Benchmark evidence workspace
Hatchery holds the recordings, outputs, grading, and files behind a benchmark, and the corrections applied to them. A run has a name and a selected revision; a revision records which attempts, grades, and policy produce the values you read. Every displayed number opens the evidence and the changes behind it.
Customers use their own agents through the page or the HTTP API. Both receive the same pinned revision and return links that open the cited result here.
Reading available workspaces