{
  "batch": "2026-08-08-global-theme-gap-review-p1",
  "seed": 42,
  "post_training": {
    "minirl": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.5,
        "mean_reward": 0.6009954910247568,
        "entropy": 1.7123607826235443,
        "kl_from_reference": 0.07939868659851035
      },
      "relative_accuracy": 1.1333333333333333,
      "diagnostics": {
        "importance_sampling": 1.0,
        "ratio_clip_fraction": 0.0,
        "routing_replay_unique_experts": 1.0,
        "policy_staleness_l1": 0.004929395593189831,
        "loss": -0.26652388641464364,
        "policy_entropy": 1.700001874078903
      }
    },
    "missing-old-logits": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.5,
        "mean_reward": 0.6009954910247568,
        "entropy": 1.7078264791391293,
        "kl_from_reference": 0.0839329900829258
      },
      "relative_accuracy": 1.1333333333333333,
      "diagnostics": {
        "old_logits_available": 0.0,
        "ewma_reconstruction": 1.0,
        "mean_discrepancy_ratio": 1.0035681045446416,
        "mean_staleness_ratio": 1.0070021608718696,
        "loss": -0.3403654639144065,
        "policy_entropy": 1.6944808903903144
      }
    },
    "stare": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.53125,
        "mean_reward": 0.6268592549565025,
        "entropy": 1.7220227280819853,
        "kl_from_reference": 0.06973674114006992
      },
      "relative_accuracy": 1.2666666666666666,
      "diagnostics": {
        "critical_token_fraction": 0.5,
        "surprisal_q25": 1.7253665562139642,
        "surprisal_q75": 2.045844371976182,
        "target_entropy": 1.2900668178441996,
        "closed_loop_gate": -0.33425035371530826,
        "loss": -0.0,
        "policy_entropy": 1.7212721080250055
      }
    }
  },
  "agent": {
    "agent-r1": {
      "benchmark": "planbench-mini",
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.9599999999999996
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "policy_updates": 120,
        "context_compressions": 360,
        "step_value_queries": 360,
        "step_gae_updates": 360,
        "transition_targets": 360,
        "transition_accuracy": 1.0
      }
    },
    "camel": {
      "benchmark": "planbench-mini",
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 1.3500000000000003
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "reasoning_steps": 360,
        "actions": 360,
        "agent_messages": 720
      }
    },
    "toolbench": {
      "benchmark": "planbench-mini",
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.8400000000000015
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "affordance_checks": 360,
        "programs_generated": 360,
        "interpreter_calls": 360,
        "task_examples_retrieved": 108,
        "task_library_updates": 12,
        "tools_exposed": 360,
        "tools_available": 1440,
        "tool_exposure_reduction": 0.75
      }
    },
    "gaia": {
      "benchmark": "gaia-mini",
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 1.3999999999999992
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 420,
        "tool_call_candidates": 420,
        "tool_calls_accepted": 420,
        "browser_queries": 120,
        "references_collected": 120,
        "interpreter_calls": 90,
        "local_verifier_calls": 120
      }
    }
  },
  "schema_version": 2,
  "manifest_ref": "experiments:global-p1-20260808-seed42",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/experiments/global-p1-20260808-seed42.json",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded",
    "dataset_fingerprint": "not recorded in historical artifact"
  }
}
