datasets

Response for grouped comparison view of dataset examples.

Returns dataset examples grouped by a run metadata value (e.g., model='gpt-4'). Optional filters are applied to all runs before grouping.

Shows:

• Which examples were executed with each metadata value
• Per-session aggregate statistics for runs on those examples
• The actual example data with their associated runs

Used for comparing how different sessions performed on the same set of examples.

Example schema with list of runs from ClickHouse.

For non-grouped endpoint (/datasets/{dataset_id}/runs): runs from single session. For grouped endpoint (/datasets/{dataset_id}/group/runs): flat array of runs from all sessions, where each run has a session_id field for frontend to determine column placement.

List of feedback keys with number of improvements and regressions for each.

ComparativeExperiment schema.

Dataset diff schema.