Resilience audit

A happy-path pass says nothing about failure behavior. Force each material failure and observe the invariant directly.

1. Map failure boundaries

Trace critical operations across process, network, storage, queue, cache, filesystem, browser, and third-party boundaries. For each, name the owner, timeout, cancellation path, retry policy, idempotency key, cleanup duty, durable state, and user-visible outcome.

Complete when: each critical path has explicit failure and recovery invariants.

2. Build a pressure matrix

Select applicable faults: