Skip to content

Create Evaluation

POST /evaluations

Always 501 today (open question G-4). The body is validated first, on purpose: a client learns the feature is missing rather than that its payload was wrong.

Request body — `EvaluationCreate`

Field	Type		Notes
`agent_id`	string (uuid)	required	—
`agent_revision_id`	string (uuid) \| null	optional	—
`dataset_name`	string \| null	optional	max length 200
`evaluators`	object[]	optional	—

Responses

Status	Description	Body
`202`	Successful Response	`EvaluationRunView`
`422`	Validation Error	`HTTPValidationError`
`501`	Always, today. No annex defines evaluators, datasets, scoring or how a result relates to a run, so there is nothing to run — see open question G-4 in docs/OPEN-QUESTIONS.md. The tables and the two read endpoints exist; the write path is deliberately absent rather than improvised. Branch on the error code `not_implemented`; do not retry.	—

Example

The body shows the fields the schema always carries. Which combination is legal is a service rule the schema cannot express, so read the description above before copying this.

curl -X POST https://your-deployment.example.com/evaluations \
  -H "Authorization: Bearer $COBRE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
       "agent_id": "0190a2f4-0000-7000-8000-000000000000"
     }'

Response — `EvaluationRunView`

Field	Type		Notes
`id`	string (uuid)	required	—
`agent_id`	string (uuid)	required	—
`agent_revision_id`	string (uuid) \| null	optional	—
`dataset_name`	string \| null	optional	—
`evaluators`	object[]	optional	—
`status`	`queued` \| `running` \| `succeeded` \| `failed` \| `cancelled`	required	—
`summary`	object	optional	—
`created_at`	string (date-time)	required	—
`finished_at`	string (date-time) \| null	optional	—