PromptEvalCalibration

A calibration of an evaluation suite's judge (1.63.0): its sample of a run's verdicts, how many are labelled, the judge compared (judge, the suite's when the calibration was created), and the agreement per rubric kind against the target. Ids, keys, counts and rates only; the labels and the transcripts are read on their own routes.

An object with 20 fields.

Fields

Field	Type	Required	Description
`agreement`	PromptEvalAgreement or null	yes	Null until `completed`.
`cost_lines`	array of PromptEvalCostLine	yes
`created_at`	Timestamp	yes
`created_by`	ActorRef	yes
`error_code`	string or null	yes
`finished_at`	Timestamp or null	yes
`from_calibration_id`	PromptEvalCalibrationId or null	yes	The calibration whose labelled sample this one reuses; null when it drew its own.
`id`	PromptEvalCalibrationId	yes
`job_id`	JobId or null	yes	The `prompt.calibration` job that compares; null while `labelling`.
`judge`	StageRef	yes
`labelled`	integer	yes	The entries that carry a label.
`meets_target`	boolean or null	yes	Whether the agreement met the target (`calibrationMeetsTarget`); null until `completed`.
`object`	`"prompt_eval_calibration"`	yes
`run_id`	PromptEvalRunId	yes	The run the sample was drawn from.
`sample`	array of PromptEvalCalibrationEntry	yes
`started_at`	Timestamp or null	yes
`status`	PromptEvalCalibrationStatus	yes
`suite_id`	PromptEvalSuiteId	yes
`suite_revision`	integer	yes
`target`	PromptEvalCalibrationTarget	yes

Used by

The operations whose parameters, request body or responses name this shape directly.