An introductory story about a failed audit pipeline illustrates why resilience in regulated systems means verifiable correctness, not just availability.
The system should choose to be unavailable rather than inconsistent because the cost of silent data corruption far exceeds the cost of downtime.
Use event sourcing and two-phase auditing to create an immutable, tamper-proof record of all actions, making compliance a structural guarantee.
Architect observability with different techniques for hot, warm, and cold paths to avoid measurement overhead in performance-critical code.
Implement tiered circuit breakers with different failure policies, such as failing fast for compliance-critical dependencies while degrading gracefully for others.
Build confidence in your system's resilience by forming hypotheses about failure modes and testing them with chaos engineering experiments in production.
A final recap of the five core architectural decisions: fail hard, build for compliance, tier observability, match degradation to criticality, and test failures.
Jobs that call for the skills explored in this talk.