Submit rollouts: specs x seeds under one judge
Submit ROLLOUTS: one episode per spec per seed, in an environment, under a judge, ending in a verdict (the manual's Rollouts section, gman://docs/llms.txt). The platform owns each rollout's identity (a digest of environment, runner, spec, seed, judge, policy, parent state and reproducibility class), its lineage, and its provenance-stamped verdict, and never interprets your spec. Two kinds: task (your container serving the environment, leased episodes from inside via $GMN_METADATA_URL/v1/tasks/lease and declaring with /v1/rollout/verdicts; catalog containers mjx-rollout and openenv-rollout) and sandbox (a fork per rollout of a baked env-... snapshot, driven and graded by your harness from OUTSIDE; nothing inside a sandbox may declare a verdict). Under profile eval, identical identities return their stored verdict unbilled (memoization, workspace-scoped); under train, memoization is off. Anything that changes the result and is not in your environment, spec, seed, judge or policy makes two different rollouts one identity.
Request
- Base URL: https://autoresearch.sfcompute.com
- URL: https://autoresearch.sfcompute.com/preview/rollouts
- Auth: HTTP bearer
Query parameters
Act in this workspace (its slug). Omitted: the token's bound workspace, then the organization's default. Never ambient - there is no switch-workspace mutation on this door.
Request body
Response
The capability's JSON result.
Changes
No recorded changes to this operation across all 1 revision of this API.