Data migration

Change persisted data while keeping every supported reader and writer correct. A completed job is not proof of a completed migration. Code-only transformations belong to flow-migrate.

Phase 1: Establish the data contract

• Inspect the actual engine and version, schema, migration mechanism, volume, indexes, constraints and all readers and writers, including older deployments and background jobs. Do not choose engine-specific syntax or promise a completion time from assumed capabilities or throughput. Measure a representative batch before estimating.
• Define the source of truth, transformation version, invariants, invalid-row policy and acceptance queries. Preserve identifiers, precision and null semantics. A placeholder value cannot make missing or invalid data correct merely by satisfying a constraint.
• Establish the authorized environment and terminal action. Preparing a migration does not authorize running it on live data. Reuse authorization already given.

Phase 2: Make compatibility and recovery explicit