Root cause analysis

Root Cause Analysis

You are performing systematic root cause analysis to find the true source of a bug. Do not apply fixes until you understand WHY the bug exists.

Core Principle

Never fix a symptom. Always find and fix the root cause.

The Five Whys Method

Ask "Why?" repeatedly to drill down to the root cause:

1. Why did the API return an error? → The database query failed
2. Why did the database query fail? → The connection pool was exhausted
3. Why was the pool exhausted? → ROOT CAUSE: Missing finally block to close connections

Investigation Phases

Phase 1: Reproduce the Bug

Before investigating:

1. Reproduce consistently - If you can't reproduce it, you can't verify a fix
2. Document reproduction steps - Exact sequence of actions
3. Note environment details - OS, versions, configuration
4. Identify minimal reproduction - Smallest case that shows the bug

Questions to answer:

• Does it happen every time or intermittently?
• Does it happen in all environments?
• When did it start happening? (recent changes)

Phase 2: Gather Evidence

Collect information before forming theories:

• Error messages and stack traces
• Log files (application, system, database)
• Recent code changes (git log, blame)
• User reports and reproduction steps
• Monitoring data (metrics, APM)
• Related issues (search issue tracker)

Do NOT:

• Make changes while gathering evidence
• Assume you know the cause without evidence
• Ignore related symptoms

Phase 3: Form Hypotheses

Based on evidence, create ranked hypotheses:

Priority	Hypothesis	Evidence	Test Plan
1	Connection leak in UserService	Stack trace shows connection pool	Add logging, check usage
2	Query timeout too short	Occurs under load	Test with longer timeout
3	Database server overload	Correlates with peak hours	Check DB metrics

For each hypothesis:

• What evidence supports it?
• What evidence contradicts it?
• How can we test it?

Phase 4: Test Hypotheses

Test each hypothesis systematically:

1. Start with highest probability
2. Design a definitive test - Should clearly confirm or reject
3. Make ONE change at a time
4. Document results

If hypothesis is rejected:

• Cross it off the list
• Re-evaluate remaining hypotheses
• Consider if new evidence suggests new hypotheses

Phase 5: Verify Root Cause

Before declaring root cause found:

• Can you explain the full causal chain?
• Does fixing it consistently prevent the bug?
• Does it explain ALL observed symptoms?
• Is there nothing earlier in the chain that could be fixed?

Common Root Cause Categories

• Code Defects: logic errors, boundary conditions, race conditions, resource leaks, null/undefined handling
• Design Issues: missing error handling, inadequate validation, poor state management, coupling
• Environment: configuration errors, resource constraints, version mismatches, network issues
• Data Issues: invalid input, data corruption, schema mismatches, encoding problems

Evidence Collection Commands

# Recent changes to relevant files
git log --oneline -20 -- path/to/file

# Who changed this line
git blame path/to/file

# Changes since last working version
git diff v1.2.3..HEAD -- src/

# Search for related error handling
grep -r "catch\|error\|throw" --include="*.ts" src/

Red Flags - You Haven't Found Root Cause

• "I'm not sure why, but this fix works"
• "The bug went away after I restarted"
• "I added a check to prevent this case"
• "It's probably a race condition somewhere"

These suggest symptom treatment, not root cause resolution.

Documentation Template

When root cause is found, document:

## Bug: [Description]

### Root Cause
[Clear explanation of why the bug occurred]

### Evidence
- [Evidence 1]
- [Evidence 2]

### Causal Chain
1. [Initial trigger]
2. [Intermediate cause]
3. [Root cause]
4. [Observed symptom]

### Fix
[Description of the fix and why it addresses root cause]

### Prevention
[How to prevent similar issues in the future]

Integration with Other Skills

After finding root cause:

• Use testing/red-green-refactor to write a test that exposes the bug
• Use planning/verification-gates to validate the fix
• Consider collaboration/structured-review for complex fixes