AI Analysis¶
The AI Analysis tab shows results from automated analysis of workflow runs. It appears on the run detail page when an auto-analysis job has been submitted.
When Analysis Runs¶
Tracker automatically submits analysis jobs when runs complete:
| Run Outcome | Analysis Type | Description |
|---|---|---|
| Failed | Failure troubleshooting | Root cause analysis with evidence and recommendations |
| Succeeded | QC summary | Quality control metric review, highlighting outliers |
Exclusions¶
ci-labeled runs: excluded from all analysisregression-test-labeled runs: excluded from QC summary- Succeeded runs without MultiQC outputs: excluded from QC summary
Analysis Content¶
Failure Troubleshooting¶
Produces a report with: summary, root cause, evidence (exit codes, log excerpts, resource metrics), and recommendations.
Submit Runs via Git
If the run was submitted using a workflow in a git repository, the analysis will clone the repo at the exact commit and examine the code to explain failures. This is a good reason to always use git-based submission, even for workflows in active development.
QC Summary¶
Reviews MultiQC metrics and highlights values outside expected ranges, flagging samples that may need attention.
Data Sources¶
- Tracker database -- Run metadata, task statuses, resource usage via MCP
- Datadog -- CPU, memory, disk, network metrics and container logs via MCP
- S3 work directories -- Task execution files (
.command.sh,.command.err,.command.out) - Workflow source code -- For failure analysis, the git repo is cloned (GitLab repos) so the analysis can reference process definitions