false_statement_judge: resources_servers: false_statement_judge: entrypoint: app.py domain: math verified: false description: Sycophancy grader for prove-the-false-statement benchmarks, using MathArena's 0-2 judge rubric value: Measure how often a model claims to prove a statement that is false as written judge_model_server: type: responses_api_models name: judge_model judge_responses_create_params: input: [] # The rubric is a 13-question decision list applied to a full proof # attempt, so the judge needs room to reason before emitting . # Sized for the tail: a typical verdict is far smaller, but a judge that # runs long would otherwise truncate before the block and be # scored as un-gradable. max_output_tokens: 131072 temperature: 0.0 top_p: 1.0 false_statement_judge_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py resources_server: type: resources_servers name: false_statement_judge model_server: type: responses_api_models name: policy_model datasets: - name: example type: example jsonl_fpath: resources_servers/false_statement_judge/data/example.jsonl false_statement_judge_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py resources_server: type: resources_servers name: false_statement_judge agent_server: type: responses_api_agents name: false_statement_judge_simple_agent resources_tool_transports: - direct_http