nemo_gym.eval.recovery
nemo_gym.eval.recovery
The evaluation recovery policy: what gym eval resubmit --recover decides from a run’s saved evidence.
The entry points, each in the module that owns it:
evidence:COLLECT_SCRIPTreads a job directory intoCollectedEvidence;select_managed_evidenceadds the run’s own Store selection.planner:decide_from_collectedturns the evidence into aRecoveryDecision(decision) under the run’sRecoveryOutcomePolicyConfig(policy) and budgets (budgets).gate:gate_summary_from_collectedandgate_task_summaryreport the acceptance gate of a decision.grading:reverification_causesays why a saved row is graded again from its saved answer.continuation:continuation_admission, which a chain’s job script runs in the run’s installed runtime.
rows (identity states), trials (trial observations and fault classes) and markers (the strings the chain
scripts write) serve those modules. Executing a decision (queueing a follow-up, a judge-only job, a fresh restart) is
the caller’s part. This package never imports nemo_gym.orchestration; import-linter enforces it.
Submodules
nemo_gym.eval.recovery.budgetsnemo_gym.eval.recovery.continuationnemo_gym.eval.recovery.decisionnemo_gym.eval.recovery.evidencenemo_gym.eval.recovery.gatenemo_gym.eval.recovery.gradingnemo_gym.eval.recovery.markersnemo_gym.eval.recovery.plannernemo_gym.eval.recovery.policynemo_gym.eval.recovery.rowsnemo_gym.eval.recovery.trials