reduced_recovery_harness
Execute a bounded soft-mode proxy experiment that validates P3O mechanisms with generated skills and numeric training evidence.
SKILL.mdbasicagent-skill/skill_library/paperbench/sapg/skill/p3o_policy_on_off/reduced_recovery_harness/SKILL.md
GitHubReduced Recovery Harness
Use this skill when full Atari or MuJoCo P3O training is blocked by runtime cost or missing benchmark packages, and soft mode permits a declared proxy. Do not report this proxy as a full paper reproduction.
Inputs
- Attempt directory with
module_plan.jsonand runtime handoff. - Generated skill root containing ESS, surrogate-loss, and replay-protocol scripts.
- A tiny constructed on-policy and replay batch.
Outputs
recovery/recovery_result.jsonwith numericloss_reduction.recovery/logs/generated_data_item.jsonandtraining_trace.json.- Command and generated-skill invocation logs.
Workflow
- Load the module-plan target so recovery metadata cannot drift.
- Invoke the generated ESS scheduler on replay policy probabilities.
- Invoke the generated replay protocol to preserve Algorithm 1 ordering.
- Invoke the generated P3O loss component calculator before and after one optimizer step.
- Record mechanism checks and validation-ready artifacts.
Source Boundary
Use this skill with the paper, module documents, generated artifacts, and ordinary package documentation. Do not read or depend on the original P3O repository.
Validation
Run python scripts/<script>.py --self-test or python -m pytest tests from the skill directory. The bundled tests use only the Python standard library.