Read-only task rendering acceptance

This slice is verified locally on feature/markdown-task-lists. Live deployment, physical assistive-technology/device testing and the full suite remain unverified. The in-page counter and wider Docs column remain separate follow-ups.

UAT correction: the initial native-disabled painting failed user visual acceptance. The CSS correction and renewed local evidence below supersede the earlier appearance assessment. Maintainer UAT remains pending.

Corpus evidence

The frozen and refreshed reports retain source/HTML hashes, tags, errors and candidate counts without source/HTML dumps. The frozen manifest uses tracked, catalog-eligible paths at the intake revision and verifies every source hash. Later report/runbook edits are outside these snapshots; the original closeout used the verified Stage 2 renderer/CSS. The UAT correction below changes CSS only; renderer/corpus results remain applicable.

All 47 measured task documents are classified against their original hashes:

Every document independently accounts for its checked and unchecked candidates. Forty documents contain displayed tasks without retained task-source candidates, three combine display/source, and four retain their tasks as source. Unsupported list blocks remain readable under the existing list contract; no source rewrite was used to improve these totals.

The audit inspects real HTML for disabled checkbox type, exactly one nonempty label, no generated IDs/names/submission/action attributes, no descendant/paragraph capture, bounded list structure and unchanged safety-filtered source recovery. It saves state and label-text hashes, not task text or HTML. These parsed labels are not a substitute for browser accessibility naming or physical assistive-technology review. Source candidates are heuristic, not a general syntax-conformance guarantee.

Reproduce from this cycle's checkout:

python3 -B scripts/markdown_task_corpus_audit.py --baseline docs/plans/markdown_task_lists/markdown_task_lists_metrics.json --output /tmp/task-audit.json

Reader and delivery evidence

Dynamic and generated readers cover desktop/390px × default/Word97/Chicago/Midnight with JavaScript disabled. Checks include ordinary Tools→Docs→browse→onboarding checklist discovery, browser role/name/checked/disabled state, repeated task names without IDs, wrapped first-paragraph labels, child/later-paragraph exclusion, pointer/Space immutability, Tab skipping and actual task-link activation. Existing list/table/code/source focus/scroll and no-overflow checks remain covered. Representative screenshots across all phone themes and desktop samples were inspected; all 47 documents were classified structurally, not individually visually reviewed.

Copied-package verification covers actual full/shared CLI publication, task/list/table/fallback parity, exact Markdown downloads, stale/deleted readers, injected renderer-failure 503, retained published task page, restoration and successful retry. Static discovery supplies eight further scenarios: 24 distinct reader/discovery theme/viewport scenarios overall.

Final focused task/list/table/reader/catalog tests pass 52/52. The complete static inventory/on-demand tests pass 2/2 in an isolated source snapshot whose hashes identify the initial tested candidate. The renderer remains unchanged; corrected CSS has separate evidence below. Metrics fixture, syntax, whitespace and artifact-link checks pass. These 54 focused cases are not a full-suite result.

No dependencies, task editing, counter, layout expansion, schema/navigation changes, deployment or recovery-storage disposal. The original closeout excluded unrelated research artifacts.

UAT contrast correction — 2026-10-23

Native disabled semantics and first-paragraph labels remain unchanged. CSS now draws an 18px outlined unchecked box, a solid checked fill and a contrasting tick, using theme colors. Forced colors use Canvas/CanvasText. Arena's transparent panel initially hid the tick; an opaque page-color tick fixes that case without theme-specific overrides.

Correction evidence records 46 dynamic and 46 generated Chromium cases: all 16 registered themes, Auto light/dark, three additional LCD palettes, forced light/dark, each at 1280px and 390px with JavaScript disabled. All checked/unchecked pair screenshots were inspected. Computed unchecked border/backdrop contrast is at least 8.46:1; tick/fill is at least 3.95:1. These are fixture color measurements, not a general accessibility certification or physical-device result. Hollow/solid fill and a check shape distinguish state without relying on hue.

The actual copied-package shared CLI refresh passed, followed by eight generated reader scenarios covering names/state, pointer/Space immutability, Tab skipping, links, source downloads and existing list/table/scroll behavior. Focused task/list/table/reader/catalog PHP checks pass 52/52. The original full-build/failure/recovery and corpus evidence remains historical; renderer/source boundaries did not change. No full-suite or production claim. User retest and merge/publication remain open in the consolidated checklist.

Reproduce the contrast sweep against an isolated host containing tests/fixtures/markdown_task_reader.md at the fixture route:

DOCS_BASE_URL=http://127.0.0.1:18087 DOCS_EVIDENCE_ROOT=/tmp/docs-task-contrast node tests/browser/docs_task_contrast.mjs

Maintainer acceptance — 2026-10-23

User accepted the corrected slice and requested merge into main. This supersedes the pending user-retest status above. Main integration preserves current origin/main work; production deployment, physical AT/devices and full-suite verification remain separate. See the consolidated checklist for current status.