rca-annotator

You can label root cause analysis outputs by interactively reviewing agent diagnoses, confirming categories, validating evidence traceability, scoring difficulty, and capturing alternative hypotheses to produce structured annotation files.

Evidence traceability in root cause analysis annotation enforces source_file, json_path, exact_value, and line references for each item, ensuring every diagnosis label is auditable and grounded in the original logs.

You can annotate RCA results on a remote MLflow artifact store by using the jumpbox synchronization feature, which downloads and uploads analysis artifacts and annotations via SSH, rsync, or scp.

Validating automated root cause analysis diagnoses requires completed jobs containing step1_job_context.json, step3_correlation.json, step4_github_fetch_history.json, and step5_summary.json in the local .analysis directory.

Yes, you can use the annotation workflow without SSH access by reading artifacts directly from a local .analysis/<job_id> directory, as jumpbox synchronization is an optional feature for remote environments.

A key limitation is that annotation depends entirely on the presence of completed RCA job artifacts; without the required step1 through step5 JSON files, the interactive labeling and evidence validation workflow cannot proceed.