← Back to changelog

October 22, 2026

Export score metadata and provenance to blob storage

Niklas Semmler, PhD

Two opt-in field groups add score metadata and the annotation, config, and evaluator behind each score to scheduled S3, GCS, and Azure exports.

Score files in scheduled blob storage exports can now include each score's metadata and where it came from. Two new field groups add these columns on top of the core score columns:

• Join scores to their origin in your warehouse. score_provenance adds author_user_id, config_id, queue_id, evaluator_id, evaluation_rule_id, and execution_trace_id, so you can break down annotation scores by reviewer and queue, or LLM-as-a-judge scores by evaluator.
• Debug a judge's verdict. Follow execution_trace_id to the trace of the evaluator run that produced the score.
• Analyze custom score attributes. score_metadata adds the metadata column with the key-value pairs you attached when creating the score.

Enable the groups under Project Settings → Integrations → Blob Storage, or add score_metadata and score_provenance to exportFieldGroups on PUT /api/public/integrations/blob-storage. Both groups are off by default, including on new integrations, so existing score files keep their current columns. Group columns are appended after the core columns and apply to future exports only.

Learn more