Use Migration Strategy Prover with your AI.

Validate migration strategy

You must: (1) INVENTORY RISKS. map every dependency, blast radius, and failure scenario. Include data volume, transformation complexity, timeline with buffer. "Straightforward migration" is never an assessment, (2) DEFINE ROLLBACK. specify trigger metrics (error rate >X%, latency >Yms, data lag >Z min), step-by-step procedure, data reconciliation plan, and whether rollback is reversible. "Just switch back" is not rollback, (3) PROVE DATA INTEGRITY. row counts, checksums, referential integrity, edge cases (NULLs, Unicode, timezones, precision). If you did not verify it, it is not verified, (4) PLAN CUTOVER. incremental traffic shifting with monitoring at each stage. Strangler Fig / Blue-Green / Canary / Big-Bang with go/no-go criteria. "Over the weekend" is not a strategy, (5) ALIGN STAKEHOLDERS. upstream producers, downstream consumers, on-call teams, customer notification, shared runbook, escalation path. "Everyone knows" is never alignment. If rejected, fix the specific migration gap before executing. Structured reflection capability for system migration strategy. forces comprehensive risk inventory, trigger-based rollback planning, data integrity verification, incremental cutover design, and stakeholder alignment before any migration executes. Catches Risk Unassessed (starting a migration without mapping blast radius. a team migrated a PostgreSQL 12 database to PostgreSQL 16 on a Saturday. They tested the migration on a staging database with 50,000 rows. Production had 340 million rows. The migration took 47 hours instead of the estimated 4. During migration, 3 downstream services that polled the database every 30 seconds began queueing requests. by hour 8, the message queue had 2.4 million unprocessed events. The blast radius included: the billing system (invoices frozen), the notification service (800,000 queued emails), and the analytics pipeline (48-hour data gap). "No major risks" is never an assessment. every migration has dependencies you have not found yet), Rollback Undefined (having no rollback plan or defining "just switch back". a monolith-to-microservices migration cut over 3 services on Monday. By Wednesday, the new payment service had processed $2.3M in transactions. A critical bug was discovered Thursday. "Roll back" meant: reconcile $2.3M in transactions written to the new database schema that the old schema cannot read. 47 hours of manual reconciliation by 6 engineers. Rollback must define: trigger metrics (error rate >X%, latency >Yms), exact procedure (not "switch back"), data reconciliation for the transition period, and whether rollback is even possible. some migrations are one-way), Data Integrity Unproven (migrating data without verification beyond "it looks right". a cloud migration moved 89 million user records. Post-migration count matched. Two weeks later: 340,000 records had truncated Unicode characters in names (Turkish İ → I, German ß → ss, Japanese characters → ?). Row count verification alone is insufficient. Checksums catch bit-level corruption. Referential integrity checks catch broken FK relationships. Edge case testing catches encoding, timezone, NULL handling, and precision loss. If you did not verify it, it is not verified), Cutover Missing (planning a "big bang over the weekend" instead of incremental traffic shifting. big bang migrations have a binary outcome: success or total failure with no middle ground. A Strangler Fig pattern routes 1% → 5% → 25% → 50% → 100% of traffic, with monitoring and go/no-go criteria at each stage. Blue-Green gives instant rollback by switching load balancer targets. Canary catches issues at 1% traffic before they affect 100% of users. The cutover strategy IS the risk mitigation), and Stakeholders Unaligned (migrating without telling downstream teams. a database migration changed column types from VARCHAR(255) to TEXT. The analytics team's ETL pipeline had VARCHAR(255) constraints hardcoded. Pipeline broke silently. 3 days of missing analytics data before anyone noticed. Every migration has upstream producers and downstream consumers. Each must be identified by name, notified with timeline, and given a shared runbook). Call once per migration plan before execution