Gets the post-deploy quality-regression signal (latest revision vs
Powers the agent-overview "quality regression" banner: compares the mean judge score of the latest deployed revision against the prior revision and reports whether it dropped significantly. Only runs the judge scored count toward the scores and sample sizes; runs where the judge itself failed are reported apart in `currentErrors` / `baselineErrors`. Read-only and side-effect-free. Defaults to comparing `PRODUCTION` executions; pass `origin` to compare a different execution origin and `environment` to scope the comparison to one environment slug. Returns `regressed=false` when there is no significant drop or not enough data to judge.
Authorization
Bearer JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.
In: header
Path Parameters
The unique identifier of the agent.
The tenant identifier
Query Parameters
Execution origin to compare. Defaults to PRODUCTION.
"PRODUCTION""UNSPECIFIED" | "PRODUCTION" | "TEST"Optional environment slug to scope the comparison to.
Response Body
curl -X GET "https://api.fruxon.com/v1/tenants/string/agents/string/judgeRegression"{
"regressed": false,
"metric": "FIRST_DRAFT_PASS_RATE",
"currentRevision": 0,
"currentValue": 0,
"currentScore": 0,
"currentSamples": 0,
"currentPassed": 0,
"currentFailed": 0,
"currentFirstDraftPassed": 0,
"currentErrors": 0,
"baselineRevision": 0,
"baselineValue": 0,
"baselineScore": 0,
"baselineSamples": 0,
"baselinePassed": 0,
"baselineFailed": 0,
"baselineFirstDraftPassed": 0,
"baselineErrors": 0,
"drop": 0
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}Gets the agent's judge-verdict counts (+ average score) for the operator overview Quality card.
Returns passed / failed / scored counts, the average run-level score, and a pass rate over the requested origin + window. Null/null window covers all time; otherwise it defaults to 30 days before to, mirroring the cost-timeseries windowing so the surfaces reconcile.
Lists the agent's revised, failed or unchecked runs, newest first.
The drill-down behind `judgeActivity.revised` (the default), `failed` or `errors`, chosen by outcome, over the same origin, window and environment. Each row carries the participant and the channel address (or, when the session recorded none, the topic) so the client can open the conversation at the accepted reply, whose messages carry this `executionId` as `executionRecordId`. Each row also carries `reasons`, the metrics the judge took points off with its shortened, redacted reason: for a reply that passed after a rewrite, why the draft was sent back. The rejected drafts themselves are in the run's trace.