qa exploration
How I think through an issue once it's reported: a symptom feeds forward through the categories most likely to explain it, then through the tools I use to confirm which one it actually is.
Nodes are stages of investigation, not literal weighted connections — read left to right.
→ Sentry payload trace
Usually a sign the request itself was malformed, unauthenticated, or rate-limited before it reached business logic. I check auth tokens, payload shape, and headers first, using Sentry to see exactly what the client sent.
→ Datadog service trace
These point further upstream — a crashed process, an unhealthy dependency, or a timeout somewhere in the chain. I use Datadog traces and logs to isolate which service actually failed, rather than assuming it's the one that returned the error.
→ Manual & automated repro
Harder to catch, because the request often succeeds and the response is just wrong. These need careful reproduction — same data state, same sequence of actions, same timing — until the bug shows itself reliably.
→ Root cause confirmed / regression coverage added
How I actually catch the categories above before release: manual exploratory sessions on new features, a Playwright regression suite for known-critical paths, and Claude to help draft and maintain test cases as coverage grows.