Results
Results: one attempt of an attack against a goal, with the request sent to the agent, its response, the judges' evaluation and the trace of steps.
List results
Returns the results of the caller's organization, newest first. Filter by `run`, `goal_index` or `evaluation_status`, and leave the traces out with `include_traces=false` for a lighter response. Paginated with `page` and `page_size` (default 10, at most 1000). Staff users see every organization's results.
Record a result
Records the outcome of one attempt against the agent in a run (`run` is required). The SDK records one per attempt, with its goal, and evaluates it later with `PATCH /result/{id}`. A cloud run's token may write results of its own run tree only.
Retrieve a result
Returns one result of the caller's organization, with its trace steps.
Replace a result
Replaces every writable field of a result (send them all). A cloud run's token may write results of its own run tree only.
Update a result
Changes only the fields sent of a result; the SDK uses it to record the judges' evaluation (`evaluation_status`, `evaluation_notes`, `evaluation_metrics`). A cloud run's token may write results of its own run tree only.
Delete a result
Deletes a result and its trace steps. This cannot be undone.
Add a trace step to a result
Appends a step (`sequence`, `step_type`, `content`) to the result's trace; `sequence` numbers the steps and is unique per result. A cloud run's token may write traces of its own run tree only.