Review 4022
#4022 fix: WCAG 2.1 AA contrast failures in core color tokens (#3654) by gonzoblasco (bucket: contributor)
c469e449dc60af29143ec0798e05dd15b02b03f0
LOOP VERSION: 1.6.0 AUDIT RUBRIC: 1.13
LANE: full WHY: The fork PR adds two executable scripts, changes an existing theme token across many visible consumers, and has a prior changes-requested review.
WHY 1: The base theme's emphasized border is below the 3:1 non-text contrast minimum against its standard light and dark surfaces. WHY 2: People with low vision can miss active field and indicator boundaries that rely on this token. WHY 3: A base design-system token repeats that perception cost across every default-theme product using those boundaries. USER-FACING PROBLEM: A person using a default-theme form can have trouble perceiving active control boundaries because the emphasized border blends into its surface. PROBLEM SEVERITY: harmful friction — the boundary is present, but its contrast is below the system's WCAG 1.4.11 contract.
VERDICT: clear
The base theme replaces the emphasized border's light and dark neutral values with a darker light value and lighter dark value. Components continue to consume the same semantic token, so ownership and override behavior do not change. The PR also adds two standalone analysis programs that are not required to ship or verify the focused token correction.
SOLUTION (2 decisions · 1 shipped-token line plus 317 standalone script lines of 319 changed lines)
- Change the base theme's emphasized-border pair from
#CCD3DB/#494D53to#898D93/#6E7276— serves the contrast problem. - Add
scripts/audit-contrast.mjsandscripts/find-compliant-colors.mjsfor one-off analysis — unnecessary to the focused source change and must be removed.
BURDEN: medium — the shipped fix is one value pair, but the PR also adds 317 lines of standalone executable maintenance. BURDEN MATCH: disproportionate — the two programs are unnecessary for the focused token correction.
VERDICT: BLOCKS — remove exactly the two added script files.
OWNER: the base theme's semantic color-token layer.
TIER 1: the existing colorDefaults and generated-theme contrast contract; no new system.
TIER 2: none.
SEAMS: default-theme consumers of --color-border-emphasized; explicit custom-theme overrides retain their own values.
BEHAVIOR UNIT: n/a — no runtime behavior or state machine changes.
Static source search finds the existing token in fields, indicators, dividers, calendar marks, resizable handles, and other boundaries; the key and cascade remain unchanged.
VERDICT: clear
Default-theme users get stronger emphasized borders across the existing consumers of this shared token. Explicit theme overrides are unchanged. Removing the scripts changes no shipped runtime behavior.
VERDICT: note — the source delta intentionally recolors a broad set of default-theme boundaries; rendered pixels were not inspected under the mandatory safety stop.
No API change.
OSSIFICATION: no prop, export, theme target, token key, or accepted value is added; an existing base-token value changes in place.
VERDICT: clear
~ --color-border-emphasized: light-dark(#CCD3DB, #494D53) → light-dark(#898D93, #6E7276) in the base theme. No target or variable is added or removed. Existing custom themes that explicitly override this token remain unchanged.
Using the landed WCAG formula and exact base surfaces, the light value is 3.02:1 against body and 3.34:1 against surface/card/popover; the dark value is 3.00:1 against popover, 3.39:1 against surface, and 3.89:1 against body. The old worst cases were 1.37:1 and 1.71:1. This meets the public rule that UI foreground/background pairs reach 3:1 in both modes.
VERDICT: clear
BEHAVIOR: no — no state, default, conditional, or event path changes. API: no — the token key and all public types remain unchanged. VISUAL: yes, intended — the existing emphasized border becomes darker in light mode and lighter in dark mode across default-theme consumers. THEME: no compatibility break — existing token references and explicit custom-theme overrides still resolve.
VERDICT: note — the visual change is the stated fix; manual rendered evidence is unavailable because the fork's executable additions triggered the mandatory safety stop.
EFFECTS: zero. RENDER: no new render work. LISTENERS/OBSERVERS: none. LAYOUT: no geometry or layout code changes. BUNDLE: no dependency or imported runtime module change; the two standalone scripts are not referenced by shipped source.
VERDICT: clear
VISUAL CHECK: manual frames required WHY: This is an intended shared-token color change across many visible endpoints. Static contrast evidence establishes the ratio, but a visual gate cannot substitute for manual frames on an intended pixel change. The untrusted fork added executable scripts, so the mandatory safety stop prohibited checkout, server, browser, and screenshot work.
VERDICT: not looked at — no browser or rendered frames were permitted.
REMEDY SEARCH: not triggered — no proven visual defect
Static accessibility contract: the changed pair meets the landed 3:1 non-text threshold against all base body/surface/card/popover backgrounds in light and dark. No strings, locale behavior, direction, semantics, focus, or ARIA are changed. Rendered contrast and component states were not driven because the mandatory safety stop prohibited browser work.
VERDICT: not looked at — static contrast math is supportive, but rendered endpoints were not measured.
| slot | verdict |
|---|---|
| PROBLEM | clear |
| SOLUTION | BLOCKS — two unnecessary scripts |
| ARCHITECTURE | clear |
| IMPACT | note — broad intended recolor; no rendered inspection |
| API | clear |
| THEMING | clear |
| BREAKING | note — intended visual change |
| PERFORMANCE | clear |
| VISUAL | not looked at — mandatory safety stop |
| A11Y & I18N | not looked at — mandatory safety stop |
GOAL: met statically — the exact new pair reaches at least 3:1 against every base body/surface/card/popover background in both modes.
DISPOSITION: the two unnecessary executable additions → blocks now; remove exactly those files and keep the focused token delta.
ADVICE: bounded outcome criteria — delete scripts/audit-contrast.mjs and scripts/find-compliant-colors.mjs; do not broaden the source change.
AUTHOR CAN PROCEED: yes — remove exactly the two script additions.
WORST OUTCOME: “A future maintainer inherits 317 lines of executable audit code that the focused one-line token correction does not need.” → request changes.
JUDGEMENT NEEDED: none — the focused token direction and script removal are settled.
request changes
- [BLOCKS] Two standalone executable programs are unnecessary to this focused token correction.
→ The contributor's one-line contrast fix is held behind 317 lines of unrelated maintenance. ·
scripts/audit-contrast.mjs:1;scripts/find-compliant-colors.mjs:1Confirmed independently: complete-diff/API search finds no integration outside the two added files, and the author already offered to split them from the token update. Exact-head CI also fails its copyright check on only these two files, so deleting them resolves that failure without a broader ask.
The prior review by @ernestt asked to leave disabled tokens unchanged; the author did so. This review accepts that resolution and independently evaluates only the surviving base-theme border change.
Thanks for narrowing this after @ernestt's review; reverting the disabled-token changes addressed that concern. The emphasized-border update is separable and aligns with the 3:1 non-text contrast contract.
Please remove the two unnecessary executable additions, scripts/audit-contrast.mjs and scripts/find-compliant-colors.mjs, and leave the base-theme token update as the focused change. If you'd rather talk it through with someone, we're in Discord.
[Reviewed by Robohands]
None.
- Stable visual regression passed at the exact head; intended pixels still require manual frames under the review rules.
- Exact-head
testfailed only because the two added scripts lack copyright headers; their required deletion removes that failure. - Current
mainhas not changedtokens.stylex.tssince this head mergedmain.
TIME total 10m setup/rules 2m full kit, active versions, and ratified ruling; no worktree install/build 0m mandatory safety stop server 0m mandatory safety stop browser/a11y 0m mandatory safety stop focused tests 0m mandatory safety stop static analysis 4m complete diff, source/history, prior review, token composition critique/wiki 4m two critic gates and public record remote CI wait 2m overlapped static analysis; exact-head run reached terminal state waste 1m initial local source query used a shell-reserved variable and was rerun
- Rendered appearance and browser-level contrast at the exact head, because the mandatory safety stop forbids checkout and execution.
- The separate Vercel preview failure was not diagnosed and is not charged to the focused source delta.
Posted as drafted: request changes. Cindy’s ruling was applied before drafting: the disabled-token changes remain reverted, the base-theme border change was reviewed independently, and the request is limited to removing the two scripts.
Pull request · exact head · prior reviewed head
#4022 fix: WCAG 2.1 AA contrast failures in core color tokens (#3654) by gonzoblasco (bucket: contributor)
519d3b0bc209d96eba7346c76dedada243773745
LOOP VERSION: 1.6.0 AUDIT RUBRIC: 1.13
LANE: full WHY: This is an intended shared-token visual change with a standing changes-requested review, so it is ineligible for the fast lane.
- Our prior review asked for exactly
scripts/audit-contrast.mjsandscripts/find-compliant-colors.mjsto be removed. The single-parent, author-authored commit519d3b0bdeletes exactly those two files (317 lines) and changes nothing else. - @ernestt's earlier request to leave disabled tokens unchanged remains satisfied.
- The prior blocking finding is deleted. Our
APPROVEDreview at this exact head clears our standingCHANGES_REQUESTED. @ernestt's earlierCHANGES_REQUESTEDstill stands, so GitHub's overall decision remainsCHANGES_REQUESTED.
VERDICT: clear
WHY 1: The base theme's emphasized border was below the 3:1 non-text contrast minimum against its standard surfaces. WHY 2: People with low vision could miss control boundaries and indicators that rely on this token. WHY 3: A base design-system token repeats that perception cost across every default-theme product using those boundaries.
USER-FACING PROBLEM: A person using a default-theme control can have trouble perceiving its boundary because the emphasized border blends into the surface. PROBLEM SEVERITY: harmful friction — the control remains usable, but its boundary misses the system's stated WCAG 1.4.11 contrast contract.
VERDICT: clear
The base theme makes the emphasized border darker in light mode and lighter in dark mode. Existing components continue to use the same semantic token, so the responsibility stays in the base theme and custom overrides keep working.
SOLUTION (1 decision · 1 runtime line)
- Change
--color-border-emphasizedfrom#CCD3DB/#494D53to#898D93/#6E7276— serves the stated contrast problem.
BURDEN: low — one existing token value pair changes; no runtime state, effect, listener, dependency, or new public surface. BURDEN MATCH: proportionate — the shared token is the existing owner for all affected boundaries.
VERDICT: clear
OWNER: the base theme's semantic color-token layer.
TIER 1: existing colorDefaults plus the generated-theme contrast contract; no new system.
TIER 2: none.
SEAMS: default-theme consumers change together; custom themes that explicitly override the token keep their own values.
BEHAVIOR UNIT: n/a — no behavior or state changes.
The affected radius includes form fields, checkbox/radio indicators, strong dividers, calendar marks, resize handles, chart guides, and other existing consumers. The key and cascade do not change.
VERDICT: clear
Default-theme users see clearer unchecked indicators, field boundaries, strong dividers, and other existing emphasized-border consumers in light and dark mode. Custom themes with an explicit override are unchanged.
VERDICT: clear — the stronger boundary is the intended result and the sampled interactive and decorative consumers retain their geometry and hierarchy.
No prop, export, accepted value, default, theme target, or token key changes.
OSSIFICATION: none — an existing token value changes in place.
VERDICT: clear
~ --color-border-emphasized: light-dark(#CCD3DB, #494D53) → light-dark(#898D93, #6E7276) at packages/core/src/theme/tokens.stylex.ts:72.
The public Design Conventions state: “A color is only meaningful relative to what it sits on, so foreground and background are decided together,” and identify any UI foreground/background pair below 3:1 as low contrast. The new base pair clears the complete base-surface matrix:
| mode | surface | ratio |
|---|---|---|
| light | body | 3.02:1 |
| light | surface/card/popover | 3.34:1 |
| dark | popover | 3.00:1 |
| dark | surface/card | 3.39:1 |
| dark | body | 3.89:1 |
No custom-theme value, target, cascade order, or override seam changes.
VERDICT: clear
BEHAVIOR: no — no state, default, conditional, or event path changes. API: no — the token key and public types remain unchanged. VISUAL: yes, intended — the existing emphasized border becomes darker in light mode and lighter in dark mode across base-theme consumers. THEME: no compatibility break — references and explicit custom-theme overrides still resolve.
VERDICT: note — the intended recolor is fully visible in the evidence below; sampled geometry is unchanged.
EFFECTS: zero. RENDER: no new render work. LISTENERS/OBSERVERS: none. LAYOUT: no geometry or layout code changes; measured indicator geometry stayed 24×24 and divider geometry stayed 804×1. BUNDLE: no dependency or imported runtime module change.
VERDICT: clear
VISUAL CHECK: manual frames required WHY: This is an intended shared-token pixel change. The exact-head stable visual job passed, but a gate cannot replace manual before/after frames for an intended visual change.
Light mode — before (left) / after (right):
Dark mode — before (left) / after (right):
The unchecked checkbox boundary moves from 1.51:1 to 3.34:1 on the light surface and 1.93:1 to 3.39:1 on the dark surface. Checked and indeterminate indicators in the same frames are controls and do not change.
Light mode — before (left) / after (right):
Dark mode — before (left) / after (right):
The strong divider takes the intended stronger token while the subtle divider remains the lower-emphasis control. Its 804×1 geometry is unchanged.
Before/after pairs match on every sensor except Build.
| frame pair | builds | story | globals / rendered mode | direction | viewport | semantic state | geometry | settled/error |
|---|---|---|---|---|---|---|---|---|
| indicator, light |
e6507a95fb13 / 519d3b0bc209
|
core-indicator--all-states |
astryxTheme:none, light |
ltr | 900×420@1 | unchecked checkbox indicator; target count 1 | 24×24, visible/in viewport | fonts loaded; 0 animations; 0 page/Storybook errors |
| indicator, dark |
e6507a95fb13 / 519d3b0bc209
|
core-indicator--all-states |
astryxTheme:none, dark |
ltr | 900×420@1 | unchecked checkbox indicator; target count 1 | 24×24, visible/in viewport | fonts loaded; 0 animations; 0 page/Storybook errors |
| divider, light |
e6507a95fb13 / 519d3b0bc209
|
core-divider--variants |
astryxTheme:none, light |
ltr | 900×520@1 | strong horizontal divider; target count 1 | 804×1, visible/in viewport | fonts loaded; 0 animations; 0 page/Storybook errors |
| divider, dark |
e6507a95fb13 / 519d3b0bc209
|
core-divider--variants |
astryxTheme:none, dark |
ltr | 900×520@1 | strong horizontal divider; target count 1 | 804×1, visible/in viewport | fonts loaded; 0 animations; 0 page/Storybook errors |
All eight source frames were opened after capture. The before server was warm main on port 6100; the after server was exact head on guarded port 6277.
VERDICT: clear — the intended contrast increase is present in both modes, controls remain visually differentiated, and sampled geometry is unchanged.
REMEDY SEARCH: not triggered — no proven visual defect.
The exact rendered unchecked-indicator boundary moves above 3:1 in both modes, and the complete static base-surface matrix remains at or above 3:1. The exact-head stable visual job passed. pr-a11y and pr-rtl were skipped because no component was modified; no strings, roles, focus, keyboard behavior, direction, or semantic state change.
VERDICT: clear
| slot | verdict |
|---|---|
| DELTA / PRIOR REVIEW | clear — both requested script deletions are exact |
| PROBLEM | clear |
| SOLUTION | clear |
| ARCHITECTURE | clear |
| IMPACT | clear |
| API | clear |
| THEMING | clear |
| BREAKING | note — intended recolor |
| PERFORMANCE | clear |
| VISUAL | clear — manual frames and receipts |
| A11Y & I18N | clear |
GOAL: met — exact rendered light/dark indicator ratios rise from 1.51/1.93 to 3.34/3.39, and every base surface reaches at least 3:1.
DISPOSITION: prior script finding → fixed and deleted; stale generated token docs → non-blocking re-review note, must be regenerated before merge and exact-head CI already enforces it; stale script references in the PR body → non-blocking re-review note.
ADVICE: proven existing path — run node scripts/generate-token-docs.mjs; it changes only the documented pair at packages/cli/assets/docs/tokens.doc.mjs:202-205.
AUTHOR CAN PROCEED: yes — regenerate the token docs; no design or architecture decision remains.
WORST OUTCOME: “A theme builder reading generated token docs still sees the old pair.” → approve with a before-merge note; CI prevents that stale output from landing.
JUDGEMENT NEEDED: none — this is a defect fix against the existing 3:1 contract.
approve
-
[not blocking] Generated token docs still carry the old values. → a theme builder sees stale defaults until the generator output is committed ·
packages/cli/assets/docs/tokens.doc.mjs:202-205Confirmed independently by the exact-head CI failure and a local--check; running the generator changes only this pair. -
[not blocking] The PR body still says the deleted scripts are included. → a reviewer sees the old three-file scope instead of the current one-file change · PR description, “Scope note” and “Methodology”
Thanks, both scripts are gone, clearing my previous block. The token change looks good. Before merge, please regenerate tokens.doc.mjs and remove stale script references from the PR body.
[Reviewed by Robohands]
None.
- Exact-head
Stable visual regression, build, Storybook, sandbox, lint, smoke, dependency, docsite, component, and theme-layer checks passed. - Exact-head
testfailed only because generatedtokens.doc.mjsstill has the old pair; reproduced locally. - Our approval cleared our own standing block. GitHub's overall decision remains
CHANGES_REQUESTEDbecause @ernestt's earlier review still stands. - Vercel failed separately; its deployment detail was not diagnosed here. All GitHub-hosted builds passed.
- CI wait: 0m — all exact-head jobs were terminal when checked.
TIME total 17m setup/rules 3m fork safety screen, complete loop kit, prior reviews/wiki, rubric 1.13 build/server 3m fast install, one 6s build, exact-head Storybook on guarded 6277 browser/a11y 4m 8 sensored Chromium frames, 2 stories × 2 modes × before/after focused tests 0m 3 files, 110 tests, 5s code/history 2m exact delta, source radius, token history, main movement, CI diagnosis critique/wiki 5m self-critique, durable frames, round/index update CI wait 0m exact-head jobs already terminal waste 1m corrected the guard location and rebuilt contact sheets after local image tooling was absent
- The separate Vercel deployment failure's cause.
- Every token consumer was not rendered; the indicator family and strong Divider cover interactive and decorative endpoints, with full source-radius enumeration.
MERGE: hold(public exact-head CI is red until generated token docs are updated; @ernestt's earlier CHANGES_REQUESTED also remains)
Posted unchanged as APPROVED at exact head 519d3b0bc209d96eba7346c76dedada243773745 on 2026-08-27T18:30:16Z. This clears our previous CHANGES_REQUESTED; @ernestt's earlier CHANGES_REQUESTED still controls GitHub's overall decision. Exit/list: approved, open, and blocked on the author/reviewer path — regenerate tokens.doc.mjs, remove stale script references from the PR body, obtain clearance of the remaining review, and resolve required red checks before merge.
Pull request · exact head · prior approved head
#4022 fix: WCAG 2.1 AA contrast failures in core color tokens (#3654) by gonzoblasco (bucket: contributor)
b5aac840f19db1657f8c137401c50a5bbeae8224
LOOP VERSION: 1.6.0 AUDIT RUBRIC: 1.13
LANE: fast WHY: The delta has zero runtime decisions and changes four generated-document lines only; it adds no API, token, state, dependency, listener, or architecture owner, changes no rendered input, leaves no prior finding unresolved, and one generated-output check settles the goal.
- The single-parent author commit
b5aac840has exact parent519d3b0band changes onlypackages/cli/assets/docs/tokens.doc.mjsby 2 additions and 2 deletions. - Our prior approval asked for the generated token reference to be refreshed. The new row now matches the source pair
#898D93/#6E7276;node scripts/generate-token-docs.mjs --checkreports all 188 tokens up to date and leaves the worktree clean. - Our original script-removal block remains fixed: neither script is present in the current PR diff. @ernestt's earlier request to leave disabled tokens unchanged also remains satisfied.
- The current PR body contains no
audit-contrast.mjs,find-compliant-colors.mjs,scripts/, or equivalent stale script-scope reference. Stale body-only scope text would be a non-blocking accuracy nit under the current rules; at this head there is no remaining nit.
VERDICT: clear
The source token changed while its generated CLI reference still showed the old light and dark values, so a builder reading the reference could copy defaults the package no longer uses.
VERDICT: clear
Regenerate the caller-facing token reference from the source token table. The generated row now exposes the same pair as the shipped source, with no hand-authored collateral change.
SOLUTION (1 generated-output decision · 0 runtime lines of 4 changed lines)
- Replace the stale documented pair with
#898D93/#6E7276.
BURDEN: low — one generated row; no runtime state, effect, dependency, or maintained parallel logic. BURDEN MATCH: proportionate — the existing generator remains the single owner.
VERDICT: clear
OWNER: the token source and its existing generated CLI reference.
TIER 1: scripts/generate-token-docs.mjs remains the source-to-reference path.
TIER 2: none.
SEAMS: the full generated reference remains byte-consistent with all 188 source tokens.
BEHAVIOR UNIT: n/a — no runtime behavior.
VERDICT: clear
CLI documentation readers now see the same emphasized-border defaults that the package ships. Runtime consumers receive no new change relative to the prior approved head.
VERDICT: clear
No prop, export, accepted value, default, target, or token key changes in this delta.
OSSIFICATION: none — generated documentation catches up to an already-reviewed existing-token value change.
VERDICT: clear
No theming runtime changes in this delta. The documented --color-border-emphasized pair now exactly matches light-dark(#898D93, #6E7276) in the source; every other generated token row remains current.
VERDICT: clear
BEHAVIOR: no — runtime and rendering trees are byte-identical to the prior approved head. API: no — no API-bearing file changed. VISUAL: no new visual change — the current delta is generated data only. THEME: no — no token key, value, cascade, target, or override changes relative to the prior approved head.
VERDICT: clear
EFFECTS: zero. RENDER: no reachable change. LISTENERS/OBSERVERS: none. LAYOUT: no rendered code changed. BUNDLE: no shipped module or dependency changed.
VERDICT: clear
VISUAL CHECK: not applicable
WHY: The only current-head delta is generated CLI reference data. packages/core, packages/lab, packages/themes, apps/storybook, and .storybook are identical to the prior approved head; the packages/core tree hash remains e72e8c647ceb285d7c1468c075878f23ab75ab8c. The prior round's exact-runtime-head screenshots therefore remain the evidence for the already-reviewed intended recolor. No Storybook server or browser was started for this docs-only delta.
VERDICT: clear
REMEDY SEARCH: not triggered — no proven visual defect
The generated values now preserve the already-reviewed 3:1 contrast result. No component, string, role, focus, keyboard, direction, locale, or semantic-state path changed; exact-head pr-a11y and pr-rtl correctly skipped because no component changed.
VERDICT: clear
| slot | verdict |
|---|---|
| DELTA / PRIOR REVIEW | clear — every prior ask is reconciled |
| PROBLEM | clear |
| SOLUTION | clear |
| ARCHITECTURE | clear |
| IMPACT | clear |
| API | clear |
| THEMING | clear |
| BREAKING | clear |
| PERFORMANCE | clear |
| VISUAL | clear — no current-delta visual obligation |
| A11Y & I18N | clear |
GOAL: met — the generated-output check passes for all 188 tokens and the documented pair matches source. DISPOSITION: prior script block → remains fixed; prior generated-doc note → fixed; prior PR-body note → fixed; no new findings. ADVICE: omitted — no work remains from this review. AUTHOR CAN PROCEED: yes — no review acceptance criterion remains. WORST OUTCOME: “No builder-facing mismatch remains; the generated reference matches its source.” → approve.
JUDGEMENT NEEDED: none.
approve
Thanks, the generated token docs now match the source and the stale script references are gone, clearing both notes from my prior approval.
[Reviewed by Robohands]
None.
- Current GitHub head is exactly
b5aac840f19db1657f8c137401c50a5bbeae8224; the commit is author-authored, single-parent, and directly follows prior approved head519d3b0bc209d96eba7346c76dedada243773745. -
node scripts/generate-token-docs.mjs --check:✓ tokens.doc.mjs is up to date (188 tokens); worktree clean. - Exact-head GitHub-hosted build, test, lint, Storybook, sandbox, smoke, dependency, docsite, component, theme-layer, stable visual, CLA, review-signal, and visual-acceptance checks are terminal and green. Vercel remains separately red, as on the prior approved head; its cause was not needed to settle this generated-doc delta.
- GitHub reports
MERGEABLE, but the overall review decision remainsCHANGES_REQUESTEDbecause @ernestt's historical review is still active.
TIME total 4m setup/rules 1m fresh loop 1.6.0 wiki, rubric 1.13, fire message, prior rounds, and safety guard build/server 0m no install, build, or server; current delta has no visual obligation browser/a11y 0m no rendered or interaction delta focused checks <1m generated-output check, clean-tree proof, and source/doc comparison code/history 1m exact parent/delta, all prior findings, PR body, runtime tree identity, and current-head CI critique/wiki 2m one critic pass and round/index publication CI wait 0m all exact-head jobs were already terminal
- The separate Vercel deployment failure's cause; it is unchanged in relevance from the prior approved round and does not affect the generated-output proof.
Posted unchanged as APPROVED at exact head b5aac840f19db1657f8c137401c50a5bbeae8224 on 2026-08-27T19:16:33Z (review id 5044764291). Exit/list: approved and open in pipeline 4; generated docs and PR-body notes are cleared, while @ernestt's historical CHANGES_REQUESTED and the separate red Vercel status still prevent merge. No merge or auto-merge action was taken.