srstomp/systematic-debugging

Systematic Debugging

Fix root causes, not symptoms. Do not edit code until you have reproduced the issue or gathered enough evidence to explain why reproduction is impossible.

Process

1. Reproduce

• Capture the exact command, input, or user steps.
• Read the full error message and stack trace.
• Confirm whether the issue is deterministic or intermittent.
• Record expected behavior versus actual behavior.

If the issue cannot be reproduced, gather logs or traces and report the missing data instead of guessing.

2. Trace Root Cause

• Check recent changes that touch the failing path.
• Compare broken code with a nearby working example.
• Trace bad data backward from the failure site to the source.
• In multi-component systems, add temporary diagnostics at boundaries to find which layer changes the data or state.

Write the hypothesis as: "I think X is the root cause because Y evidence."

3. Test One Hypothesis

• Change one variable at a time.
• Prefer the smallest diagnostic or failing test that proves the hypothesis.
• If the hypothesis fails, remove incidental diagnostic changes and form a new hypothesis from the new evidence.
• After three failed fix attempts, stop and question whether the architecture or task framing is wrong.

4. Fix And Verify

• Add a regression test before the fix whenever feasible.
• Make the minimal code change that addresses the root cause.
• Run the regression test and relevant existing tests.
• Use verification-before-completion before reporting success.

Anti-Patterns

• Patching the line where the error surfaced without tracing the source.
• Changing tests to match broken behavior.
• Bundling cleanup or refactors into a bug fix.
• Trying multiple fixes at once.
• Declaring "fixed" after code changes without rerunning the failing scenario.

Report Template

## Debugging Report

### Reproduction
- Command/steps:
- Expected:
- Actual:

### Root Cause
- Hypothesis:
- Evidence:

### Fix
- Files:
- Why this is minimal:

### Verification
- Command:
- Result: