Playbook: Analysis Failures

Analysis Failures

When to use: Customer reports “analysis not completing”, “stuck in progress”, or AI quality issues.

Diagnosis Steps

1. Check AI Analytics

MCP: get_ai_analytics
  workspace_id: "<workspace_id>"

Look for elevated error rates, increased latency, or low completion rates.

2. Check Analysis Engine Health

MCP: remote_debug
  workspace_id: "<workspace_id>"
  command: "analysis_status"

3. Review Recent Analyses

MCP: remote_debug
  workspace_id: "<workspace_id>"
  command: "recent_analyses"
  params: { "limit": 10 }

Look for patterns — all failing, specific incident types failing, intermittent failures.

Common Causes

Symptom	Likely Cause	Fix
All analyses failing with timeout	LLM provider degraded	Check provider status page, wait or switch provider
Analyses stuck “in_progress”	Worker process crashed	Restart analysis engine pods
Partial results, low quality	Context window exceeded	Check incident group size, may need log filtering
Rate limit errors	Too many concurrent analyses	Check concurrency settings, consider queue backpressure

LLM Provider Issues

If the LLM provider is degraded:

1. Check the provider status page
2. Review error codes in AI analytics (429 = rate limit, 503 = provider down)
3. If persistent, consider temporarily increasing timeout or switching provider

Escalation

Collect AI analytics output + recent analysis IDs and escalate to the engineering team.