Detections
Runless automatically analyzes your workflow runs and surfaces actionable insights called Detections. These are issues that, if fixed, will improve your CI reliability, speed, or cost.
How Detections Work
Detections are generated by analyzing patterns in your workflow run data:
- We aggregate metrics per workflow (success rate, re-run rate, duration, etc.)
- We compare against thresholds and historical baselines
- We generate findings when metrics exceed thresholds
- Detections update automatically as new runs come in
Detections require at least 10 runs in the last 30 days to generate meaningful insights.
Detection Categories
Reliability
Issues affecting the consistency of your CI:
High Re-run Rate
- Trigger: Re-run rate > 20%
- Severity: High if > 30%, Medium otherwise
- Cause: Usually flaky tests or unreliable external dependencies
- Fix: Identify the specific failing tests, add retries for external calls, or improve test isolation
High Failure Rate
- Trigger: Failure rate > 15%
- Severity: High if > 25%, Medium otherwise
- Cause: Broken tests, misconfigured workflows, or legitimate frequent failures
- Fix: Review failed runs to identify patterns, fix broken tests
Performance
Issues affecting CI speed:
Duration Regression
- Trigger: p50 duration increased > 20% vs previous period
- Severity: Medium
- Cause: New slow tests, cache invalidation, dependency changes
- Fix: Profile the workflow, identify slow steps, optimize or parallelize
High p95/p50 Ratio
- Trigger: p95 is > 3x p50
- Severity: Low
- Cause: Inconsistent infrastructure, conditional slow paths, resource contention
- Fix: Investigate slow runs, look for patterns in timing or inputs
Cost
Issues affecting CI spend:
Expensive Workflow
- Trigger: Workflow in top 10% by total duration
- Severity: Medium
- Cause: Long-running jobs, inefficient parallelization
- Fix: Optimize slow steps, improve caching, consider splitting the workflow
Severity Levels
| Level | Meaning | Action |
|---|---|---|
| High | Significant impact, fix soon | Investigate this week |
| Medium | Notable impact, worth fixing | Add to backlog |
| Low | Minor impact or informational | Fix when convenient |
Viewing Detections
On the Dashboard
The Findings Preview shows the top 5 detections. Click any finding to expand details.
On the Detections Page
The full /detections page shows all findings with:
- Category filtering (sidebar)
- Severity filtering
- Affected repository/workflow
- Estimated impact
Detection Lifecycle
Detections are recalculated periodically:
- When new workflow runs complete
- At least once per hour
If you fix the underlying issue, the detection will automatically resolve once enough new runs show improvement.
Future Detections
We're working on additional detection types:
- Cache miss analysis
- Unpinned action versions
- Redundant job detection
- Optimal parallelization suggestions
Have ideas for detections? Let us know.