Degrade or fail fast? The answer depends on whether the error is reversible

A software engineering analysis examines how distributed systems with multiple interdependent services should handle partial failures. The core principle proposed is that services should degrade gracefully only when a wrong guess is reversible, such as serving a stale product catalog, while services must fail fast when errors can cause irreversible harm, like overselling inventory. Using a six-service e-commerce pipeline as a case study, the author argues that a single universal failure policy applied across all services is fundamentally flawed. For example, inventory checks must refuse requests outright during downtime to prevent overselling, whereas a settlement service can safely queue events and catch up later without affecting the buyer experience. The framework centers not on service importance but on whether the business can recover from acting on incorrect or missing data.