When testing systems that depend on third-party APIs (payment processors, identity providers, SMS gateways), provider-maintained sandboxes are unreliable regression baselines because they drift from production without notice. Key failure modes include sandbox drift, poor error path coverage, inconsistent state management, and behavioral changes that don't alter response schemas. Reliable teams record real API interactions as version-controlled fixtures and replay them in CI, test failure conditions explicitly with dedicated fixtures, and treat contract testing as complementary to—not a substitute for—behavioral regression coverage. Fixture maintenance requires explicit ownership to avoid recreating the same drift problem you were trying to solve.
