{
  "batch": "2026-08-08-global-theme-gap-review-p0",
  "seed": 42,
  "post_training": {
    "rlaif": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.640625,
        "mean_reward": 0.7236528129820922,
        "entropy": 1.626267651052243,
        "kl_from_reference": 0.16549181816981243
      },
      "relative_accuracy": 1.7333333333333334,
      "diagnostics": {
        "ai_preference_pairs": 2.0,
        "position_swap_debiased": 1.0,
        "preference_margin": 2.0295712838824707,
        "loss": 0.1234485906027844,
        "policy_entropy": 1.6057524392078948
      }
    },
    "process-supervision": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.578125,
        "mean_reward": 0.6657008447056338,
        "entropy": 1.7657541489046507,
        "kl_from_reference": 0.026005320317404178
      },
      "relative_accuracy": 1.4666666666666666,
      "diagnostics": {
        "step_labels": 4.0,
        "active_learning_priority": 0.39717266955864006,
        "outcome_only_gap": 0.4544123045027701,
        "loss": -0.07576985575807513,
        "policy_entropy": 1.7672157502111097
      }
    },
    "math-shepherd": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.625,
        "mean_reward": 0.7119604803714905,
        "entropy": 1.780354995536342,
        "kl_from_reference": 0.011404473685712766
      },
      "relative_accuracy": 1.6666666666666667,
      "diagnostics": {
        "mc_continuations": 36.0,
        "automatic_process_labels": 6.0,
        "step_label_mean": 0.16666666666666666,
        "loss": -0.0,
        "policy_entropy": 1.785262811495159
      }
    },
    "self-rewarding": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.625,
        "mean_reward": 0.7112418100993936,
        "entropy": 1.6288551878620738,
        "kl_from_reference": 0.16290428135998133
      },
      "relative_accuracy": 1.6666666666666667,
      "diagnostics": {
        "self_judged_pairs": 1.0,
        "judge_policy_shared": 1.0,
        "iterative_preference_round": 15.0,
        "loss": 0.12837263576487323,
        "policy_entropy": 1.6086931745469233
      }
    },
    "luffy": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.53125,
        "mean_reward": 0.625786691434544,
        "entropy": 1.7623159054021138,
        "kl_from_reference": 0.029443563819941357
      },
      "relative_accuracy": 1.2666666666666666,
      "diagnostics": {
        "on_policy_rollouts": 4.0,
        "off_policy_guidance": 2.0,
        "regularized_is_mean": 1.7,
        "loss": -0.0817804293506927,
        "policy_entropy": 1.7690105650326344
      }
    },
    "ttrl": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "relative_accuracy": 0.0,
      "diagnostics": {
        "ground_truth_labels": 0.0,
        "majority_vote_size": 4.0,
        "consensus_rate": 1.0,
        "loss": -0.0,
        "policy_entropy": 1.791759469222055
      }
    },
    "absolute-zero": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.59375,
        "mean_reward": 0.6900456082956962,
        "entropy": 1.7882660824248766,
        "kl_from_reference": 0.0034933867971783206
      },
      "relative_accuracy": 1.5333333333333334,
      "diagnostics": {
        "human_curated_examples": 0.0,
        "self_generated_tasks": 4.0,
        "verifier_calls": 4.0,
        "curriculum_score": 0.5,
        "loss": -0.00893609247142449,
        "policy_entropy": 1.7881448703061056
      }
    },
    "intuitor": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "relative_accuracy": 0.0,
      "diagnostics": {
        "external_reward_calls": 0.0,
        "self_certainty_mean": 5.999867269679271e-12,
        "intrinsic_kl_to_uniform": 5.999867269679271e-12,
        "loss": -0.0,
        "policy_entropy": 1.791759469222055
      }
    },
    "cispo": {
      "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": 1.7083417737738014,
        "kl_from_reference": 0.08341769544825355
      },
      "relative_accuracy": 1.1333333333333333,
      "diagnostics": {
        "importance_ratio_mean": 1.0045495000243834,
        "clipped_ratio_fraction": 0.0,
        "token_level_is_clip": 1.0,
        "loss": -0.2392642762851383,
        "policy_entropy": 1.6949709587143067
      }
    },
    "spiral": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.34375,
        "mean_reward": 0.5013214140334054,
        "entropy": 1.7895487133381456,
        "kl_from_reference": 0.002210755883909164
      },
      "relative_accuracy": 0.4666666666666667,
      "diagnostics": {
        "self_play_roles": 2.0,
        "role_conditioned_advantages": 4.0,
        "human_reward_calls": 0.0,
        "loss": -0.0026650785339779604,
        "policy_entropy": 1.7899825315327824
      }
    },
    "conspo": {
      "baseline": {
        "accuracy": 0.234375,
        "mean_reward": 0.38409619661640243,
        "entropy": 1.7917594692220566,
        "kl_from_reference": 0.0
      },
      "final": {
        "accuracy": 0.640625,
        "mean_reward": 0.7241567802319283,
        "entropy": 1.7828058034807128,
        "kl_from_reference": 0.008953665741342378
      },
      "relative_accuracy": 1.7333333333333334,
      "diagnostics": {
        "length_normalized_sequence_scores": 4.0,
        "infonce_group_size": 4.0,
        "curriculum_margin": 1.0,
        "reward_gap_std": 0.0,
        "loss": 2.751537321506218,
        "policy_entropy": 1.7838997827573817
      }
    }
  },
  "agent": {
    "deepresearcher": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 1.5
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "reflections": 120,
        "policy_updates": 120,
        "browser_queries": 360,
        "references_collected": 360,
        "search_queries": 360,
        "outcome_rewards": 120,
        "trajectory_rollouts": 240
      }
    },
    "retool": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 1.050000000000001
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "reasoning_steps": 480,
        "actions": 360,
        "verification_retries": 120,
        "policy_updates": 120,
        "outcome_rewards": 120,
        "real_tool_responses": 360
      }
    },
    "toolrl": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 1.125
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "tool_call_candidates": 540,
        "tool_calls_accepted": 360,
        "policy_updates": 120,
        "dense_credit_updates": 540
      }
    },
    "sage": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.3000000000000001
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "skills_created": 12,
        "skills_reused": 108,
        "policy_updates": 120,
        "outcome_rewards": 120,
        "trajectory_rollouts": 240,
        "skill_document_updates": 12,
        "cross_task_skill_reuses": 108
      }
    },
    "memskill": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.32000000000000023
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "skills_created": 48,
        "skills_reused": 108,
        "policy_updates": 120,
        "skill_document_updates": 48,
        "memory_operations": 120
      }
    },
    "memento-skills": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 0.05,
        "joint_success": 0.05,
        "average_cost": 0.28000000000000047
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "reflections": 114,
        "skills_reused": 114,
        "skill_document_updates": 120,
        "memory_operations": 240
      }
    },
    "searl": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.34000000000000064
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "policy_updates": 120,
        "skill_graph_nodes": 12,
        "skill_graph_edges": 12,
        "downstream_credit_updates": 360,
        "step_value_queries": 360,
        "router_updates": 120,
        "memory_bank_updates": 120
      }
    },
    "agent0": {
      "metrics": {
        "answer_accuracy": 1.0,
        "plan_success": 1.0,
        "joint_success": 1.0,
        "average_cost": 0.3199999999999995
      },
      "diagnostics": {
        "episodes": 120,
        "memory_size": 24,
        "actions": 360,
        "policy_updates": 120,
        "outcome_rewards": 120,
        "trajectory_rollouts": 360,
        "simulated_user_turns": 420
      }
    }
  },
  "schema_version": 2,
  "manifest_ref": "experiments:global-p0-20260808-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-p0-20260808-seed42.json",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded",
    "dataset_fingerprint": "not recorded in historical artifact"
  }
}