Vector Database Index Migration Checklist Builder

You are a senior AI infrastructure engineer helping a team migrate a vector database index safely. Build a migration checklist for:

Current vector store: [Pinecone, Weaviate, Qdrant, Milvus, Chroma, Elasticsearch, pgvector, custom]
Target change: [new index, namespace split, metadata schema change, embedding model upgrade, dimension change, provider migration, shard redesign]
Application use case: [RAG search, recommendations, semantic support search, entity matching, image search]
Embedding model details: [current model, target model, dimensions, normalization, chunking strategy]
Data volume: [documents, vectors, metadata size, update rate, tenant count]
Retrieval path: [API, background jobs, hybrid search, filters, reranker, cache, streaming answer pipeline]
Current quality baseline: [golden queries, recall metrics, click data, human evals, support tickets]
Migration constraints: [zero downtime, tenant isolation, cost limit, compliance, data residency, SLA]
Write path: [batch ingestion, event stream, dual writes, retries, idempotency, delete/update semantics]
Rollback options: [old index retained, traffic switch, feature flag, backup export, rehydration plan]
Monitoring tools: [logs, traces, vector DB metrics, eval dashboard, alerts]

Create:
1. Pre-migration inventory for index settings, schema, embeddings, filters, tenants, and dependencies.
2. Compatibility checklist for dimensions, distance metric, normalization, metadata filters, and chunk IDs.
3. Backfill plan with batching, idempotency, rate limits, cost controls, and verification.
4. Dual-write or shadow-read strategy for zero-downtime migration.
5. Retrieval quality test plan with golden queries, recall, precision, latency, and answer-grounding checks.
6. Data consistency checks for counts, deletes, updates, duplicate chunks, and metadata coverage.
7. Traffic cutover plan with feature flags, canary cohorts, and owner approvals.
8. Rollback plan and conditions that trigger rollback.
9. Monitoring dashboard and alert thresholds for latency, errors, empty results, cost, and quality drift.
10. Post-migration cleanup plan for old indexes, code paths, credentials, and documentation.

Make the checklist implementation-ready for engineers. Call out assumptions and risks instead of treating vector search as a black box.