root-cause-trace

Root Cause Tracing

ultrathink — use extended reasoning for root cause analysis.

Process

1. Observe — Gather All Evidence

• Read the full error message and stack trace
• Check logs at multiple levels (application, database, message broker)
• Note the exact error, not a paraphrase
• Determine: is this error the root cause, or a symptom?

2. Hypothesize — Form Multiple Theories

• Generate at least 3 possible root causes
• Rank by probability based on the evidence
• Consider: timing, data flow, state mutations, race conditions

3. Trace Backwards

• Start from the error location
• Walk the call chain backwards through the code
• At each step, verify: is the data correct here? Where did it come from?

4. Isolate — Binary Search

• If the trace is long, check the midpoint
• Is the data correct at the midpoint? → Problem is downstream
• Is the data wrong at the midpoint? → Problem is upstream
• Narrow until you find the exact point where data goes wrong

5. Verify — Prove the Root Cause

• Can you reproduce the error with specific input?
• Does fixing the identified root cause prevent the error?
• Are there other code paths that could trigger the same issue?

Anti-Patterns

• ❌ Don't guess — trace with evidence
• ❌ Don't fix symptoms — find the root cause
• ❌ Don't assume "it works on my machine" — check environment differences
• ❌ Don't add try/catch to silence errors — fix the cause