math_proof_judgement:
  resources_servers:
    math_proof_judgement:
      entrypoint: app.py
      domain: math
      verified: false
      description: "Binary judgement of math proofs — the policy model reads a problem plus a candidate proof and outputs Judgement: Yes/No."
      value: "Improve the NVIDIA ProofBench judge benchmark and math-proof verification skill."

math_proof_judgement_simple_agent:
  responses_api_agents:
    simple_agent:
      entrypoint: app.py
      resources_server:
        type: resources_servers
        name: math_proof_judgement
      model_server:
        type: responses_api_models
        name: policy_model
      datasets:
        - name: example
          type: example
          jsonl_fpath: resources_servers/math_proof_judgement/data/example.jsonl