Improve Search Index

Goal

Fix the selected search/index boundary across indexing pipeline, document schema, query/filter logic, authorization, UI/API state, tests, and docs/maps when used.

Scope selection

Use the selected index/query/resource, stale/missing/unauthorized result bug, audit finding, failing test, or explicit search improvement task. Avoid broad ranking redesigns unless the selected product acceptance requires them.

Change targets

• Index document construction, update/delete events, backfill/reindex scripts, and freshness markers.
• Query builder, filters/facets/sorting/pagination, language/tokenization settings, and result hydration.
• Permission/tenant/privacy filters and no-results/error states.
• Tests/replay fixtures for recall, freshness, deletion, permissions, and important ranking/order expectations.

Implementation principles

• Keep search index as a projection unless the repository explicitly makes it authoritative.
• Enforce access control before exposing results; do not rely on UI-only filtering.
• Preserve existing public query contracts or version them explicitly.
• Include reindex/rollback/forward-fix notes when document schema changes.

Validation

Run repository-native tests and bounded indexing/query scenarios for create/update/delete, stale index, permission filter, empty/error state, and reindex/backfill where relevant. Report managed search gates honestly.

Fix the document/normalization/ranking/filter or freshness owner using a versioned query corpus with known relevance judgments and misleading neighbours. Keep explicit restrictions intact and preserve public query/ID contracts. Rebuild the required projection when its schema changes, then feed the same corpus and update/delete sequence into the recall/freshness check. Recheck failures at that owner before Patch Checking; tuning only a few attractive queries is not evidence of broad retrieval quality.