Schema drift

Every supplier changes. Drift shows up as a new enum value, a field changing type, a suddenly mandatory object, or a response that arrives in a different order. Left unchecked, it becomes customer‑visible errors or silent data loss.

Treat mappings as code with observability. Watch live traffic against a contract of expected shapes; when a deviation appears, log it with samples. An agent proposes a mapping delta that absorbs the change by adjusting transforms or introducing a field with a safe default.

Prove the proposal offline against cached traffic. The goal is confidence: failures turn green, healthy cases stay green and performance does not regress. Only then promote the mapping as a new version with canary percentage and an instant rollback switch.

Record who approved what, when and why. Version numbers should be visible in dashboards and correlatable to error rates. If a regression appears after promotion, roll back automatically and open an investigation with the captured samples.

The user‑facing API should remain stable during all of this. That separation — supplier mapping vs public schema — is what prevents a supplier release from becoming your breaking change.