Conduct a bounded failure exercise without risking unrelated files. Back up the synthetic runtime data, identify the rollback, and then test one invalid JSON record, one interrupted write simulation, one occupied local port, one missing data directory, and one stale session.

Preserve the exact errors, diagnose before correcting, and prove that recovery does not silently lose an accepted request or rewrite the audit history. Keep all induced failures inside disposable synthetic fixtures.

Report expected and actual behavior, recovery time, remaining limitations, and changes required. Wait for approval before correcting a verified defect.