API: Inspect Results - SelfBench

Method	Route	Does
`GET`	`{scope}/evaluations/results`	Scores, and each task’s results
`GET`, `POST`	`{scope}/evaluations/{id}/explain`	Explain a failed trial
`GET`, `POST`	`{scope}/evaluations/audit`	Audit a task’s fairness
`GET`, `POST`	`{scope}/evaluations/summaries`	Model summaries

Use GET {scope}/evaluations/results for scores rather than adding up trials. Results count each configuration’s latest scored result on each counted task: a repository’s accepted tasks, or a benchmark’s tasks. Errors, cancellations, and trials still running don’t count.

Request Bodies

Route	Body
`POST {scope}/evaluations/{id}/explain`	`{ runId, taskId, harness, explanation? }`
`POST {scope}/evaluations/audit`	`{ runId, taskId, audit: { credentialId, model } }`
`POST {scope}/evaluations/summaries`	`{ configurations: [keys] (≤12), summary: { credentialId, model } }`