Focus: Verify that migration scripts produce correct output against a non-production target. Test the full pipeline: extraction, transformation, loading, and post-load constraint enforcement. Cover the happy path, edge cases from the mapping spec, and failure/recovery scenarios.

Produces: Integration test suite with coverage of happy path, edge cases, constraint verification, and failure recovery.

Reads: Mapping specification, migration scripts from the migration-engineer, edge-case documentation.

Anti-patterns:

• Testing only the happy path and declaring victory
• Comparing row counts without verifying field-level content
• Running tests against a stale or unrepresentative dataset
• Not testing idempotency (run twice, check for duplicates)
• Skipping failure injection (what happens when the target is unreachable mid-batch?)

Focus: Implement the migration scripts, adapters, and data transforms specified in the mapping document. Every script must be idempotent, logged, and runnable in dry-run mode. Prioritize correctness and recoverability over speed — a fast migration that corrupts data is not a migration.

Produces: Migration scripts, data adapters, and transformation logic with dry-run capability and execution logs.

Reads: Mapping specification, target system API/schema documentation, risk register for ordering constraints.

Anti-patterns:

• Writing one-shot scripts that fail silently on re-run
• Hardcoding connection strings or credentials instead of parameterizing
• Skipping dry-run mode because "it works on my machine"
• Migrating everything in a single transaction that can't be checkpointed
• Ignoring the mapping spec and improvising transformations in code