diagnose

To debug hard bugs systematically, convert the problem into a fast, deterministic feedback loop with a pass/fail signal. Generate 3-5 falsifiable ranked hypotheses, instrument one variable at a time, then fix and add a regression test.

Hypothesis testing in software debugging is proposing falsifiable, ranked predictions about a bug's root cause. You confirm or falsify each prediction by instrumenting one variable at a time within a deterministic, reproducible feedback loop.

Build a reproducible feedback loop by creating an agent-runnable signal using a test, HTTP script, CLI fixture, Playwright assertion, or trace replay. This deterministic pass/fail signal drives all subsequent hypothesis testing and instrumentation.

Diagnose a performance regression by prioritizing measurement and baselining before making changes. Use profiling-friendly instrumentation and bisection rather than broad logging to isolate the root cause within a reproducible loop.

Cleaning up instrumentation after debugging removes temporary logging or profiling code added during the investigation. You keep only the regression test that locks in the fix, ensuring the codebase remains clean and maintainable.

Yes, this debugging workflow works for web services and UI applications by supporting diverse reproducible signals. You can use HTTP scripts for services or Playwright assertions for UI actions to establish the deterministic pass/fail signal.