List-structure rendering acceptance

The approved list slice is technically complete locally on feature/markdown-list-structure. Live deployment, physical devices, full-suite verification and maintainer acceptance remain pending. No external component or wider Docs layout change was introduced.

Frozen and refreshed evidence

The frozen and refreshed summaries retain per-document source/HTML hashes, tags, errors, source candidates and report hashes without source/HTML dumps. The frozen render verifies every original source hash; later acceptance/runbook/report edits are outside these snapshots. Renderer/CSS were unchanged after the Stage 4 delivery candidate.

The nested-list audit covers all 453 original measured documents / 9,716 nested-item candidates. Actual output has balanced semantic parent/item relationships, explicit ordered-item values and maximum observed nesting depth three; adversarial fixtures separately verify eight supported levels and nine-level source recovery.

• 447 documents: structured nested lists, no list-source regions.

• 5 documents: structured nesting alongside unsupported source regions.

• 1 document: source-only nested candidates (docs/runbooks/qdb_archive_import.md, containing an embedded fenced example).

• 9 source regions across these six documents retain their complete safety-filtered source. Source indicators identify embedded fences/tables/quotes or remaining indentation/ownership ambiguity; they are nonexclusive clues, not universal parser diagnoses.

Source equality uses the canonical prose-comment boundary, with independent regression fixtures verifying comment privacy and example-byte preservation. Candidate counts are heuristic; classification does not prove all candidates are semantic nesting or that all 453 documents were visually reviewed. Representative real-guide and controlled-fixture checks establish the specific ownership behavior.

Reproduce from this cycle’s checkout with the recorded renderer (unchanged since 4876f2337; later documentation commits do not alter it):

python3 scripts/markdown_metrics.py --baseline docs/plans/markdown_renderer_extension/render_current_summary.json --output /tmp/list-frozen.json
python3 -B scripts/markdown_list_corpus_audit.py --metrics /tmp/list-frozen.json --output /tmp/list-audit.json

Reader and delivery

Dynamic and generated scenarios cover desktop/390px × default/Word97/Chicago/Midnight with JavaScript disabled: ordinary Tools→Docs→guide entry, six owned procedure paragraphs, mixed ancestry, literal repeats/gaps, parent paragraphs after children, deep-list and source Tab/arrow/focus, code bytes, safe list/table destinations and no page overflow. Representative phone screenshots in all themes and desktop samples were inspected; this is not exhaustive per-document or physical-device review.

Copied-package acceptance verifies complete full/shared CLI publication, dynamic/static lists/tables/fallback parity, stale/deleted-reader refresh, injected renderer-failure 503, retention of the active published release, restoration and successful retry. Static discovery adds eight scenarios. Together, reader/discovery evidence covers 24 distinct theme/viewport scenarios.

Final catalog/reader/list/table checks pass 46/46; the complete static inventory/on-demand checks pass 2/2. Hand-counted metrics, PHP/Node/Python syntax and whitespace checks pass. These 48 focused cases and release checks are not a full-suite or production-deployment claim.