Practice Bank: Vector Store Decisions
Use starter.py to recommend collection and release strategies.
Scenario 1: Multi-tenant support
Collection boundaries and tenant filters matter.
Scenario 2: New embedding model
Dimension and index version change.
Scenario 3: Policy retirement
Deletes and active-status filters matter.
Scenario 4: Large rebuild
Blue-green release reduces risk.
Scenario 5: Low-latency search
Approximate index needs recall checks.
Scenario 6: Legal citations
Stable ids and source references matter.
Scenario 7: Failed embedding job
Partial index should not be promoted.
Scenario 8: Access leak
Metadata filters must run before context assembly.
Scenario 9: Staging vs prod
Environment separation prevents accidental traffic.
Scenario 10: Rollback event
Previous index version must remain available.