eval.run.completed
⚠️
Preview — this feature is being built; the design below is the spec.
Sent when: An evaluation run finished, failed or was canceled.
| Property | Value |
|---|---|
| Scope | One workspace. Register with an inference key at /v1/webhooks. |
| Area | Evaluations |
| Subscribe with | "events": ["eval.run.completed"], or a prefix such as "eval.*" |
Payload
{
"id": "evt_01JCA5D7F9H1K3M5P7R9",
"type": "eval.run.completed",
"created_at": "2026-11-05T10:00:00Z",
"workspace_id": "wrkspc_01JC0A2C4E6G8J0L2N4Q",
"data": {
"object": {
"id": "evrun_01JD35D7F9H1K3M5P7R9",
"type": "eval_run",
"status": "succeeded",
"summary": [
{
"label": "baseline",
"pass_rate": 0.82
},
{
"label": "short",
"pass_rate": 0.86
}
]
}
}
}data.object matches what the corresponding GET endpoint returned when the event happened. Prompt and output text are never included.
What to do
- Post the summary to your team channel, or fail a CI job when the candidate’s pass rate drops below the baseline.
Delivery
Deliveries are signed, retried with back-off for up to 3 days, and may arrive more than once or out of order. De-duplicate on the syntic-event-id header and compare timestamps before acting. See Webhooks.
Was this page helpful?