Validation

Ask review

Verify data integrity, functional parity, and performance

Hats
2
Review Agents
1 +1
Review
Ask
Unit Types
Validation
Inputs
Migrate

Dependencies

Migratemigration-artifacts

Hat Sequence

1

Regression Tester

Focus: Confirm that downstream consumers and application logic produce identical results when reading from the migrated target instead of the original source. Run existing test suites, replay production queries, and compare outputs. Surface any behavioral difference, no matter how small.

Produces: Regression test results with before/after comparisons for downstream consumers and critical query paths.

Reads: Validation report from the validator, downstream consumer contracts, existing test suites.

Anti-patterns (RFC 2119):

  • The agent MUST NOT only test the data layer without exercising application logic on top of it
  • The agent MUST NOT ignore performance regressions (correct but 10x slower is still a regression)
  • The agent MUST NOT assume that passing unit tests means the integration is correct
  • The agent MUST replay real query patterns from production logs
  • The agent MUST NOT treat "no errors in logs" as equivalent to "functionally correct"
2

Validator

Focus: Perform quantitative verification that the migrated data matches the source. Reconcile row counts, compute checksums, and run spot-check comparisons on randomly sampled records. Verify that constraints, indexes, and referential integrity hold in the target. The goal is proof, not confidence.

Produces: Validation report with reconciliation results, sample diffs, constraint verification, and discrepancy analysis.

Reads: Migration artifacts, source data for comparison, mapping specification for expected transformations.

Anti-patterns (RFC 2119):

  • The agent MUST NOT declare validation complete after checking only row counts
  • The agent MUST NOT sampl records non-randomly (e.g., only the first 100 rows)
  • The agent MUST NOT ignore records that were intentionally dropped or transformed — they still need accounting
  • The agent MUST NOT treat zero errors as proof of correctness without verifying test coverage
  • The agent MUST NOT validate against the mapping spec but not against actual source data

Review Agents

Parity

Mandate: The agent MUST verify the migrated system achieves functional parity with the source.

Check:

  • The agent MUST verify that all critical user workflows produce the same results in the target system
  • The agent MUST verify that performance benchmarks meet or exceed the source system
  • The agent MUST verify that edge cases identified in the assessment are explicitly tested
  • The agent MUST verify that no regression in features that were not part of the migration scope

Included from other stages

Accuracyfrom Mapping

Validation

Criteria Guidance

Good criteria examples:

  • "Row-count reconciliation shows zero discrepancy between source and target for every entity"
  • "Spot-check validation compares at least 100 randomly sampled records per entity with field-level diff"
  • "Performance benchmarks show target query latency within 10% of source for critical paths"

Bad criteria examples:

  • "Data looks correct"
  • "Validation is complete"
  • "Performance is acceptable"

Completion Signal (RFC 2119)

Validation report MUST exist with quantitative reconciliation results (row counts, checksum comparisons, sample diffs). Functional parity tests confirm that downstream consumers produce identical results against the migrated data. Performance benchmarks are within defined thresholds.