debugging

Root cause analysis for failing unit tests follows a 5-phase protocol: evidence gathering, pattern recognition, hypothesis testing, implementation, and verification. This workflow converts failure uncertainty into testable claims so you apply the smallest regression-backed fix that eliminates the entire defect class.

Isolating a deterministic crash from a stack trace uses zoom-out/zoom-in isolation and hierarchical repair-surface selection. This method patches the highest architectural layer capable of removing the defect class, ensuring the fix addresses the actual fault rather than localized symptoms.

Errors recurring across turns happen when repeated error classes suggest an approach-level issue rather than a localized fault. Conversation state tracking distinguishes independent events to avoid false causal inferences, triggering escalation and plateau stopping when an autonomous self-repair loop stalls.

Automating self-repair for compiler and lint errors requires a bounded run-observe-patch-rerun loop with explicit contracts. Plateau detection stops the autonomous retries when repeated error classes appear, escalating the issue instead of looping infinitely on approach-level problems.

Regression testing documentation is needed immediately after post-fix verification confirms the smallest patch removes the defect class. Documenting the regression prevents the same error class from recurring in future iterations by locking the verified fix behind a test.

Yes, this error isolation workflow applies to any platform with available error evidence such as stack traces, failing tests, lint/compiler errors, or deterministic crashes. It requires no specific dependencies or external components to execute the structured 5-phase protocol.