Add evidence-based UI patterns with governed search coverage

[changeroa]

Summary

Add 12 experimental, locally authored Design Engineering patterns with explicit source dates, measurement boundaries, counterexamples, accessibility requirements, and local examples:

• Inline validation timing
• Error messages and recovery summaries
• Persistent form labels
• Localized input masks
• Faceted filter selection
• Empty states and no results
• Data table interaction
• Tabs and segmented controls
• Autocomplete acceptance
• Breadcrumb hierarchy
• Bottom navigation destinations
• Dark mode contrast

Extend the existing loading/feedback, native interaction, input/focus, and delete/undo owners rather than duplicating them. Route the collection from the domain hub and manifest. Admit the 12 leaves and index as 13 new materials, resealing the exact inventory at 193 documents and regenerating the registry. Package the new Markdown files without changing the package version.

New aliases, questions, and evaluation questions are English-only. Existing Korean metadata, queries, and support code are unchanged.

Evidence

All 29 cited primary URLs returned HTTP 200 and their source bodies were checked on 2026-10-19. The Apple HIG body was read through its official JSON representation. W3C's document fetcher returned empty bodies, so those sources were verified directly over HTTP.

Separate WCAG requirements, informative APG/platform conventions, observed usability findings, and local proposals. Do not present benchmark site prevalence as a user benefit or an observed extreme as an average. Consumer examples and proposed verification cases are not represented as executed consumer conformance evidence.

CI inventory correction

The first CI run found stale A2A and AG-UI source inventories after this change updated package exposure and material test fixtures. Refresh both inventories through the existing authoritative generator; preserve owners, unknown caller status, review dates, extension dispositions, and immutable archive bindings. No lifecycle validator or gate is changed.

Verification

Passed documentation gates and fixture tests for OKF, links, IA, domains, and governance; material validation and all material tests; Registrar and Conservator with controls and their tests; search contracts and the unchanged retrieval gate; consumer-reference/conformance/page-evidence/freshness dependent contracts; generated Layout, reference-artifact, evidence, and material drift checks; JavaScript syntax checks.

Checks ran on an isolated tracked snapshot. The shared checkout has three pre-existing ignored .tmp/*.md files that fail the whole-directory OKF walk; they were not edited or deleted. The new index was corrected to the repository's frontmatter-free OKF index format. Domain fixture inventories and material cardinality assertions were updated, not weakened. LSP diagnostics were unavailable because the workspace lacks TypeScript/Biome; runtime, syntax, and repository tests passed.

Evaluation	Questions	Hit@5	nDCG@5
Before, original set	75	0.9733	0.8618
After, same original set	75	0.9733	0.8577
After, expanded set	91	0.9780	0.8772
Added English questions only	16	1.0000	0.9688

The comparable original set has a small nDCG reduction; expanded-set improvement is not a like-for-like gain. Must-pass questions: 35/35. The two existing non-mandatory misses remain. Gate thresholds and ranker are unchanged.

No version bump, npm publication, or Slack send. .omo/ is untouched.

consumer_reference: not_applicable because this corpus expansion selects no consumer reference record.

Additional local verification: lifecycle validation, extension-boundary tests, dependent governance, and material drift pass after the inventory refresh. The full macOS lifecycle adversarial suite stops at the unchanged archive-rename fixture (scripts/test-lifecycle-disposition.mjs:507, ENOENT). Required Linux CI determines the full-suite merge result; no test was weakened or skipped. Final retrieval gate rerun: n=91, Hit@5=0.9780, nDCG@5=0.8772, must-pass 35/35.