ldaume Meilisearch
Use this for search-backed product features and operational search workflows.
Project Fit Check
Before changing search:
- Read existing index definitions, sync/backfill jobs, search clients, auth rules, settings, and tests.
- Identify whether indexes are global, tenant-scoped, user-scoped, or mixed.
- Confirm which fields are searchable, filterable, sortable, displayed, and safe to expose.
- Preserve existing relevance policy unless the task is to tune it.
- Plan reindexing and rollback before changing settings or document shape.
Index Rules
- Keep document IDs stable.
- Store only fields needed for search and display.
- Avoid indexing secrets, private prompts, tokens, or raw sensitive content.
- Make tenant/user ownership enforceable before query results reach the user.
- Treat settings changes as migrations: apply consistently and verify.
Relevance Rules
- Tune searchable attributes before adding complex ranking hacks.
- Use filters/facets for structured constraints.
- Add synonyms only when they reflect real domain language.
- Test typo tolerance against real queries, not only ideal examples.
- Record meaningful relevance changes in docs or changelog.
Sync And Backfill
- Prefer idempotent indexing jobs.
- Track source record id, index name, version, and last indexed timestamp when operations need observability.
- Handle deletes and permission changes, not only creates/updates.
- Batch conservatively and retry safely.
Verification
- fixture search tests for relevance-critical queries
- permission tests for tenant/user boundaries
- backfill smoke test on a small sample
- manual query check for changed facets or ranking rules
Red Flags
- search results bypass authorization
- private fields indexed for convenience
- settings changed without reindex plan
- relevance tuned from one anecdotal query
- delete path missing from sync