Skip to main content
Reports whether a session actually worked, from a signal only your app has: the order shipped, the test passed. Same API-key auth as ingest. The SDK wraps this as reportOutcome(sessionId, outcome) (Python: report_outcome).
  • Body: { "session_id": "...", "outcome": "success" | "fail" } (over 16 KB returns 413). session_id is truncated to 200 characters, so a longer id will not match the session it was meant for.
  • Scope: a project key may report on any of the project’s sessions; the per-run key a benchmark run mints may only report on sessions from its own run.
Reported ground truth beats Belvedir’s own judgment of the session, asymmetrically: a fail overrides the session’s tasks to fail, so bad sessions drop out of training data and environments everywhere task outcomes are read, while a success is stored on the session only (it never promotes tasks Belvedir judged failed).
A report for a session the platform hasn’t seen yet returns 404; call flush() first so the session’s spans have arrived.

​
Limits

Rate limit: 2 requests per second sustained per API key, burst 120; 429 with a Retry-After header beyond it.