#Contrast

Generated. Do not edit — packages/core/src/theme/contrast.test.ts writes this file, and fails if the committed copy and a fresh run disagree, so a re-tinted token cannot land without the number moving here too. Regenerate with npx vitest run -u.

Every ratio is WCAG 2.2 relative luminance over sRGB, rounded to two decimals, computed from the token values themselves. The arithmetic is checked in the same test against published reference values (#767676 on white is 4.54:1; #949494 on white is 3.03:1) so the audit cannot be wrong and self-consistent at once.

#What is gated, and why each pair

A pair is gated when the shipped code paints it. The floor follows from what that line draws — 4.5:1 for text (WCAG 1.4.3), 3:1 for a graphic or a control boundary (1.4.11) — and never from the token name.

This is the part a blanket rule gets wrong in both directions. accent on canvas is exactly 3.00:1 in the Twitter light palette: a blanket 4.5:1 would call the deliberate choice a defect, when the vivid blue is kept precisely because it is a mark on a map. And nothing in the name hover suggests the toolbar puts body text on it, which it does.

#Summary

140 gated pairs across 7 themes: 104 at or above their floor, 36 below.

#blaeu-default

#twitter-light

#twitter-dim

#twitter-black

#survey-paper

#high-contrast

#imagery-dark

#Below the floor

Grouped by rule, because these do not scatter — a pair that fails tends to fail in most themes at once, and that is the signal. A rule failing everywhere is one token doing a job it was not chosen for, not seven separate mistakes.

#text on hover — needs 4.5:1

plugin-ui/src/styles.ts:99,224 — the button and issue-row hover fill; the label keeps color: inherit

#textMuted on surfaceMuted — needs 4.5:1

plugin-ui/src/styles.ts:158-160 — the attribution, at 11 px

#onAccent on hover — needs 4.5:1

plugin-ui/src/styles.ts:99 over :108-111 — a hovered pressed toggle. :hover:not(:disabled) is specificity (0,4,0) and [aria-pressed='true'] is (0,3,0), so the hover fill wins the background while the pressed rule still supplies color: onAccent

#error on hover — needs 3.0:1

plugin-ui/src/styles.ts:224-225 — the same stripe, on a hovered row

#accent on hover — needs 3.0:1

plugin-ui/src/styles.ts:224,227 — the info severity stripe on a hovered row, exactly as error and warning above

#warning on hover — needs 3.0:1

plugin-ui/src/styles.ts:224,226 — the same stripe, on a hovered row

#border on canvas — needs 3.0:1

plugin-ui/src/styles.ts:143 and :240 — the only boundary of the scale bar and of the status line. Neither carries .bl-ui-control (ScaleBar.ts:34, plugin.ts:160), so neither gets its box-shadow, and both sit over the map

These are recorded, not waived. contrast.test.ts pins every one of them to the exact ratio printed above, and the pin is an equality: any movement at all, in either direction, fails until the row is updated or deleted. It also asserts that this list and the set of pairs actually below their floor are the same set — so a pair cannot begin failing without being written down here, and a fixed one cannot stay written down.

#Measured but not gated

A pair with no rule is a pair nobody has thought about; a pair here is one somebody has.

#vertexActive on vertex

plugin-edit/src/handles.ts:226-229 — a 2 px ring around a white disc.

A handle is perceivable because its ring clears 3:1 against the ground, which is gated above. How much the ring differs from its own fill is a legibility judgement, not a WCAG floor.

#vertexActive on midpoint

plugin-edit/src/handles.ts:239-242 — a 1 px ring, the thinnest boundary in the system.

Same as above, and worth watching: this is the pair that runs lowest.

#accent on vertex

plugin-draw/src/preview.ts:39-44, plugin-measure/src/layers.ts:69-74 — the draw and measure vertex stroke.

Same as above.

#surface on canvas

plugin-ui/src/styles.ts:57 — a panel floating over the map.

Every shipped theme sits a panel within ~1.2:1 of its own ground, deliberately: the chrome is meant to recede. That is precisely why border on canvas is gated instead.

#surfaceMuted on canvas

plugin-ui/src/styles.ts:158 — the attribution, which has neither border nor shadow.

Same, with a sharper edge: the attribution has no boundary at all, so nothing here can carry 3:1. Recorded so the number is on file.

#labelHalo on canvas

preset-cadastre/src/layers.ts:51 — the halo behind an on-map label.

A halo does its job by matching the ground, not by contrasting with it. Near 1:1 is correct here, and a high number would be the bug.

#Tokens with no rule at all

accentMuted and success are declared in ThemeTokens, given a value by every theme, mirrored to a CSS custom property — and read by nothing in packages/*/src. There is no call site, so there is no pair, so there is no floor. Several themes carry a hand-computed ratio in a comment for success, which is arithmetic performed for a consumer that does not exist.

#What this document does not cover

The three preset themes — preset-cadastre, preset-urban, preset-game — are not audited. They are applied through theme.set() rather than registered, and they live in packages @blaeu/core may not import, so gating them needs a test inside each of those packages. Nothing checks their contrast today.

Opacity is not modelled either. Several of the gated marks are painted with a fill opacity — the selection fill at 0.25, the parcel fill at 0.08 — and the ratios here are for the token at full strength, which is what the accompanying line or outline is drawn at. A composited fill over an unknown basemap has no single ratio to compute.

One unmodelled opacity does land on a text pair, with no full-strength companion: plugin-ui/src/styles.ts:100 composites a disabled button — label included — at 0.45. The text on surface row below is therefore optimistic for every disabled toolbar and history button, by an amount nothing here computes.