Failure, reconstructed.
Synthetic production incidents, investigated without hindsight. Each file follows the signals from the first alert to a verified causal chain—or the point where the evidence runs out.
Lead investigation · 8 min readpool cap reduced→connection queue grows→payment latency→checkout p99 alert
A routine payment-service release changes one connection-pool default. Checkout owns the alert; a dependency two hops away owns the failure.
deadline mismatch→unbounded retries→database saturation→order latency
gateway errors→telemetry gap→competing hypotheses→no verified RCA
These are not customer case studies and they are not Tiravan performance benchmarks. We author plausible services, alerts, metrics, spans, logs, and deploy events to examine one question: what can the available evidence actually support?
Every file distinguishes the initiating trigger from downstream effects, records rejected hypotheses, and stops when a conclusion would require invented evidence.
Follow the evidence here. Bring your own telemetry when you are ready to evaluate Tiravan.