 / 29.5 (current deadline-vs-live banners: 27.6 / 12.1 / 4.2). Red/green pairs outside that tier are separated by hue on the blue–yellow axis (teal vs red) plus glyph, as Okabe–Ito prescribe.
3. Optic lime #C6E84B is correct as a low-area level accent, but the stated reason is half wrong. It is the brightest role (Y 0.702, ink numerals Lc 84.1) — that is V(λ)/luminance. Helmholtz–Kohlrausch does the opposite of what the brief assumes: Nayatani Γ (VCC) is 0.930 for optic, among the lowest of any role (warning amber 0.890; error 1.311, info 1.590). It must never be text or a bare shape on white (Lc 19.1 on white, 13.2 on the basement — under APCA's Lc 15 invisibility line), and it merges with warning amber under deutan (5.8) and turns pale grey-green under tritan (#D1DCC9).
4. The content hues collide with the semantic ones; fix three pairs, not "add colours". Court grass #006633 vs brand 2.3 (identical); court clay vs grass protan 1.7; clay #B84A2E vs error 6.6 (deutan 2.9); tiers: bronze vs grandmaster deutan 3.4, gold on white 2.38:1. Proposed: one 5-series set that also is the court trio (#3566B8, #8D3D00, #83971F, #883170, #089AC3; n/p/d/t min 22.6 / 15.1 / 14.9 / 8.8, all ≥ 3:1 on white and basement), tiers bronze #B2580B, gold #D9A00F, grandmaster #9A0F3A (min 25.1 / 22.7 / 18.1 / 11.4 vs 14.8 / 10.0 / 3.4 / 8.7 now). Strictly equal-L is not viable: the best equal-L 5-set at OKLCH L 0.60 has min ΔE 11.5 / 9.3 / 8.2 / 5.2; staggering L by ±0.08 lifts deutan to 15.0 — Okabe–Ito and Tol also stagger (L* spread 43 / 36).
5. Neutrals are inside the industry envelope; the tint cannot shift the brand. Canvas C* 1.5, basement 3.2, track 3.0, hairline 4.2 vs Radix sage steps 1–8 C* 0.9–1.8, Wise's green-tinted surface 2.8, iOS grouped 2.6. CAT16 appearance shift of #006D37 on the basement: ΔE2000 0.76 (full von Kries, upper bound) / 0.55 (CAM16 D = 0.85) — at or below the ~1 level conventionally treated as a just-noticeable difference. Recommended ceiling: Lab C* ≤ 4.5 for L* ≥ 88, hue 145–165.
6. Dark parity fails in three places today: dark $error text #D4736E is Lc 40.5 on the card (the gate has no APCA watch on $error; helper LUT: 58 px at weight 400); dark success #68BF8E vs dark primary #70D5A2 is ΔE 5.5 (the same collision as light); and the dark deadline/live banners are ΔE 3.1 under deutan. Proposed dark: error #FA9B94 (Lc 60.4), success #75D1C6 (ΔE vs primary 12.0 / 13.9 / 16.3), warning text #E1B75C (Lc 64.9), dark deadline banner = amber #E2B13C with ink (ΔE vs live 62.2 / 56.5 / 55.9), hero block #1E5236 (optic 6.50:1).

---

1. Verdict table per role

Legend: keep = no change; hex pairs are light / dark. Numbers are helper outputs; full tables in §2.

Role	Current	Issue	Evidence (URL)	Final (light / dark)	Key numbers
Brand solid (CTA, hero block, toggle on)	#006D37 / #70D5A2	none; L* 39.9 / 78.3 are the Material tone-40 / tone-80 logic	Material primary tone 40 light, 80 dark, on-primary 100/20: color_spec_2021.ts	keep #006D37 / #70D5A2	white-on Lc −86.6 (6.48:1; 16.3 px @400, 14 px @700); dark ink-on #70D5A2 Lc 70.4; #70D5A2 on card Lc −67.8
Brand as text	$primary on white/basement	fine for links ≥ 16 px/500; the helper LUT wants ≥ 15.8 px @500 on white (Lc 81.5) and 16 px @500 on basement (Lc 75.7)	APCA nutshell (Lc 75 = body-column minimum)	keep; inline links only (budget §5)	on basement Lc 75.7, WCAG 5.95
Success	#14804A / #68BF8E	ΔE vs brand 6.7/7.0/6.4/7.0; on-white Lc 74.0 < 75; vs error deutan 9.5; dark vs dark primary 5.5	Polaris success fill #047B5D (h168) and text #014B40, brand fill #303030 — @shopify/polaris 13.9.5 styles.css; Seed positive #079171 (h170) — daangn/seed-design color.yaml; Wise positive #054D28 — neptune-tokens; Okabe–Ito: bluish green is the recommended green so it is not confused with red or brown — jfly.uni-koeln.de/color; Polaris: "Avoid using green to entice merchants or to draw unnecessary attention" — shopify.dev	$success #00796A / #75D1C6 (OKLCH 0.516/0.094/180) for icon, fill, figure; $successText #014B40 / #75D1C6 for text; tint #E0F8F3 / #14433E	#00796A: vs brand 13.4/14.7/17.2/6.0; vs error 55.5/16.2/24.2/58.9; on-white Lc 75.9 (17.9 px @400); white-on Lc −81.3. #014B40: Lc 93.0 on white (15.7 px @400), 85.7 on tint (9.07:1); vs brand 16.3/17.6/18.1/10.4; vs neutral badge text 23.6/13.6/13.1/18.5. Dark #75D1C6: Lc 67.6 on card; vs dark primary 12.0/13.9/16.3/1.8; vs dark neutral text 21.9/6.7/3.0
Warning	#946200 (text, fill, icon) / #D4A849	text-on-fill and fill both at L* 45.7, same lightness as error 42.4: deutan 4.2; deadline vs live banner identical to deutan; peers use a bright fill + dark ink	Polaris warning fill #FFB800 + text #5E4200; Atlassian warning.bold #FBC828 with inverse text #292A2E; Carbon yellow30 #F1C21B; Wise #FFD11A — packages and Carbon, @atlaskit/tokens 20.4.0, @transferwise/neptune-tokens 8.20.2	keep $warning #946200 for icons/graphics (5.24:1); add $warningText #6B4A00 / #E1B75C; add $warningFill #F5B301 / #E2B13C + $onWarningFill #0D100E	fill vs error 46.4/40.4/29.5/30.7 (WCAG 3.19:1); text vs error 28.9/7.4/10.3/17.4 (was 27.6/12.1/4.2/12.0); text on tint Lc 81.4 (was 69.4), 7.36:1 (was 4.78); ink on fill Lc 68.8
Error	#B8332E / #D4736E	light is right: OKLCH h 27.0 vs peers 24.5–28.8; dark text is Lc 40.5	Polaris critical #C70A24 (h24.5), Wise #CB272F (h24.8), Carbon red60 #DA1E28 (h25.9), Atlassian #C9372C (h28.8), M3 error hue 25 chroma 84 — core_palette.ts	light keep; dark #FA9B94 (M3 tone-80 #FFB4AB is Lc 70.7 if the protan/deutan price vs dark primary is accepted)	light white-on Lc −83.6, on-white 78.3; dark Lc 40.5 → 60.4 (min px @500 46.5 → 20.8). Irreducible: brand vs error protan 6.5 for every red tested (table B2, scan in §2.2)
Info	#4B0082 / #C7B5F2	none in the set; purple is shared with tier master (h 302 vs 315)	—	keep	on-white Lc 97.2 (12.95:1); vs brand 50.5/50.6/44.3/36.7; dark Lc −65.7
Optic (level accent)	#C6E84B	zero consumers today; not a surface colour (Lc 19.1 on white, 13.2 on basement)	APCA Lc 15 = "invisible" for non-text, Lc 30 = absolute minimum for spot text — APCA nutshell; Nayatani Γ — colour-science hke	keep #C6E84B both modes; $levelChipBg = optic, $levelChipText = #0D100E	ink-on Lc 84.1 (13.71:1; 16.6 px @400); on brand block Lc −63.2 (4.64:1), on #1E5236 6.50:1; deutan sim #F2DA57; tritan sim #D1DCC9 (ΔE vs basement 10.2)
Ink / neutral	gray ladder, canvas #F8FBF9, basement #F0F7F3, track #E5EBE7	within envelope; basement is 2.7× Radix sage chroma	Radix tinted greys pair with accent hue — Radix scales; M3 neutral chroma 4 / variant 8 — core_palette.ts	keep all; dark $controlTrack #374039, $controlPill #515B53	basement ΔL* 3.4 vs white; track vs white 1.209:1, vs basement 1.111:1; dark track vs card 1.544:1
Court trio	hard #247BA0, clay #B84A2E, grass #006633	grass = brand (ΔE 2.3); clay ≈ error (6.6; deutan 2.9); clay vs grass protan 1.7	Okabe–Ito: avoid colours between yellow and green, and red/green combinations — jfly.uni-koeln.de/color	alias the series: hard #089AC3, clay #8D3D00, grass #83971F (dark #67D3FC, #DA875A, #BBD068)	trio n/p/d/t min 42.6 / 27.4 / 20.0 / 22.2; grass vs brand 25.9/20.5/23.8/26.1; clay vs error 15.9/7.4/7.5/8.5
Tier quintet	bronze #CD7F32, silver #8E8E93, gold #D4A017, master #9B59B6, grandmaster #E74C3C	bronze–GM deutan 3.4; bronze–gold deutan 7.5; gold 2.38:1 on white	WCAG 1.4.11 3:1 — Understanding	bronze #B2580B, gold #D9A00F, grandmaster #9A0F3A; dark bronze #DA8A11, grandmaster #D8516A	light set min n/p/d/t 25.1/22.7/18.1/11.4 (was 14.8/10.0/3.4/8.7); dark 19.7/14.6/12.8/11.2 (was 19.7/13.2/4.8/10.6). Gold still needs an ink outline (2.34:1)
Series 1–5 (charts)	none	no governed chart palette; Carbon/Spectrum/Tol all publish one	Carbon first 5: min 25.7/9.5/15.8/18.0 — Carbon; Tol — sronpersonalpages.nl/~pault	$series1–5 light #3566B8 #8D3D00 #83971F #883170 #089AC3; dark #6C9EF1 #DA875A #BBD068 #D57CB8 #67D3FC	light min 22.6/15.1/14.9/8.8, ≥ 3:1 on white and basement; dark min 18.9/10.0/10.1/7.1, 5.85–9.74:1 on card
Club identity hue	continuous hue picker (hueFlagPalette, hsl)	not a fixed set; cannot be CVD-gated	—	no change; rule: identity is never colour-only (monogram/photo always present)	—
Hue budget	brand, ink, neutral, semantic ×4, optic, content ≤ 5	10 hue families exist in the token set (27 error, 48 clay, 80 warning, 119–121 optic/grass, 153 brand, 180 success, 225 hard, 260 blue, 302 info, 340 plum); ≤ 7 per viewport is the defensible cap	Healey 1996: "up to seven isoluminant colors" identified rapidly — Healey	per-viewport cap 7 hue families, role-exclusive contexts (§2.7)	—

---

2. Evidence by question

2.1 Brand vs success — how green-brand systems keep a distinct success colour

What the peers actually do (read from published packages, not screenshots; table I). Only Wise is a green-brand system whose tokens could be read; the others show where a "positive" colour goes relative to their own brand.

• Wise (lime brand): interactive-primary #163300 (ink-green), interactive-accent #9FE870 (lime), sentiment-positive #054D28 (L* 28.1). Brand accent and positive are separated by lightness (85.1 vs 28.1), and the lime is only ever a fill with forest-green ink on it — our optic pattern. Dark positive #BAE5A0, negative #CB272F, warning #FFD11A (neptune-tokens 8.20.2, themes/personal*).
• Polaris (Shopify): the chrome brand fill token is #303030 (near-black), not green; success fill #047B5D (OKLCH h 168) and text-success #014B40 (h 179, L* 27.9, Lc 93.0 on white) (@shopify/polaris 13.9.5 styles.css). Shopify's guidance: "Green should be used to indicate that a status is positive or that an action has been completed successfully," "Avoid using green to entice merchants or to draw unnecessary attention," and "avoid relying on color only to provide context" (shopify.dev). I could not verify any statement about Polaris' older green brand.
• Atlassian: brand and information are the same blue #1868DB; success is olive #5B7F24 / text #4C6B1F (OKLCH h 129); warning is a bright fill #FBC828 with dark inverse text (@atlaskit/tokens 20.4.0). Brand shares a hue only with the least critical status, never with success/error.
• 당근 Seed (orange brand #FF6600): fg.positive #079171 (h 170), dark #22B27F, critical #FA342C (color.yaml). Positive is a teal-leaning green.
• Carbon: success is plain green60 #198038 (ΔE vs our brand 7.2), but status must "rely on at least two of the following elements: color, shape, or symbol" (status indicators).
• Robinhood: the only primary evidence found is the "Accessible colors" setting, "to improve the colors for red-green colorblindness" (support). Spotify, Naver and Grab: no primary token or guideline source could be fetched, so I make no claim about them.

Reading: where positive and brand could both be read, positive is moved off the brand hue by hue (Polaris, Seed, Atlassian) or by lightness (Wise), and every system that discusses accessibility asks for a second cue.

Why ΔE ≥ 20 cannot be the rule. Table P scans every in-gamut colour satisfying the lane's own constraints (on-white Lc ≥ 75, white-on Lc ≥ 60, OKLCH C ≥ 0.08). The best achievable min(n,p,d) against #006D37 by hue: 150° → 7.0, 160° → 9.3, 170° → 11.3, 175° → 12.6, 180° → 14.1, 185° → 15.8, 190° → 17.9, 195° → 19.8, 200° → 22.0. ΔE ≥ 20 therefore needs a blue-teal such as #0D7677/#13797E (h 195–200), which (a) reads as blue, not "success", (b) lands only 13–15 ΔE from the current hard-court blue #247BA0 (8.6–10.0 under deutan), and (c) still fails tritan (6.2 / 7.5). Table D shows the realistic bar: Okabe–Ito's worst pair is 12.3 / 11.5 / 11.1 (protan / deutan / tritan), Tol bright 14.0 / 14.7 / 8.7, and Tol, who designed his sets with CIEDE2000 differences and red-blind / green-blind simulation (Tol), publishes no numeric threshold.

Final rule (replaces ΔE ≥ 20): every pair of roles that can sit in one component must reach OI parity — ≥ 12 under protan and deutan, ≥ 20 normal — and carry meaning by glyph + label (WCAG 1.4.1: "Color is not used as the only visual means of conveying information"; a luminance difference counts as a second cue only at ≥ 3:1, and an additional indicator is still required to identify a specific colour whatever the ratio — Understanding 1.4.1). #00796A gives 13.4 / 14.7 / 17.2 against brand (tritan 6.0 is flagged, not gated: every candidate fails it). But a teal moves the confusion partner to the neutral grey: under deutan #00796A simulates to #66676B, ΔE 3.0 from the neutral-badge text #60646C (bg 0.7, table B3) — the old green was 19.1 from it. Pushing the text to the deep end (#014B40, L* 27.9) restores 23.6 / 13.6 / 13.1. Hue scan against {brand, error, neutral, warning text}: best compromise for a single mid-tone is hue 168–172 (#137B5E: worst n/p/d 9.2 / 9.9 / 10.5) — still below OI parity vs brand — so no single mid-tone satisfies all three confusers; the two-tier split does.

One product consequence to decide. #00796A sits at the same OKLCH hue (180) as the shipped badge.accent text #007B6B (ΔE 1.1). "모집 중 / open" (accent) and "완납 / paid" (success) become the same teal. Today success vs accent is already only 11.6 / 12.8 / 15.2 / 2.1 apart, so the merge removes an existing near-duplicate rather than creating one. If the owner wants "open" distinct from "paid", the alternative is to render "open" as ink-on-neutral with an "open door" glyph, not to return to a second green. Also note the pastel tints: brand-subtle #E5FAEA vs success tint is ΔE 4.5 (current #EBF8F1: 4.7) — brand-tint fills and success chips must not share a row.

2.2 Semantic set under CVD — how much lightness separation is needed?

Method (Okabe–Ito / Tol): ensure the pair differs on luminance or sits on opposite sides of the blue–yellow axis. The Okabe–Ito page recommends vermilion for red (it stays recognizable to protanopes) and bluish green for green (so it is not confused with red or brown), and advises against colours that share brightness and differ only in hue. Measured:

• Window. Text-capable colours must satisfy WCAG 4.5:1 (L* ≤ 49.9) and, by the lane's APCA bar, Lc 75 on white (L* ≤ 46.7 at the brand hue; Lc 60 falls at L* 59.3). Brand 39.9, error 42.4, success fill 45.2: pairwise ΔL* ≤ 5.3 (the deep-teal success text at L* 27.9 adds ΔL* 12–15 but only at text level). Table N: only pairs involving the warning fill (L 77.1)* or optic reach the 3:1 luminance ratio; error/info (ΔL* 21.9) is 2.19:1 and brand/info (19.5) is 2.00:1 — so the brief's "ΔL* ≥ 20" heuristic is weaker than the WCAG 3:1 route by about a third.
• Therefore two tiers. Tier 1, text/solid-with-white-ink: brand, success, error, info, warning text — separated by hue and glyph, plus a deep-tone text variant for success (L* 27.9). Tier 2, bright fill with ink: warning (and optic).
• Hue placement. Moving success from green (h 155) to teal (h 180) changes success-vs-error deutan from 9.5 to 24.2 and protan from 13.2 to 16.2 (table B1 vs B3): under deutan/protan red collapses to dark olive (#7B6F29 / #5A522C) while #00796A becomes blue-grey (#66676B / #737169).
• What stays weak (honest list). Brand vs error protan 6.5 (best red found in the scan: 9.2, a vivid #BD2703 that leaves the peer hue consensus; lighter reds are worse, 0.8–5.3). Warning text vs error protan 7.4 (deutan 4.2 → 10.3). Terracotta series vs warning text protan 1.0 (table O). These are resolved by glyph + label + position, not by hex.
• Peers agree on the tier split: danger/success/info fills cluster at L* 42–49 (Polaris 42.0/45.6, Wise 44.7, Atlassian 46.1/48.9/45.9, Carbon 47.0/46.9) while warning fills sit at L* 79–86 with dark ink (table I).

Proposed text-level and fill-level sets are tables B2 and B3.

2.3 Optic lime as level accent

• Numbers. Ink #0D100E on optic: Lc 84.1, 13.71:1 (helper LUT: 16.6 px @400, 14 px @700). Optic on white: 1.40:1, Lc 19.1; on basement 1.28:1, Lc 13.2 (below APCA's Lc 15 "designers should treat anything below this level as invisible", APCA nutshell); on brand #006D37: Lc −63.2, 4.64:1 (22.6 px @400, 15.4 px @700 — numerals only, ≥ 16 px bold); on dark card 11.88:1, Lc −82.5; on dark canvas 14.08:1.
• Why it works at small area. Relative luminance Y = 0.702 (brand 0.112, error 0.128). The CIE photopic V(λ) peaks at 555 nm (luminous efficiency function, secondary; CIE S 010 not fetched), so a yellow-green of given chroma carries more luminance than any other hue — that is a lightness (L* 87.1) advantage, and it is why a 22 pt chip with dark ink reads on a white card (ΔE vs white 29.2 even at 1.4:1 luminance — chroma carries it).
• Correction to the premise. Helmholtz–Kohlrausch does not help it. Nayatani 1997 Γ (object colours, L_a 12.7 cd/m², implementation): optic 0.930 (VCC) / 1.045 (VAC); warning amber 0.890 / 1.052; brand 1.056 / 1.074; error 1.311 / 1.184; indigo 1.590 / 1.259 (table L). H–K adds apparent brightness mostly to saturated reds, blues and purples, and least to yellow-green. An earlier run at L_a = 60 cd/m² gives the same ordering (optic 0.952 / 1.067, error 1.363 / 1.236, indigo 1.654 / 1.323).
• CVD. Simulated: protan #F6DB33, deutan #F2DA57 (both yellow — still ΔE 40.3 / 43.9 from brand, but only 9.7 / 5.8 from warning fill #F5B301), tritan #D1DCC9 (pale grey-green; ΔE vs white 13.2, vs basement 10.2). Rules: never put optic and warning amber in one component; give optic chips on white a 1 px $borderSubtle edge (chip shape otherwise fails in greyscale/tritan).
• Small-text guidance. Not as text: Lc 19.1 on white is below APCA's "Lc 30 absolute minimum for other text and solid non-text". The peer pattern is yellow/lime only as a fill carrying dark ink — Wise (#9FE870 with #163300), Atlassian (warning.bold with inverse #292A2E), Playtomic (lane D).

2.4 Categorical content set

What the references say.

• Okabe–Ito (7 hues + black): alternate warm and cool hues; "Do not convey information in color only. Show difference BOTH in color and shape" (jfly.uni-koeln.de/color); hexes via Wong 2011 (values read from a package that cites it, secondary).
• Tol: distinguishability limits the count; beyond nine colours he recommends a sequential/banded scheme; his default Bright set has 7 hues (Tol).
• Carbon: 14 categorical colours applied in order "curated to maximize contrast between neighboring colors" (Carbon); the 14-set measured here has min deutan ΔE 3.0 (table D) — the order, not the set, carries the safety.
• ColorBrewer: qualitative schemes "rely on differences in hue with only subtle lightness differences" (ColorBrewer) — the equal-L tradition for maps, not for CVD.
• Healey 1996: a hue circle at constant L* = 67.1 (CIELUV); "up to seven isoluminant colors can be displayed simultaneously, while still allowing for the rapid and accurate identification" of one — with neighbour distance shrinking 129.9 → 88.2 → 65.1 ΔE*uv as the count goes 3 → 5 → 7 (UBC TR-96-10; read from the author's PDF).
• Spectrum: categorical palette "optimized to be distinguishable for users with color vision deficiencies"; becomes difficult "starting at 6 colors" and "extremely difficult… at 12" (spectrum.adobe.com; via search excerpt — the page body could not be fetched). Leonardo checks "measurable color difference for each color vision deficiency type" and a 3:1 minimum (leonardocolor.io).
• Size matters: "perceived color difference varies inversely with size" (Szafir 2018, IEEE TVCG 24(1), DOI 10.1109/TVCG.2017.2744359) — so ΔE thresholds for 12–16 px dots must exceed those for swatches. WCAG 1.4.11 asks 3:1 for "each line in a graph" (Understanding).
• Count: ≤ 5–8 per chart is the defensible bound (Spectrum 6 / Healey 7).

Benchmark (table D, helper). Okabe–Ito 21.7 / 12.3 / 11.5 / 11.1; Tol bright 23.7 / 14.0 / 14.7 / 8.7; Tol vibrant 18.3 / 16.1 / 14.5 / 8.6; Carbon first 5: 25.7 / 9.5 / 15.8 / 18.0; Tableau 10 first 5: 22.9 / 6.2 / 0.7 / 9.6; d3 category10 first 5: 26.4 / 1.4 / 4.8 / 11.9. So "≥ 15–20 under CVD" is not met by any published set; the evidence-based bar is OI/Tol parity, i.e. normal ≥ 20, protan/deutan ≥ 12 (OI 11.5), tritan ≥ 8 (Tol 8.7).

Our current sets (table D, E1, F1). Court trio light 36.2 / 1.7 / 15.7 / 13.0 (clay vs grass protan 1.7); dark trio L* spread 4.4 and 0/3 reach 3:1 on white (irrelevant on dark, 6.4–7.3:1 on the card); tiers 14.8 / 10.0 / 3.4 / 8.7; gold 2.38:1.

Equal-L vs staggered-L (the question as asked). Constraint: 3:1 on white and basement ⇒ L* ≲ 58.6. Searches over hue (5° grid, C cap 0.13–0.14), excluding the error and brand/success hue windows:

search	min ΔE n / p / d / t
equal L = 0.60, free hues	11.5 / 9.3 / 8.2 / 5.2
equal L = 0.58, one hue per family (blue, orange, gold, magenta, cyan)	23.2 / 7.5 / 2.2 / 6.9
one hue per family, L 0.56 / 0.60	19.4 / 5.2 / 5.2 / 10.4
one hue per family, L 0.52 / 0.58 / 0.64	20.5 / 13.4 / 12.2 / 7.3
one hue per family, L 0.48 / 0.56 / 0.64	20.9 / 15.1 / 15.0 / 11.9
free hues (not family-constrained), L 0.48 / 0.56 / 0.64	20.4 / 19.6 / 19.6 / 12.8
final: families fixed by use (blue, terracotta, olive-lime, plum, sky), L 0.46 / 0.52 / 0.64, C ≥ 0.11	22.6 / 15.1 / 14.9 / 8.8

So a strictly equal-L set cannot reach OI parity under deutan (best 8.2); a three-step stagger can. This is consistent with the sets that do pass (Okabe–Ito L* spread 43.1, Tol bright 35.8, Carbon first-5 23.4). Proposed: $series1–5 = #3566B8 (OKLCH 0.52/0.14/260), #8D3D00 (0.46/0.125/48), #83971F (0.64/0.14/119), #883170 (0.46/0.14/340), #089AC3 (0.64/0.12/226); court tokens alias series 5 / 2 / 3. Equal chroma ≈ 0.12–0.14, L split 0.46 / 0.52 / 0.64 (so "two ranks": dark = blue, terracotta, plum; light = olive, sky). Tables G1–G3, O.

Known collisions left in the final set (table O): terracotta vs warning text protan 1.0 / deutan 4.0 and vs error 15.9 / 7.4 / 7.5 / 8.5; plum vs info 18 / 9 / 17; olive-lime shares hue 119 with optic (ΔE 21, ΔL* 28). Rule: series never appear inside a status chip or beside an optic chip; a legend always carries direct labels (Okabe–Ito: show the difference "BOTH in color and shape").

Tiers. The gaming-convention hues are kept (bronze orange-brown, silver neutral, gold, master purple, grandmaster red-family), only lightness/hue are moved to clear the CVD failures: bronze #B2580B (4.89:1 on white; was 3.14), gold #D9A00F (unchanged class; 2.34:1 — needs an ink 1.5 px outline or numeral to meet SC 1.4.11), grandmaster crimson #9A0F3A (8.39:1; vs error 13.6 / 16.6 / 12.1; was 12.4 / 10.4 / 12.6). Dark: bronze #DA8A11, grandmaster #D8516A, others unchanged.

2.5 Neutrals with a hue

Data (table J). Radix tinted greys, chroma C* steps 1–8 (backgrounds, borders): gray 0, sage 0.9–1.8 (mean 1.3), olive 0.6–2.4, sand 0.5–3.2, slate 0.5–5.9, mauve 0.6–7.0; chroma peaks at step 9 (sage 3.6, olive 4.4, mauve 7.4) where it is a solid, not a ground. Radix's guidance: use the tinted grey whose hue is closest to the accent, "the difference is subtle" (Radix, via search excerpt); step 1/2 is the dark-mode app background and white the light one (understanding the scale). Tailwind v4.1.11 OKLCH chroma: slate ≤ 0.046, gray ≤ 0.034, zinc ≤ 0.017, neutral 0 (npm tarball); surfaces (steps 50–200) are ≤ 0.013 (slate) / 0.006 (gray). Material: neutral palette chroma 4, neutral variant 8 (HCT, core_palette.ts). Peers' grounds: Seed basement #F3F4F5 C* 0.6, iOS grouped #F2F2F7 2.6, Wise surface #E8EBE6 2.8.

Ours. Canvas C* 1.5 (OKLCH 0.004), basement 3.2 (0.009), track 3.0 (0.008), hairline 4.2 (0.012), border 5.6 (0.015), mid greys 6–7.5 (0.016–0.020), text greys 5.8–6.3. In OKLCH terms every step sits at or below Tailwind gray and far below slate, and at or above Radix sage. Verdict: keep; ceiling C ≤ 4.5 for L ≥ 88** (largest area, avoids the mint-tint read that lane D measured) and hue 145–165 so the tint stays in the brand family. Hue-consistent greys (all 150–161°) match Radix's "pick the tinted grey whose hue is closest to the accent".

Does the tinted canvas shift the brand? Table K, CAT16 (Li et al. 2017, doi:10.1002/col.22131; matrix checked against colour-science): adapting to the basement #F0F7F3 instead of #FFFFFF, the corresponding colour of #006D37 is #106C38 (full von Kries) — ΔE2000 0.76, 0.55 at CAM16 D = 0.85 (L_A = 12.7 cd/m²); canvas 0.27; error 0.65; optic 1.07; success 1.03. All ≲ 1. Dark-card adaptation (3–7) is an upper bound: the eye adapts to the near-white text, not the card.

2.6 Dark theme parity

Reference logic. Material 3 primary is tone 40 in light and 80 in dark, onPrimary 100 / 20, error 40 / 80, surface 98 / 6 (color_spec_2021.ts); a 40-step tone difference guarantees ≥ 3:1 and 50 guarantees ≥ 4.5:1 (Material blog, via search summary of the source comments). Ours: brand L* 39.9 (tone ≈ 40) and dark primary L* 78.3 (≈ 80): the brand pair follows the logic. Seed's dark foregrounds are lighter and flatter (positive #22B27F L* 64.7, critical #FF6E60, warning #CA901C, informative #41A2F9; all |Lc| ≈ 47–49 on #16171B); Wise's are much lighter (positive #BAE5A0 L* 86.5, negative #FFA8AD 77.5, warning #FADC65 88.2; |Lc| 68–86 on #121511). Proposed semantic text tones sit between them at L* 73–78 (|Lc| 60–68 on our card), anchored to M3's tone 80.

What fails today (table H, helper LUT): dark $error #D4736E — Lc 40.5 on card (min px 58.3 @400, 46.5 @500, 31.3 @700; fine as a graphic at 5.12:1, unusable as text); $success #68BF8E and $warning #D4A849 Lc 56.5 / 56.8; dark success vs dark primary ΔE 5.5 (same collision as light); dark live vs deadline banners deutan 3.1.

Each role in dark (all verified, tables H–H5):

role	dark value	Lc on card / canvas / tertiary	note
primary	#70D5A2 (kept)	67.8 / 69.5 / 60.7	ink-on Lc 70.4
success, successText	#75D1C6	67.6 / 69.3 / 60.4	vs dark primary 12.0 / 13.9 / 16.3 / 1.8 (tritan); vs dark neutral text 21.9 / 6.7 / 3.0 — the aqua-vs-grey limit is inherent to any light aqua; the badge bg #14433E and the glyph carry it
warning text	#E1B75C	64.9 / 66.6 / 57.7	was 56.8
warning fill + ink	#E2B13C + #0D100E	ink-on Lc 65.2 (9.65:1)	deadline banner: vs dark live banner 62.2 / 56.5 / 55.9 / 53.6 (was 20.2 / 9.3 / 3.1 / 10.8)
error	#FA9B94	60.4 / 62.1 / 53.2	M3 #FFB4AB is Lc 70.7 but 6.9 / 4.8 from dark primary under protan / deutan; #FA9B94 is 10.0 / 5.1 — protan/deutan primary-vs-error is weak in dark for every tone-80 red (M3 itself pairs them at the same tone)
info	#C7B5F2 (kept)	65.7 / 67.4 / 58.5
optic	#C6E84B (kept)	82.5 / 84.2 / 75.4	brightest object in dark by design; chip area ≤ 1%
brand block (new)	#1E5236	white-on Lc 95.0 (9.07:1); optic-on 6.50:1; block vs card 1.83:1, ΔE 20.6	#70D5A2 cannot be the dark hero block: optic on it is invisible
$controlTrack	#374039 (existing gray400)	vs card 1.544:1, vs canvas 1.829:1
$controlPill	#515B53 (existing gray500)	vs track 1.520:1; label #EFF3F0 Lc 80.9
series 1–5	#6C9EF1 #DA875A #BBD068 #D57CB8 #67D3FC	5.85–9.74:1 on card; Lc 45.8–70.5	min ΔE 18.9 / 10.0 / 10.1 / 7.1
tiers	bronze #DA8A11, grandmaster #D8516A (silver/gold/master kept)	6.0 / 4.2:1 on card	min 19.7 / 14.6 / 12.8 / 11.2 (was 19.7 / 13.2 / 4.8 / 10.6)

Tamagui dark values reuse existing ramp hexes wherever possible (#374039, #515B53, #1E5236, #14433E) so the dark theme gains eleven new hexes (success #75D1C6, warning text #E1B75C, warning fill #E2B13C, error #FA9B94, tier bronze #DA8A11, tier grandmaster #D8516A, and the five series).

2.7 How many colours is "clean"?

Measured / published counts.

• Healey 1996 (CIELUV, constant L* = 67.1): "up to seven isoluminant colors can be displayed simultaneously, while still allowing for the rapid and accurate identification of any one of the seven"; the introduction records anecdotal military-display guidance of "no more than five to seven colours at a time", with no explanation (Healey, Choosing Effective Colours for Data Visualization, PDF read directly). The 2012 Healey & Enns survey (IEEE TVCG 18(7)) is the usual citation for preattentive features, but the file served there was a different Healey document, so I rely on the 1996 paper and do not attribute "≤ 5–7 preattentively distinguishable" to the 2012 survey.
• Spectrum: categorical colour "become[s] more difficult to comprehend starting at 6 colors, and extremely difficult… at 12" (spectrum.adobe.com, search-excerpt only — page body not fetched). EU data-visualisation guide: "Seven is a good maximum" and CVD-safe categorical palettes "are rare and hard to design" (data.europa.eu, secondary).
• Ware: nominal colour coding is usable up to roughly 12 hues (UBC lecture slides crediting Ware, slide text, and the book Information Visualization: Perception for Design) — secondary; the book was not accessible.
• ISO 9241-125:2017 covers visual-information coding (iso.org) but the colour-count provision is paywalled and was not read; the standard states it does not address charts. ISO 22324 (colour-coded alert levels) is summarised as "not exceed seven colors" (Wikipedia summary, secondary). NN/g: no article on a colour-count limit was found, so no NN/g claim is made.
• Carbon (status): "at least two of the following elements: color, shape, or symbol" (Carbon).

Reconciling with the budget. The token set contains 10 hue families at 30° resolution: error 27, terracotta/clay 48, warning amber 80, optic and grass 119–121, brand 153, success 180, hard/sky 225, blue 260, info 302, plum 340. That exceeds 7 only if all appear at once, which the role-exclusive budget (§5 of the benchmark) prevents:

context	hue families on screen	count
홈 hub	brand block 153 · optic chips 121 · success/open chip 180 · warning chip 80 · court dots 225 + 48 + 119	7
기록 hub	brand block 153 · optic progress 121 · tier ring (1 hue) · success delta 180 · 1 series line colour	≤ 6
session detail	brand CTA 153 · court dot (1) · success/warning/error chip (1–2)	≤ 4
chart	5 series hues, no semantic hue inside the plot	5

Rules that follow: (1) ≤ 7 hue families per viewport (Healey 7) and ≤ 5 series per chart (Spectrum 6); (2) role-exclusive: semantic hues only on chips/banners/validation, content hues only on dots/charts/rings, brand only on action + one block; (3) every colour-coded fact has a second cue (WCAG 1.4.1; Carbon). The budget's "content hues ≤ 5" is the series set; court dots alias three of its five hues, tiers are the only content hues outside it, and they never share a row with series.

---

3. Full computed tables (all from bench/colour_lab.py unless marked in the header note)

Letters are referenced from §0–§2. Scripts: bench/lane_g/final_tables.py, extra_tables.py, series_slots.py, series_final.py, court_search.py, adapt.py.

3.0 APCA / WCAG pairs (python3 colour_lab.py pairs FG:BG ..., published-LUT font minima in px by weight)

fg	bg	Lc	WCAG	min px @400	@500	@700
#FFFFFF	#006D37	-86.6	6.48	16.3	15.6	14
#006D37	#FFFFFF	81.5	6.48	17	15.8	14
#006D37	#F0F7F3	75.7	5.95	17.9	16	14
#FFFFFF	#14804A	-79.4	4.98	17.3	15.8	14
#14804A	#FFFFFF	74.0	4.98	18.3	16.4	14.1
#FFFFFF	#00796A	-81.3	5.33	17.1	15.8	14
#00796A	#FFFFFF	75.9	5.33	17.9	16	14
#00796A	#E0F8F3	68.7	4.79	20.1	18.3	14.6
#00796A	#F0F7F3	70.1	4.90	19.5	18	14.5
#FFFFFF	#B8332E	-83.6	5.91	16.7	15.7	14
#B8332E	#FFFFFF	78.3	5.91	17.5	15.9	14
#B8332E	#FDECEC	69.3	5.18	19.8	18.1	14.6
#946200	#FFF4D6	69.4	4.78	19.8	18.1	14.6
#6B4A00	#FFF4D6	81.4	7.36	17	15.8	14
#6B4A00	#FFFFFF	87.7	8.06	16.2	15.6	14
#0D100E	#F5B301	68.8	10.32	20.1	18.2	14.6
#4B0082	#F2E8FF	85.9	10.95	16.4	15.6	14
#FFFFFF	#4B0082	-100.8	12.95	14.9	14.4	12.8
#0D100E	#C6E84B	84.1	13.71	16.6	15.7	14
#C6E84B	#006D37	-63.2	4.64	22.6	19.7	15.4
#C6E84B	#FFFFFF	19.1	1.40	n/a	n/a	n/a
#C6E84B	#F0F7F3	13.2	1.28	n/a	n/a	n/a
#0D100E	#F0F7F3	99.7	17.59	15	14.5	13
#0D100E	#FFFFFF	105.5	19.13	14.4	13.9	11.9
#4F5851	#F0F7F3	79.8	6.78	17.3	15.8	14
#626C64	#F0F7F3	71.4	5.02	19.1	17.4	14.4
#4F5851	#FFFFFF	85.6	7.38	16.4	15.6	14
#626C64	#FFFFFF	77.3	5.46	17.7	15.9	14
#0D100E	#70D5A2	70.4	10.69	19.4	17.8	14.5
#70D5A2	#1A201B	-67.8	9.26	20.5	18.4	14.7
#75D1C6	#1A201B	-67.6	9.23	20.6	18.5	14.7
#E1B75C	#1A201B	-64.9	8.79	21.8	19	15
#FA9B94	#1A201B	-60.4	8.04	23.8	20.8	15.9
#D4736E	#1A201B	-40.5	5.12	58.3	46.5	31.3
#C7B5F2	#1A201B	-65.7	8.94	21.4	18.9	14.9
#C6E84B	#1A201B	-82.5	11.88	16.9	15.7	14
#0D100E	#E2B13C	65.2	9.65	21.6	19	15
#EFF3F0	#1A201B	-97.4	14.81	15.3	14.8	13.3
#BBC3BD	#1A201B	-67.2	9.20	20.7	18.6	14.8

P. Brand-vs-success frontier: best achievable min(ΔE normal, protan, deutan) vs brand #006D37 per OKLCH hue, subject to on-white Lc >= 75, white-on Lc >= 60, OKLCH chroma >= 0.08 (grid L 0.40-0.60 step .005, C 0.08-0.17 step .005)

OKLCH hue	best hex	L / C	ΔE vs brand n / p / d / t	on-white Lc	white-on Lc	reads as
150	#1F542E	0.400 / 0.085	8.8 / 8.5 / 7.0 / 8.1	90.1	94.6	brand green
155	#1C5434	0.400 / 0.080	9.3 / 9.6 / 8.1 / 8.0	90.1	94.5
160	#16563A	0.405 / 0.080	9.5 / 10.0 / 9.3 / 7.0	89.4	93.9	green
165	#0A553D	0.400 / 0.080	10.6 / 11.7 / 11.0 / 7.3	89.8	94.3
170	#035844	0.410 / 0.080	11.3 / 12.5 / 12.4 / 6.1	88.7	93.3	jade green
175	#2D7968	0.525 / 0.080	12.6 / 13.6 / 14.7 / 6.0	75.5	80.9	jade
180	#29796C	0.525 / 0.080	14.1 / 15.3 / 17.4 / 6.1	75.5	80.8	teal (= shipped accent badge hue)
185	#1B7A71	0.525 / 0.085	15.8 / 17.7 / 20.3 / 6.8	75.2	80.6	teal
190	#227975	0.525 / 0.080	17.9 / 20.7 / 22.4 / 6.9	75.3	80.7	blue-teal
195	#0D7677	0.515 / 0.085	19.8 / 23.4 / 24.5 / 6.2	76.6	81.9	blue-teal
200	#13797E	0.525 / 0.085	22.0 / 25.3 / 26.0 / 7.5	75.2	80.5	cyan-teal
205	#147882	0.525 / 0.085	24.4 / 27.8 / 28.0 / 7.7	75.4	80.7	cyan
210	#047887	0.525 / 0.090	26.7 / 30.0 / 29.7 / 8.3	75.2	80.6	cyan

A. Role numbers: current vs final (light)

role	hex	L*	C*	h(Lab)	OKLCH L/C/h	on white Lc	on white WCAG	white-on Lc	ink-on Lc	on tint Lc (WCAG)
brand (kept)	#006D37	39.9	46.3	151	0.469/0.121/153	81.5	6.48	-86.6	22.0	-
success CURRENT	#14804A	47.1	47.1	153	0.530/0.125/155	74.0	4.98	-79.4	29.6	67.9 (4.56)
success FINAL (icon / fill / large figure)	#00796A	45.2	32.4	180	0.516/0.094/180	75.9	5.33	-81.3	27.6	68.7 (4.79)
successText FINAL (small text, badge text)	#014B40	27.9	23.5	178	0.370/0.067/179	93.0	10.09	-97.1	10.4	85.7 (9.07)
warning text CURRENT	#946200	45.7	55.0	76	0.537/0.114/74	75.7	5.24	-81.1	27.8	69.4 (4.78)
warning text FINAL	#6B4A00	34.0	43.8	79	0.434/0.090/79	87.7	8.06	-92.4	15.8	81.4 (7.36)
warning fill FINAL (ink text)	#F5B301	77.1	80.7	81	0.807/0.166/82	34.6	1.85	-38.7	68.8	-
error (kept)	#B8332E	42.4	63.1	33	0.524/0.171/27	78.3	5.91	-83.6	25.2	69.3 (5.18)
info (kept)	#4B0082	20.5	74.3	314	0.339/0.179/302	97.2	12.95	-100.8	0.0	85.9 (10.95)
optic (kept)	#C6E84B	87.1	76.2	115	0.878/0.182/121	19.1	1.40	-21.8	84.1	-

B1. CURRENT set: brand, success, warning, error, info, optic

pair	normal	protan	deutan	tritan	flagged (<10)
brand vs success	6.7	7.0	6.4	7.0	normal, protan, deutan, tritan
brand vs warning	36.1	7.7	14.9	46.9	protan
brand vs error	61.0	6.5	11.3	58.7	protan
brand vs info	50.5	50.6	44.3	36.7	-
brand vs optic	43.6	40.3	43.9	42.1	-
success vs warning	36.5	10.2	13.5	48.2	-
success vs error	61.9	13.2	9.5	60.9	deutan
success vs info	52.7	52.4	45.4	40.5	-
success vs optic	37.4	33.7	37.4	35.7	-
warning vs error	27.6	12.1	4.2	12.0	deutan
warning vs info	59.2	58.3	59.9	28.1	-
warning vs optic	42.8	37.3	31.8	46.9	-
error vs info	39.7	46.4	55.8	33.2	-
error vs optic	67.4	48.5	33.3	53.9	-
info vs optic	102.5	87.5	85.3	69.1	-

B2. FINAL set, TEXT level (light): brand, successText, warningText, error, info, optic, neutral-badge text

pair	normal	protan	deutan	tritan	flagged (<10)
brand vs successText	16.3	17.6	18.1	10.4	-
brand vs warnText	31.9	9.6	9.8	43.2	protan, deutan
brand vs error	61.0	6.5	11.3	58.7	protan
brand vs info	50.5	50.6	44.3	36.7	-
brand vs optic	43.6	40.3	43.9	42.1	-
brand vs neutral	27.2	23.1	20.0	17.8	-
successText vs warnText	31.6	18.5	24.3	40.7	-
successText vs error	51.5	12.9	28.3	53.9	-
successText vs info	37.0	30.5	23.3	32.1	-
successText vs optic	60.5	58.1	62.2	54.8	-
successText vs neutral	23.6	13.6	13.1	18.5	-
warnText vs error	28.9	7.4	10.3	17.4	protan
warnText vs info	55.0	53.4	53.7	20.1	-
warnText vs optic	51.9	50.0	45.4	55.4	-
warnText vs neutral	27.9	28.2	28.6	27.0	-
error vs info	39.7	46.4	55.8	33.2	-
error vs optic	67.4	48.5	33.3	53.9	-
error vs neutral	29.5	21.9	27.7	33.7	-
info vs optic	102.5	87.5	85.3	69.1	-
info vs neutral	27.0	25.0	24.4	25.2	-
optic vs neutral	50.9	50.3	49.6	37.4	-

B3. FINAL FILL / ICON level (light): brand, success, warningFill, error, info, optic, neutral-badge text

pair	normal	protan	deutan	tritan	flagged (<10)
brand vs success	13.4	14.7	17.2	6.0	tritan
brand vs warnFill	49.6	31.9	40.7	57.4	-
brand vs error	61.0	6.5	11.3	58.7	protan
brand vs info	50.5	50.6	44.3	36.7	-
brand vs optic	43.6	40.3	43.9	42.1	-
brand vs neutral	27.2	23.1	20.0	17.8	-
success vs warnFill	48.3	33.8	43.8	54.3	-
success vs error	55.5	16.2	24.2	58.9	-
success vs info	44.9	37.0	27.1	39.7	-
success vs optic	44.1	40.8	46.1	36.9	-
success vs neutral	23.3	10.4	3.0	18.4	deutan
warnFill vs error	46.4	40.4	29.5	30.7	-
warnFill vs info	85.4	79.7	83.7	57.3	-
warnFill vs optic	25.8	9.7	5.8	33.9	protan, deutan
warnFill vs neutral	45.5	43.8	47.5	41.3	-
error vs info	39.7	46.4	55.8	33.2	-
error vs optic	67.4	48.5	33.3	53.9	-
error vs neutral	29.5	21.9	27.7	33.7	-
info vs optic	102.5	87.5	85.3	69.1	-
info vs neutral	27.0	25.0	24.4	25.2	-
optic vs neutral	50.9	50.3	49.6	37.4	-

B4. CURRENT set plus neutral badge text (for comparison)

pair	normal	protan	deutan	tritan	flagged (<10)
brand vs success	6.7	7.0	6.4	7.0	normal, protan, deutan, tritan
brand vs warning	36.1	7.7	14.9	46.9	protan
brand vs error	61.0	6.5	11.3	58.7	protan
brand vs info	50.5	50.6	44.3	36.7	-
brand vs optic	43.6	40.3	43.9	42.1	-
brand vs neutral	27.2	23.1	20.0	17.8	-
success vs warning	36.5	10.2	13.5	48.2	-
success vs error	61.9	13.2	9.5	60.9	deutan
success vs info	52.7	52.4	45.4	40.5	-
success vs optic	37.4	33.7	37.4	35.7	-
success vs neutral	27.5	23.5	19.1	19.4	-
warning vs error	27.6	12.1	4.2	12.0	deutan
warning vs info	59.2	58.3	59.9	28.1	-
warning vs optic	42.8	37.3	31.8	46.9	-
warning vs neutral	29.5	29.1	30.8	29.0	-
error vs info	39.7	46.4	55.8	33.2	-
error vs optic	67.4	48.5	33.3	53.9	-
error vs neutral	29.5	21.9	27.7	33.7	-
info vs optic	102.5	87.5	85.3	69.1	-
info vs neutral	27.0	25.0	24.4	25.2	-
optic vs neutral	50.9	50.3	49.6	37.4	-

C. Brand vs success: ΔE2000 (normal / protan / deutan / tritan) against brand #006D37

candidate	hex	OKLCH h	vs brand	vs error #B8332E	on-white Lc	white-on Lc	note
CURRENT success	#14804A	155	6.7 / 7.0 / 6.4 / 7.0	61.9 / 13.2 / 9.5 / 60.9	74.0	79.4	shipped
Polaris bg-fill-success (npm @shopify/polaris 13.9.5)	#047B5D	168	9.0 / 9.6 / 10.3 / 6.1	58.7 / 13.0 / 17.1 / 59.7	75.6	80.9	h168
Polaris text-success	#014B40	179	16.3 / 17.6 / 18.1 / 10.4	51.5 / 12.9 / 28.3 / 53.9	93.0	97.1
Wise sentiment-positive (npm @transferwise/neptune-tokens 8.20.2)	#054D28	154	10.5 / 10.8 / 9.8 / 10.1	56.7 / 4.3 / 20.3 / 55.5	92.8	96.9	dark forest
Atlassian text.success (npm @atlaskit/tokens 20.4.0)	#4C6B1F	129	11.4 / 5.6 / 7.3 / 11.0	51.2 / 10.0 / 5.5 / 44.6	80.4	85.6	olive
Seed fg.positive (daangn/seed-design color.yaml)	#079171	170	15.3 / 16.1 / 15.7 / 14.2	61.4 / 20.9 / 18.2 / 62.8	66.4	71.9
Carbon green60	#198038	148	7.2 / 7.6 / 7.3 / 6.3	63.2 / 14.0 / 5.1 / 61.0	74.2	79.6
FINAL success (icon/fill)	#00796A	180	13.4 / 14.7 / 17.2 / 6.0	55.5 / 16.2 / 24.2 / 58.9	75.9	81.3	h180
FINAL successText	#014B40	179	16.3 / 17.6 / 18.1 / 10.4	51.5 / 12.9 / 28.3 / 53.9	93.0	97.1	h179, L* 27.9
cyan end of window (ΔE>=20 feasible)	#0D7677	196	19.8 / 23.4 / 24.5 / 6.2	50.7 / 23.4 / 32.1 / 57.8	76.6	81.9	h195-200: reads blue-teal

D. Categorical sets: min pairwise ΔE2000 per vision type, L* spread, 3:1 on white / basement

set	n	min ΔE normal / protan / deutan / tritan	L* range (spread)	>=3:1 on white	>=3:1 on basement #F0F7F3	worst deutan pair	worst protan pair
Okabe-Ito 7 (no black)	7	21.7 / 12.3 / 11.5 / 11.1	46.0-89.1 (43.1)	4/7	3/7	#E69F00/#F0E442 11.5	#0072B2/#CC79A7 12.3
Tol bright 6 (no grey)	6	23.7 / 14.0 / 14.7 / 8.7	41.6-77.4 (35.8)	4/6	3/6	#4477AA/#AA3377 14.7	#4477AA/#AA3377 14.0
Tol vibrant 6 (no grey)	6	18.3 / 16.1 / 14.5 / 8.6	46.0-71.2 (25.2)	4/6	4/6	#EE7733/#CC3311 14.5	#0077BB/#EE3377 16.1
Tol muted 9	9	15.0 / 11.8 / 14.0 / 11.6	22.4-81.8 (59.4)	6/9	5/9	#CC6677/#44AA99 14.0	#332288/#882255 11.8
Tableau 10	10	18.1 / 1.3 / 0.7 / 7.0	49.6-82.0 (32.4)	5/10	4/10	#E15759/#59A14F 0.7	#FF9DA7/#BAB0AC 1.3
Tableau 10 first 5	5	22.9 / 6.2 / 0.7 / 9.6	49.6-70.1 (20.5)	3/5	2/5	#E15759/#59A14F 0.7	#F28E2B/#59A14F 6.2
Carbon categorical first 5	5	25.7 / 9.5 / 15.8 / 18.0	35.2-58.5 (23.4)	5/5	5/5	#005D5D/#9F1853 15.8	#005D5D/#9F1853 9.5
Carbon categorical 14	14	12.2 / 5.8 / 3.0 / 4.5	15.2-58.5 (43.4)	14/14	14/14	#1192E8/#A56EFF 3.0	#1192E8/#A56EFF 5.8
d3 category10 first 5	5	26.4 / 1.4 / 4.8 / 11.9	46.8-66.9 (20.0)	4/5	4/5	#2CA02C/#D62728 4.8	#FF7F0E/#2CA02C 1.4
OURS court trio (light, current)	3	36.2 / 1.7 / 15.7 / 13.0	37.4-48.4 (11.0)	3/3	3/3	#B84A2E/#006633 15.7	#B84A2E/#006633 1.7
OURS court trio (dark, current)	3	28.2 / 9.1 / 14.7 / 6.0	66.0-70.4 (4.4)	0/3	0/3	#E09050/#4DB380 14.7	#E09050/#4DB380 9.1
OURS tiers (light, current)	5	14.8 / 10.0 / 3.4 / 8.7	48.9-68.9 (20.0)	4/5	2/5	#CD7F32/#E74C3C 3.4	#CD7F32/#D4A017 10.0
OURS tiers (dark, current)	5	19.7 / 13.2 / 4.8 / 10.6	64.0-86.9 (22.9)	0/5	0/5	#D4944A/#FF6B6B 4.8	#D4944A/#FF6B6B 13.2
PROPOSED series 1-5 (light)	5	22.6 / 15.1 / 14.9 / 8.8	35.2-59.1 (23.9)	5/5	5/5	#3566B8/#883170 14.9	#3566B8/#883170 15.1
PROPOSED series 1-5 (dark)	5	18.9 / 10.0 / 10.1 / 7.1	63.4-80.0 (16.6)	0/5	0/5	#DA875A/#BBD068 10.1	#6C9EF1/#D57CB8 10.0

E1. Court trio CURRENT: hard #247BA0, clay #B84A2E, grass #006633

pair	normal	protan	deutan	tritan	flagged (<10)
hard vs clay	47.9	38.7	43.6	55.4	-
hard vs grass	36.2	37.3	35.0	13.0	-
clay vs grass	54.9	1.7	15.7	56.3	protan

E2. Court trio PROPOSED (= series 5, 2, 3): hard #089AC3, clay #8D3D00, grass #83971F

pair	normal	protan	deutan	tritan	flagged (<10)
hard vs clay	54.4	51.1	49.9	60.2	-
hard vs grass	48.1	46.9	48.4	22.2	-
clay vs grass	42.6	27.4	20.0	38.0	-

Court vs semantic: current grass vs brand ΔE 2.3 / 2.6 / 2.2 / 2.4 ; current clay vs error 6.6 / 5.4 / 2.9 / 4.1 ; proposed grass vs brand 25.9 / 20.5 / 23.8 / 26.1 ; proposed clay vs error 15.9 / 7.4 / 7.5 / 8.5 ; proposed clay vs warningText 14.4 / 0.8 / 3.5 / 10.3

F1. Tiers CURRENT (light)

pair	normal	protan	deutan	tritan	flagged (<10)
bronze vs silver	27.5	27.3	28.0	24.7	-
bronze vs gold	14.8	10.0	7.5	8.7	deutan, tritan
bronze vs master	48.6	51.9	51.1	18.7	-
bronze vs grandmaster	21.1	12.2	3.4	10.5	deutan
silver vs gold	31.4	31.2	31.9	24.3	-
silver vs master	25.1	22.7	20.1	21.4	-
silver vs grandmaster	29.2	25.1	26.7	30.0	-
gold vs master	61.8	57.9	55.9	22.2	-
gold vs grandmaster	36.5	22.1	10.9	18.2	-
master vs grandmaster	35.5	44.6	49.2	23.5	-

F2. Tiers PROPOSED (light): bronze #B2580B, gold #D9A00F, grandmaster #9A0F3A (silver, master kept)

pair	normal	protan	deutan	tritan	flagged (<10)
bronze vs silver	30.3	31.8	29.3	28.3	-
bronze vs gold	25.8	23.5	18.4	20.5	-
bronze vs master	45.2	51.1	50.1	18.0	-
bronze vs grandmaster	30.1	26.8	18.1	11.4	-
silver vs gold	31.7	31.3	32.3	24.9	-
silver vs master	25.1	22.7	20.1	21.4	-
silver vs grandmaster	36.1	31.1	28.0	36.2	-
gold vs master	61.7	58.0	56.2	23.0	-
gold vs grandmaster	56.4	47.4	36.9	34.5	-
master vs grandmaster	28.0	29.6	39.4	24.6	-

tier	current	on white WCAG	proposed	on white WCAG
bronze	#CD7F32	3.14	#B2580B	4.89
silver	#8E8E93	3.26	#8E8E93	3.26
gold	#D4A017	2.38	#D9A00F	2.34
master	#9B59B6	4.67	#9B59B6	4.67
grandmaster	#E74C3C	3.82	#9A0F3A	8.39

GM vs error: current 12.4 / 10.4 / 12.6 / 11.9 ; proposed 13.6 / 16.6 / 12.1 / 7.0 ; gold(proposed) vs warningFill 6.0 / 5.7 / 6.1 / 4.5 ; bronze(proposed) vs clay(proposed) 10.8 / 10.0 / 11.3 / 10.5

G1. PROPOSED series 1-5 (light)

pair	normal	protan	deutan	tritan	flagged (<10)
s1 blue vs s2 terracotta	45.8	50.5	52.7	58.5	-
s1 blue vs s3 olive-lime	61.7	56.4	57.0	24.0	-
s1 blue vs s4 plum	30.4	15.1	14.9	54.4	-
s1 blue vs s5 sky	22.6	16.6	15.5	16.3	-
s2 terracotta vs s3 olive-lime	42.6	27.4	20.0	38.0	-
s2 terracotta vs s4 plum	36.7	40.8	35.8	8.8	tritan
s2 terracotta vs s5 sky	54.4	51.1	49.9	60.2	-
s3 olive-lime vs s4 plum	68.0	53.5	42.8	38.7	-
s3 olive-lime vs s5 sky	48.1	46.9	48.4	22.2	-
s4 plum vs s5 sky	54.3	30.2	21.0	66.5	-

G2. PROPOSED series 1-5 (dark)

pair	normal	protan	deutan	tritan	flagged (<10)
s1 vs s2	42.9	46.1	50.1	65.9	-
s1 vs s3	58.2	53.2	53.8	25.4	-
s1 vs s4	32.1	10.0	14.5	61.7	-
s1 vs s5	18.9	13.5	12.1	12.6	-
s2 vs s3	37.3	17.0	10.1	33.3	-
s2 vs s4	32.8	37.3	33.5	7.1	tritan
s2 vs s5	49.2	42.1	45.2	59.6	-
s3 vs s4	62.8	46.1	37.0	34.0	-
s3 vs s5	45.5	44.5	46.6	24.0	-
s4 vs s5	50.4	17.4	13.3	69.9	-

G3. Series values

token	light	OKLCH (light)	WCAG on white / basement	dark	WCAG on dark card / canvas	Lc on dark card
$series1	#3566B8	0.52/0.140/260	5.61 / 5.16	#6C9EF1	6.16 / 7.30	47.9
$series2	#8D3D00	0.46/0.125/48	7.50 / 6.89	#DA875A	6.00 / 7.11	46.8
$series3	#83971F	0.64/0.140/119	3.28 / 3.01	#BBD068	9.74 / 11.54	70.5
$series4	#883170	0.46/0.140/340	7.72 / 7.10	#D57CB8	5.85 / 6.93	45.8
$series5	#089AC3	0.64/0.120/226	3.27 / 3.01	#67D3FC	9.71 / 11.51	70.5

H. Dark parity: current vs proposed on card #1A201B / canvas #090C0A / tertiary #374039

role	current	Lc card	WCAG card	proposed	Lc card	Lc canvas	Lc tertiary	WCAG card
primary	#70D5A2	67.8	9.26	#70D5A2	67.8	69.5	60.7	9.26
success	#68BF8E	56.5	7.45	#75D1C6	67.6	69.3	60.4	9.23
warning text	#D4A849	56.8	7.50	#E1B75C	64.9	66.6	57.7	8.79
error	#D4736E	40.5	5.12	#FA9B94	60.4	62.1	53.2	8.04
info	#C7B5F2	65.7	8.94	#C7B5F2	65.7	67.4	58.5	8.94
optic	#C6E84B	82.5	11.88	#C6E84B	82.5	84.2	75.4	11.88

H2. DARK current: primary, success, warning, error, info, optic

pair	normal	protan	deutan	tritan	flagged (<10)
primary vs success	5.5	5.5	5.3	5.5	normal, protan, deutan, tritan
primary vs warning	33.3	15.0	16.4	48.4	-
primary vs error	57.6	21.2	12.8	61.4	-
primary vs info	38.9	35.8	29.3	27.6	-
primary vs optic	22.1	17.9	20.2	20.0	-
success vs warning	31.7	12.6	15.9	46.5	-
success vs error	54.8	16.1	8.2	58.0	deutan
success vs info	38.7	35.8	30.4	27.3	-
success vs optic	23.4	20.0	22.2	21.0	-
warning vs error	32.5	20.7	13.9	13.1	-
warning vs info	47.6	47.6	45.6	21.9	-
warning vs optic	23.8	14.4	10.3	33.8	-
error vs info	31.3	35.3	37.4	29.5	-
error vs optic	56.1	32.9	23.0	42.0	-
info vs optic	61.8	51.7	48.4	17.3	-

H3. DARK proposed: primary, success, warningText, error, info, optic

pair	normal	protan	deutan	tritan	flagged (<10)
primary vs success	12.0	13.9	16.3	1.8	tritan
primary vs warnText	31.9	12.2	15.6	47.3	-
primary vs error	54.3	10.0	5.1	57.1	deutan
primary vs info	38.9	35.8	29.3	27.6	-
primary vs optic	22.1	17.9	20.2	20.0	-
success vs warnText	34.4	23.7	30.7	47.1	-
success vs error	48.2	12.0	21.4	56.7	-
success vs info	32.4	19.6	12.3	27.7	-
success vs optic	30.5	28.1	34.1	20.6	-
warnText vs error	30.1	16.7	11.0	8.4	tritan
warnText vs info	47.1	46.4	45.0	21.0	-
warnText vs optic	21.6	11.4	7.3	32.3	deutan
error vs info	27.6	30.4	34.8	24.4	-
error vs optic	51.7	25.3	16.3	37.0	-
info vs optic	61.8	51.7	48.4	17.3	-

Dark warning fill #E2B13C: ink #0D100E on it Lc 65.2 (WCAG 9.65); vs card WCAG 8.36 ; vs error(dark) 32.4 / 19.7 / 13.8 / 6.8

H4. Dark tiers PROPOSED vs current

dark tiers CURRENT

pair	normal	protan	deutan	tritan	flagged (<10)
bronze vs silver	26.0	26.1	26.2	25.9	-
bronze vs gold	22.8	18.9	16.7	15.6	-
bronze vs master	41.8	43.1	41.6	15.4	-
bronze vs grandmaster	27.1	13.2	4.8	10.6	deutan
silver vs gold	34.4	33.7	34.6	23.2	-
silver vs master	19.7	14.6	12.8	16.6	-
silver vs grandmaster	27.8	18.5	22.5	21.5	-
gold vs master	58.6	51.0	49.4	15.7	-
gold vs grandmaster	49.7	29.7	19.5	25.0	-
master vs grandmaster	28.0	32.9	37.5	22.6	-

dark tiers PROPOSED (bronze #DA8A11, grandmaster #D8516A; silver, gold, master kept)

pair	normal	protan	deutan	tritan	flagged (<10)
bronze vs silver	29.7	30.1	29.5	27.2	-
bronze vs gold	23.3	19.3	15.9	18.7	-
bronze vs master	46.4	47.2	45.2	18.6	-
bronze vs grandmaster	37.7	27.5	17.7	11.2	-
silver vs gold	34.4	33.7	34.6	23.2	-
silver vs master	19.7	14.6	12.8	16.6	-
silver vs grandmaster	29.5	21.0	19.3	32.6	-
gold vs master	58.6	51.0	49.4	15.7	-
gold vs grandmaster	58.1	41.3	30.1	29.4	-
master vs grandmaster	26.3	28.3	32.7	25.4	-

H5. Control track / pill

• track #E5EBE7 on #FFFFFF: WCAG 1.209, ΔL* 7.5
• track #E5EBE7 on #F0F7F3: WCAG 1.111, ΔL* 4.1
• track #E5EBE7 on #F8FBF9: WCAG 1.161, ΔL* 5.8
• track #374039 on #1A201B: WCAG 1.544, ΔL* 14.6
• track #374039 on #090C0A: WCAG 1.829, ΔL* 23.0
• track #2A332C on #1A201B: WCAG 1.270, ΔL* 8.7
• dark pill #515B53 on track #374039: WCAG 1.520 ; text #EFF3F0 on pill Lc 80.9
• light pill #FFFFFF on track #E5EBE7: WCAG 1.209

I. Peer semantic/brand values read from published packages (hex -> our metrics)

system / token	hex	L*	Lab h	OKLCH h	on-white Lc	white-on Lc	source
Polaris --p-color-bg-fill-brand (light)	#303030	19.9	158	90	99.6	102.8	@shopify/polaris 13.9.5 build/esm/styles.css
Polaris bg-fill-success	#047B5D	45.6	167	168	75.6	80.9	same
Polaris text-success	#014B40	27.9	178	179	93.0	97.1	same
Polaris bg-fill-critical	#C70A24	42.0	32	24	77.5	82.8	same
Polaris bg-fill-warning	#FFB800	79.4	80	81	31.1	35.0	same
Polaris text-warning	#5E4200	30.1	81	81	91.3	95.6	same
Wise interactive-primary (personal)	#163300	18.1	130	135	100.3	103.4	@transferwise/neptune-tokens 8.20.2 themes/personal/tokens.css
Wise interactive-accent	#9FE870	85.1	130	135	22.2	25.3	same
Wise sentiment-positive	#054D28	28.1	152	154	92.8	96.9	same
Wise sentiment-negative	#CB272F	44.7	31	25	75.3	80.7	same
Wise sentiment-warning	#FFD11A	85.5	89	92	21.6	24.6	same
Wise dark sentiment-positive	#BAE5A0	86.5	132	134	20.1	22.9	themes/personal--dark
Atlassian background.brand.bold = information.bold	#1868DB	45.9	288	259	75.1	80.4	@atlaskit/tokens 20.4.0 figma/atlassian-light.json
Atlassian success.bold	#5B7F24	48.9	122	129	72.2	77.7	same
Atlassian text.success	#4C6B1F	41.5	123	129	80.4	85.6	same
Atlassian danger.bold	#C9372C	46.1	36	29	74.2	79.7	same
Atlassian warning.bold	#FBC828	82.9	87	89	25.7	29.1	same
Atlassian text.warning	#9E4C00	41.9	60	52	79.5	84.7	same
Carbon green60 / red60 / yellow30 / orange40	#198038	46.9	145	148	74.2	79.6	carbondesignsystem.com/data-visualization/color-palettes (alert palette)
red60	#DA1E28	47.0	33	26	72.5	77.9	same
yellow30	#F1C21B	80.4	88	90	29.6	33.4	same
Seed (당근) fg.positive light	#079171	53.5	169	170	66.4	71.9	github.com/daangn/seed-design packages/rootage/color.yaml
Seed fg.critical light	#FA342C	55.0	36	28	63.3	68.7	same
Seed bg.brand-solid (carrot-600)	#FF6600	62.3	52	43	54.8	60.1	same
Seed fg.positive dark	#22B27F	64.7	162	163	52.1	57.3	same

J. Tinted neutrals: CIELAB C* by step (Radix 3.0.0 light; first sRGB block of each CSS file)

scale	C* steps 1-12	mean C* steps 1-8	max C* steps 1-8	step 9	step 11
radix gray	0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0	0.0	0.0	0.0	0.0
radix sage	0.9 0.9 1.2 1.2 1.5 1.2 1.8 1.8 3.6 3.6 2.8 4.1	1.3	1.8	3.6	2.8
radix olive	0.6 1.2 1.3 1.3 1.9 1.9 1.9 2.4 4.4 4.5 4.2 4.2	1.6	2.4	4.4	4.2
radix sand	0.5 0.5 0.6 1.1 1.6 1.6 2.1 3.2 4.0 3.4 3.0 2.9	1.4	3.2	4.0	3.0
radix slate	0.5 1.0 1.5 2.1 2.6 3.7 4.2 5.9 6.3 5.8 5.0 3.4	2.7	5.9	6.3	5.0
radix mauve	0.6 1.1 2.3 2.8 3.9 4.3 5.4 7.0 7.4 7.0 6.1 5.2	3.4	7.0	7.4	6.1

Tailwind CSS 4.1.11 theme.css neutral families, OKLCH chroma (steps 50/100/200/300/400/500/600/700/800/900/950):

family	chroma by step
slate	0.003 0.007 0.013 0.022 0.04 0.046 0.043 0.044 0.041 0.042 0.042
gray	0.002 0.003 0.006 0.01 0.022 0.027 0.03 0.034 0.033 0.034 0.028
zinc	0 0.001 0.004 0.006 0.015 0.016 0.017 0.013 0.006 0.006 0.005
neutral	0 0 0 0 0 0 0 0 0 0 0
stone	0.001 0.001 0.003 0.005 0.01 0.013 0.011 0.01 0.007 0.006 0.004

ours	hex	L*	C* (Lab)	OKLCH C	hue (OKLCH)
canvas	#F8FBF9	98.3	1.5	0.0041	157
basement (proposed)	#F0F7F3	96.6	3.2	0.0091	161
controlTrack	#E5EBE7	92.5	3.0	0.0083	157
hairline gray200	#DFE7E1	90.9	4.2	0.0117	154
border gray300	#CAD4CC	84.0	5.6	0.0154	152
gray400	#A8B5AB	72.4	7.3	0.0202	153
gray500 (borderStrong)	#7E8A80	56.3	7.5	0.0204	150
textTertiary gray600	#626C64	44.6	6.3	0.0174	151
textSecondary gray700	#4F5851	36.4	5.8	0.0160	152
ink gray950	#0D100E	4.4	1.5	0.0063	156
dark canvas	#090C0A	3.1	1.2	0.0066	157
dark card	#1A201B	11.5	4.8	0.0129	150
dark gray300	#262D28	17.7	4.8	0.0133	154
dark gray500	#515B53	37.6	6.5	0.0179	151

Reference grounds: Seed basement #F3F4F5 C* 0.6 ; iOS grouped #F2F2F7 C* 2.6 ; Wise light surface #E8EBE6 C* 2.8 ; Radix sage-3 #EEF1F0 C* 1.2 ; Seed dark layer #16171B C* 3.0

K. Chromatic-adaptation shift of a stimulus when the adapting white is the tinted ground instead of #FFFFFF (CAT16, ΔE2000 between the stimulus and its corresponding colour)

stimulus	canvas #F8FBF9 (D=1)	basement #F0F7F3 (D=1)	basement (D=0.85, CAM16 at L_A=12.7)	track #E5EBE7 (D=1)	dark card #1A201B (D=1, upper bound)
brand #006D37	0.27	0.76	0.55	0.69	3.71
success #00796A	0.47	1.03	0.98	1.01	7.04
error #B8332E	0.24	0.65	0.46	0.46	3.32
optic #C6E84B	0.53	1.07	0.88	1.07	5.98
info #4B0082	0.16	0.21	0.21	0.21	1.57

L. Relative luminance and Helmholtz-Kohlrausch luminance factor Gamma (Nayatani 1997 object-colour estimate, colour-science 0.4.7, L_a = 12.7 cd/m2, D65 white)

colour	hex	Y (white=1)	L*	Gamma VCC	Gamma VAC
optic	#C6E84B	0.702	87.1	0.930	1.045
warning fill	#F5B301	0.517	77.1	0.890	1.052
brand	#006D37	0.112	39.9	1.056	1.074
success	#00796A	0.147	45.2	1.057	1.054
hard (current)	#247BA0	0.171	48.4	1.121	1.074
clay (current)	#B84A2E	0.153	46.0	1.196	1.140
error	#B8332E	0.128	42.4	1.311	1.184
info	#4B0082	0.031	20.5	1.590	1.259
grey 8E8E93	#8E8E93	0.272	59.2	1.012	1.006

M. Optic lime #C6E84B: appearance under simulated deficiency, and contrast to grounds

vision	simulated optic	vs white dE	vs basement dE	vs brand dE	vs warningFill dE
normal	#C6E84B	29.2	27.7	43.6	25.8
protan	#F6DB33	29.5	28.1	40.3	9.7
deutan	#F2DA57	27.7	26.6	43.9	5.8
tritan	#D1DCC9	13.2	10.2	42.1	33.9

N. Final-fill luminance separation (SC 1.4.1 "3:1 luminance difference" alternative)

pair	dL*	WCAG luminance ratio	>= 3:1
brand / success	5.3	1.22	no
brand / error	2.5	1.10	no
brand / info	19.5	2.00	no
brand / warningText	5.9	1.24	no
brand / warningFill	37.1	3.49	yes
brand / optic	47.2	4.64	yes
success / error	2.8	1.11	no
success / info	24.8	2.43	no
success / warningText	11.2	1.51	no
success / warningFill	31.8	2.87	no
success / optic	41.9	3.82	yes
error / info	21.9	2.19	no
error / warningText	8.4	1.36	no
error / warningFill	34.7	3.19	yes
error / optic	44.7	4.24	yes
info / warningText	13.6	1.61	no
info / warningFill	56.6	6.99	yes
info / optic	66.6	9.28	yes
warningText / warningFill	43.0	4.35	yes
warningText / optic	53.1	5.78	yes
warningFill / optic	10.0	1.33	no

O. Series vs semantic roles: ΔE2000 normal/protan/deutan/tritan (! = any of n/p/d < 10)

series	brand	success	warnText	warnFill	error	info	optic
s1 blue	44/45/41/12	32/29/22/9	49/50/52/50	63/62/68/59	42/43/51/64	28/20/17/39	72/66/69/38
s2 terracotta	47/10/11/53	46/23/26/54	14/1/4/10 !	43/43/37/37	16/7/7/8 !	47/53/56/27	60/51/41/56
s3 olive-lime	26/20/24/26	30/26/31/23	31/27/23/35	26/12/17/32	54/27/13/38	81/68/65/41	21/21/21/21
s4 plum	66/38/26/53	45/26/10/54	45/41/35/10	68/64/56/38	26/35/35/14	18/9/17/21 !	83/70/58/57
s5 sky	40/42/40/22	25/27/22/16	50/51/50/52	53/49/56/50	59/44/46/63	49/37/30/51	55/53/56/27

---

4. Final token proposal (ready for classic-grass.theme.ts, shared-semantics.ts, status-badge-palette.ts, packages/ui/src/tokens.ts, tier.config.ts)

All hexes verified in §3. "New" = token does not exist today. Dark values reuse existing ramp hexes where marked (=).

token	light	dark	role	allowed uses	notes / replaces
$primary	#006D37	#70D5A2	brand solid	the screen's one primary CTA, toggle on, checkbox, selected date disc; inline link	unchanged
$brandBlock (new)	#006D37	#1E5236 (= ramp 300)	hero block fill	one per hub screen (홈 day block, 기록 ELO block, 프로필 identity)	$onBrandBlock #FFFFFF / #EFF3F0; optic on block 4.64:1 light, 6.50:1 dark
$basement (new)	#F0F7F3	#090C0A (= canvas)	hub ground	hub/dashboard screens behind white cards; lists stay white	ΔL* 3.4 vs white; C* 3.2
$controlTrack (new)	#E5EBE7	#374039 (= gray400)	segmented/tab/progress track	neutral controls	1.209 vs white, 1.111 vs basement; dark 1.544 vs card
$controlPill (new)	#FFFFFF	#515B53 (= gray500)	selected segment	with 1 px $borderSubtle in light	pill vs track 1.209 / 1.520
$success	#00796A	#75D1C6	success icon / fill / large figure	paid, RSVP yes, +delta glyph, validation OK, always with a check glyph; solid fill with white ink	was #14804A / #68BF8E; SHARED_STATUS_*.success, tokens.ts success, THEME_HEX_FALLBACK.success, toast inverse.successIcon
$successText (new)	#014B40	#75D1C6	success small text, badge text	badge text, inline "+24" at < 18 px	badge pair: light {bg #E0F8F3, text #014B40}, dark {bg #14433E, text #75D1C6}; badge.accent becomes an alias of badge.success (was #E0F8F3/#007B6B and #14433E/#44D2BA)
$warning	#946200	#D4A849	warning icon / graphic only	lighting icon, facility accents	unchanged (5.24:1 on white, 7.50:1 on dark card)
$warningText (new)	#6B4A00	#E1B75C	warning text	badge text on #FFF4D6, inline warning copy	light badge text was #946200 (Lc 69.4, 4.78:1) → Lc 81.4, 7.36:1
$warningFill (new)	#F5B301	#E2B13C	warning solid	deadline banner, "마감 임박" solid chip	$onWarningFill #0D100E / #0D100E; replaces deadlineBannerBg (#946200 / #4A3106) and deadlineBannerText
$error	#B8332E	#FA9B94	error	text, icon, fill (white ink in light)	dark was #D4736E (Lc 40.5); SHARED_STATUS_DARK.error, toast inverse.errorIcon
$info	#4B0082	#C7B5F2	info	badge, notice	unchanged
$levelChipBg (new)	#C6E84B	#C6E84B	level accent (= optic)	NTRP/ELO chip under avatars and in rows, today dot, progress fill inside $brandBlock	never text, never on white without 1 px edge, never beside $warningFill; $levelChipText #0D100E both (Lc 84.1, 13.71:1)
$series1 (new)	#3566B8	#6C9EF1	chart/content hue 1	chart series 1	OKLCH 0.52/0.14/260
$series2 (new)	#8D3D00	#DA875A	hue 2 (terracotta)	chart series 2; = $courtClay	OKLCH 0.46/0.125/48
$series3 (new)	#83971F	#BBD068	hue 3 (olive-lime)	chart series 3; = $courtGreen	OKLCH 0.64/0.14/119
$series4 (new)	#883170	#D57CB8	hue 4 (plum)	chart series 4	OKLCH 0.46/0.14/340
$series5 (new)	#089AC3	#67D3FC	hue 5 (sky)	chart series 5; = $courtHard	OKLCH 0.64/0.12/226
$courtHard / $courtClay / $courtGreen	#089AC3 / #8D3D00 / #83971F	#67D3FC / #DA875A / #BBD068	court surface identity	12–40 px dots and CourtLinesTile fills	were #247BA0 / #B84A2E / #006633 (dark #67B7D1 / #E09050 / #4DB380); other three themes keep their own court values unless the owner extends
$tierBronze	#B2580B	#DA8A11	tier hue	tier ring/dot beside a text label	tier.config.ts color/colorDark; bgTint #B2580B15
$tierSilver	#8E8E93	#A8A9AD	tier hue	same	unchanged
$tierGold	#D9A00F	#FFD700	tier hue	same; 2.34:1 on white ⇒ needs an ink outline or numeral	bgTint #D9A00F15
$tierMaster	#9B59B6	#C39BD3	tier hue	same	unchanged
$tierGrandmaster	#9A0F3A	#D8516A	tier hue	same	was #E74C3C / #FF6B6B; bgTint #9A0F3A15

Wiring notes (all repo paths read, none edited): SHARED_STATUS_LIGHT/DARK, SHARED_SEMANTICS_*.deadlineBanner* in packages/app/src/config/themes/shared-semantics.ts (theme-invariant, so the change reaches all four themes at once); STATUS_BADGE_LIGHT/DARK.success/accent/warning in status-badge-palette.ts; colors.court/colorsDark.court, inverse.successIcon/errorIcon and the new fields in classic-grass.theme.ts; tokens.ts. Side finding: tokens.ts primitives textPrimary #131A14, textSecondary #455D49, surfaceSecondary #F6F8F5, border #D5E3D6, borderDark #617A65 differ from the live theme (#0D100E, #4F5851, #F8FBF9, #CAD4CC, #7E8A80); tokens-theme-agreement.test.ts pins only warning and textTertiary, so these have drifted unguarded.

Blast radius (grep, excl. tests): $success 40 sites, $warning 40, $error 112, $info 2, $optic 0, court tokens 14, deadlineBannerBg 4, $badgeSuccess 11, $badgeAccent 4. The value changes to $success (40 sites) and dark $error (112 sites) reach the most call sites; keeping $success as the icon/fill token and adding $successText for badge text means no call site has to change, only colour values (and the badge palette).

---

5. What to add to scripts/check-color-contrast.mjs

What it asserts today (read from scripts/check-color-contrast.mjs and scripts/lib/color-contrast-core.mjs): WCAG 4.5:1 for derived {bg,text} pairs, text tiers on every surface, brand-as-ink, the primary button; 3:1 for border.dark and icon.muted; ≥ 1.5:1 for dark badge fills; four APCA watch entries (dark badge.muted hard at 45; dark badge.success/accent, dark text.tertiary on tertiary/elevated, dark button warn at 60); empty baseline; exemptions for the Naver logotype and disabled text.

Gaps (verified by reading derivePairs):

1. status.success/warning/error/info are not derived as text/icon on a surface — they enter only through badge pairs and the toast icons. Today's $error dark #D4736E on the card (Lc 40.5, 5.12:1) and light $success (Lc 74.0) therefore pass silently.
2. court.* and medal.* are info only; tiers (tier.config.ts), optic, basement, controlTrack, series do not appear at all.
3. No CVD assertion of any kind.
4. APCA is gated on four hand-picked pairs only.

Additions (each row = a derived pair or an assertion; thresholds are the numbers the final tokens already meet):

id	assertion	threshold (kind)	final value	canary (current value that must fail)
A1	status.{error,info}, success, successText, warningText as text on surface.primary/secondary and basement	WCAG 4.5 (fail) · APCA ≥ 60 (fail) · APCA ≥ 75 on surface.primary (warn)	success 5.33 / Lc 75.9; successText 10.09 / 93.0; warningText 8.06 / 87.7; error 5.91 / 78.3	dark $error #D4736E Lc 40.5 → fail; light $success #14804A Lc 74.0 → warn
A2	dark status text on surface.primary, secondary, tertiary, elevated	APCA ≥ 60 on primary/secondary (fail), ≥ 55 on tertiary/elevated (warn)	success 67.6/69.3/60.4; warning 64.9/66.6/57.7; error 60.4/62.1/53.2	#D4736E
A3	solid fills with their ink: success/error/info/brandBlock + text.inverse; warningFill + onWarningFill; levelChipBg + levelChipText	APCA ≥ 75 (fail) for white-ink fills; ≥ 60 for warningFill (fail), ≥ 65 (warn); levelChip ≥ 75	81.3 / 83.6 / 100.8 / 86.6; 68.8; 84.1	—
A4	optic on brandBlock light/dark	WCAG 4.5 (fail)	4.64 / 6.50	optic on dark primary #70D5A2 (invisible)
A5	series1–5 vs surface.primary and basement (light), vs card and canvas (dark)	WCAG 3.0 (fail, SC 1.4.11)	min 3.27 / 3.01 light; 5.85 dark	—
A6	controlTrack vs surface.primary; vs basement; dark vs card; pill vs track	WCAG ≥ 1.2 / ≥ 1.1 / ≥ 1.5 / ≥ 1.2 (fail)	1.209 / 1.111 / 1.544 / 1.209	gray100 #F0F5F1 (1.10 / 1.01)
A7	basement vs card: ΔL* ∈ [3.0, 5.5]; canvas of list screens = #FFFFFF; neutral chroma: Lab C* ≤ 4.5 for L* ≥ 88 and hue 145–165	fail	3.4; 3.2 / 4.2	#EEF2EF (grey-everywhere reading withdrawn in the benchmark)
A8	CVD role separation (new cvdSimulate + deltaE2000 in color-contrast-core.mjs; Machado 2009 severity 1.0, matrices applied in linear RGB)	config table below	below	below
A9	hue-family count of {brand, success, warning fill, error, info, optic, series1–5} at 30° buckets	report-only; warn above 10	10	—

A8 config (scripts/color-separation.config.json), thresholds chosen at Okabe–Ito parity (worst OI pair 21.7 / 12.3 / 11.5 / 11.1):

pair (light unless noted)	normal	protan	deutan	tritan	final measured (n/p/d/t)
brand ↔ success	≥ 12	≥ 12	≥ 12	warn ≥ 5	13.4 / 14.7 / 17.2 / 6.0
brand ↔ successText	≥ 14	≥ 14	≥ 14	warn ≥ 8	16.3 / 17.6 / 18.1 / 10.4
successText ↔ neutral badge text	≥ 20	≥ 12	≥ 12	—	23.6 / 13.6 / 13.1 / 18.5
success ↔ error	≥ 40	≥ 15	≥ 15	—	55.5 / 16.2 / 24.2 / 58.9
warningFill ↔ error (+ WCAG luminance ratio ≥ 3.0)	≥ 25	≥ 25	≥ 25	—	46.4 / 40.4 / 29.5 / 30.7 (3.19:1)
warningText ↔ error	≥ 20	warn ≥ 7	warn ≥ 8	—	28.9 / 7.4 / 10.3 / 17.4
brand ↔ optic	≥ 30	≥ 30	≥ 30	≥ 30	43.6 / 40.3 / 43.9 / 42.1
series 1–5 pairwise (light)	≥ 20	≥ 12	≥ 12	warn ≥ 8	22.6 / 15.1 / 14.9 / 8.8
series 1–5 pairwise (dark)	≥ 18	≥ 9.5	≥ 9.5	warn ≥ 7	18.9 / 9.99 / 10.09 / 7.13
court trio pairwise (aliases of series 5, 2, 3)	≥ 20	≥ 12	≥ 12	warn ≥ 8	42.6 / 27.4 / 20.0 / 22.2
grass ↔ brand; clay ↔ error	≥ 12	warn ≥ 7	warn ≥ 7	—	25.9 / 20.5 / 23.8; 15.9 / 7.4 / 7.5
tiers pairwise (light / dark)	≥ 15	≥ 10	≥ 10	warn ≥ 8	25.1 / 22.7 / 18.1 / 11.4; 19.7 / 14.6 / 12.8 / 11.2
dark primary ↔ dark success	≥ 9	≥ 12	≥ 12	warn	12.0 / 13.9 / 16.3 / 1.8
dark live banner ↔ dark deadline banner	≥ 25	≥ 25	≥ 25	—	62.2 / 56.5 / 55.9 / 53.6

Canaries (project law: a gate needs a known positive). Ship the gate with fixtures that must fail: current #14804A vs brand (6.7 / 7.0 / 6.4); current tiers (bronze↔grandmaster deutan 3.4); current court trio (clay↔grass protan 1.7, grass↔brand 2.3); dark #68BF8E vs #70D5A2 (5.5); dark error Lc 40.5; light/dark banner pair deutan 4.2 / 3.1.

Porting test vectors (helper → JS), to 0.1 / 0.01: cvd('#006D37','deutan') = #625B3B, protan #6D6333, tritan #006B60; cvd('#14804A','deutan') = #736C4E; cvd('#00796A','deutan') = #66676B; cvd('#B8332E','protan') = #5A522C, deutan #7B6F29; cvd('#F5B301','deutan') = #DDC518; ΔE2000 #006D37/#14804A = 6.699 (normal), 6.389 (deutan); #006D37/#00796A = 13.415 / 14.695 (protan) / 17.173 (deutan); apca('#00796A','#FFFFFF') = 75.936, wcag = 5.326; apca('#D4736E','#1A201B') = −40.463. The repo's APCA implementation (apcaContrast, constants 0.1.9) and the helper's (0.0.98G-4g) agree to ≤ 0.01 Lc on the 11 pairs compared (bench/lane_g/apca_cmp.mjs: e.g. #00796A/white 75.936 vs 75.936, #D4736E/card −40.463 vs −40.463), and WCAG ratios to ≤ 0.01; the self-test should still pin both.

---

6. Caveats

• CVD simulation. Machado 2009 at severity 1.0 is dichromacy, the worst case; most colour-vision-deficient people are anomalous trichromats (Machado et al., IEEE TVCG 15(6)). The page does not state the colour space the matrices act in; the helper applies them in linear RGB (the usual implementation). ΔE2000 on simulated colours is a proxy, not a discrimination experiment, and it is not size-aware: smaller marks need more difference (Szafir 2018), so 12 px dots deserve more margin than the 40 px court tiles measured here.
• Thresholds are benchmarks, not standards. "OI parity" (≥ 11.5 under deutan) is derived from the best published sets measured with the same tool (table D); no standard fixes a ΔE for colour coding. ISO 9241-125 was not readable (paywall).
• Peers. Polaris, Wise, Atlassian values come from published npm packages (@shopify/polaris 13.9.5, @transferwise/neptune-tokens 8.20.2, @atlaskit/tokens 20.4.0); Seed from color.yaml via fetch; Carbon from its docs page. Spotify, Naver, Grab: nothing primary found, no claim made. Robinhood: only the accessibility setting. Okabe–Ito hexes were read from a package that cites Wong 2011 (secondary); the Okabe–Ito and Tol guidance quotes come from fetched pages summarised by the fetch tool.
• Not from the helper: CAT16 adaptation (adapt.py; matrix equals colour-science's CAT_CAT16), Helmholtz–Kohlrausch Γ (colour-science 0.4.7, Nayatani 1997 object-colour method), the OKLCH→hex inverse used by the scans. Adaptation assumes the whole field adapts to the ground colour (an upper bound) and ignores simultaneous contrast; the dark-card rows are overestimates because the eye adapts to the near-white text.
• Healey & Enns 2012 survey was not read (wrong file served); the seven-colour finding is from Healey 1996. Spectrum's 6 / 12 statement is from a search excerpt. Ware figures are via slides. Adobe Leonardo and Spectrum token values were not retrieved (JS-rendered).
• APCA. The helper implements 0.0.98G-4g; the font-size minima use the published LUT (strict: for example #0D100E on white is 14.4 px @400). Under that LUT no pair in this app reaches 14 px at weight 400 (even #0D100E on white needs 14.4 px); secondary and tertiary grey text (Lc 79.8 / 71.4 on basement) need ≥ 15.8 / 17.4 px at weight 500 — a typography finding outside this lane, noted because the neutrals verdict (keep) holds on WCAG/Lc but not on the LUT.
• Residual weak pairs (honest list): brand ↔ error protan 6.5; warning text ↔ error protan 7.4; success icon ↔ neutral badge deutan 3.0 (hence $successText); terracotta series ↔ warning text 14 / 1 / 4 and ↔ error 16 / 7 / 7; plum ↔ info 18 / 9 / 17; olive-lime shares hue 119 with optic (ΔE 21); dark primary ↔ dark error deutan 5.1; every teal vs brand under tritan ≤ 8.3 (table P). All are mitigated by glyph + label + role-exclusive placement, not by hex.
• Eyeball on device before shipping: #6B4A00 (reads brown), #8D3D00 (terracotta, darker than today's clay), #83971F (olive-lime grass), #014B40 (very dark teal). They are computed, not judged; a Pixel-class low-gamut panel may crush the three dark ones.
• Product decisions embedded: merging badge.accent ("open") into badge.success; replacing the deadline banner's brown fill by amber with ink; grandmaster moving from red to ruby; court dots becoming series hues (clay darker).
• Tamagui v5 styleCompat: 'legacy' and the tokens-theme-agreement test were not exercised; token names above are proposals.

---

7. Sources (every claim above traces to one of these; "S" = secondary)

• WCAG 2.2 Understanding 1.4.1: https://www.w3.org/WAI/WCAG22/Understanding/use-of-color.html · 1.4.11: https://www.w3.org/WAI/WCAG22/Understanding/non-text-contrast.html
• APCA in a Nutshell (Lc 90/75/60/45/30/15): https://git.apcacontrast.com/documentation/APCA_in_a_Nutshell.html
• Okabe & Ito, Color Universal Design: https://jfly.uni-koeln.de/color/ · Wong 2011 (palette): https://www.nature.com/articles/nmeth.1618 (S for hexes: https://www.npmjs.com/package/safepalette)
• Paul Tol, colour schemes: https://sronpersonalpages.nl/~pault/
• Machado, Oliveira, Fernandes 2009: https://www.inf.ufrgs.br/~oliveira/pubs_files/CVD_Simulation/CVD_Simulation.html
• Brewer, ColorBrewer scheme types: https://colorbrewer2.org/learnmore/schemes_full.html
• Healey 1996: https://healey.csc.ncsu.edu/publications/15843-choosing-effective-colours-for-data-visualization · UBC TR-96-10: https://www.cs.ubc.ca/tr/1996/tr-96-10 · Healey & Enns 2012 (not read): https://www.csc2.ncsu.edu/faculty/healey/download/tvcg.12a.pdf
• Szafir 2018, Modeling Color Difference for Visualization Design: https://cmci.colorado.edu/visualab/VisColors/ (paper read from https://par.nsf.gov/servlets/purl/10074616)
• IBM Carbon: palettes https://carbondesignsystem.com/data-visualization/color-palettes/ · status indicators https://carbondesignsystem.com/patterns/status-indicator-pattern/
• Adobe Spectrum data viz (S, search excerpt): https://spectrum.adobe.com/page/color-for-data-visualization · Leonardo (S): https://leonardocolor.io/
• Material 3: color_spec_2021.ts https://raw.githubusercontent.com/material-foundation/material-color-utilities/main/typescript/dynamiccolor/color_spec_2021.ts · core_palette.ts https://raw.githubusercontent.com/material-foundation/material-color-utilities/main/typescript/palettes/core_palette.ts · blog (S, summary) https://material.io/blog/science-of-color-design
• Radix Colors: https://www.radix-ui.com/colors/docs/palette-composition/understanding-the-scale · composing a palette (S, excerpt) https://www.radix-ui.com/colors/docs/palette-composition/composing-a-palette · package https://www.npmjs.com/package/@radix-ui/colors
• Tailwind CSS 4.1.11 theme.css: https://www.npmjs.com/package/tailwindcss
• Shopify Polaris: https://www.npmjs.com/package/@shopify/polaris (13.9.5 build/esm/styles.css) · https://shopify.dev/docs/apps/design/visual-design
• Wise Neptune tokens: https://www.npmjs.com/package/@transferwise/neptune-tokens (8.20.2 themes/personal*/tokens.css)
• Atlassian tokens: https://www.npmjs.com/package/@atlaskit/tokens (20.4.0 figma/atlassian-light.json)
• 당근 Seed: https://github.com/daangn/seed-design/blob/dev/packages/rootage/color.yaml
• Robinhood accessibility options: https://www.robinhood.com/us/en/support/articles/accessibility-options
• CIE 1924 V(λ) (S): https://en.wikipedia.org/wiki/Luminous_efficiency_function · Nayatani 1997 implementation: https://colour.readthedocs.io/en/latest/_modules/colour/appearance/hke.html
• CAT16: Li et al. 2017 https://doi.org/10.1002/col.22131 (matrix checked against colour-science)
• Colour count (S): https://data.europa.eu/apps/data-visualisation-guide/colour-for-categories · https://www.cs.ubc.ca/~tmm/talks/sig03.multimodal/mgp00025.txt · https://shop.elsevier.com/books/information-visualization/ware/9780123814647 · ISO 9241-125 scope https://www.iso.org/standard/64839.html (body paywalled) · ISO 22324 https://en.wikipedia.org/wiki/ISO_22324
• Repo (read-only): packages/app/src/config/themes/{classic-grass.theme,shared-semantics,status-badge-palette}.ts, packages/ui/src/{tokens,flag-medallion-presets,theme-hex}.ts, packages/app/src/domain/rules/tier.config.ts, packages/app/src/config/visual-accents.ts, scripts/check-color-contrast.mjs, scripts/lib/color-contrast-core.mjs, docs/design/research/visual-benchmark-2026-10-08.md (+ lane D appendix).