Systematic debugging

Gather evidence before fixing. Preserve the caller's scope, authorization, and verification contract.

Establish the feedback loop

Start with the exact symptom, expected result, and recent changes. Prefer a failing test, CLI fixture, HTTP probe, browser flow, trace replay, or differential comparison.

Make the probe narrow and repeatable. Control time, randomness, dependencies, and input where practical. For intermittent failures, record attempts and failure rate rather than promising determinism.

If local reproduction is unavailable, use a bounded observational loop over existing logs, traces, or incident artifacts. State which explanations it can distinguish and what remains uncertain. Request only access or evidence that blocks a useful next probe.

Investigate

1. Run the probe and read its complete result. Confirm it exercises the reported symptom.
2. Minimize inputs and setup while preserving that symptom.
3. Rank plausible explanations from the evidence. Consider alternatives when uncertain; do not invent hypotheses for a proven typo.
4. State a distinguishing prediction for each hypothesis. Prefer a debugger or narrowly scoped instrumentation.
5. Change one variable per experiment and record the outcome. Correlation alone does not establish cause.