{
  "schema": "cain.e129.unification_4arm.v1",
  "seeds": [
    12901,
    12902,
    12903,
    12904,
    12905
  ],
  "total_scored_evaluations": 800,
  "duration_s": 81.07,
  "arms": {
    "A": {
      "n_paired_tasks": 200,
      "composite_success_rate": 0.34,
      "composite_ci_95": [
        0.2779,
        0.4081
      ],
      "generalization_macro_rate": 0.25,
      "causal_pass_rate": 0.43,
      "causal_mean_mae": 0.7229,
      "total_evaluations_used": 735024,
      "compute_efficiency_solved_per_10k_evals": 0.9251,
      "llm_tokens_used": 218,
      "governance_violations": 0,
      "memory_retention": 0.0,
      "improvement_inheritance": false
    },
    "B": {
      "n_paired_tasks": 200,
      "composite_success_rate": 0.385,
      "composite_ci_95": [
        0.3203,
        0.454
      ],
      "generalization_macro_rate": 0.28,
      "causal_pass_rate": 0.49,
      "causal_mean_mae": 0.7114,
      "total_evaluations_used": 1490763,
      "compute_efficiency_solved_per_10k_evals": 0.5165,
      "llm_tokens_used": 218,
      "governance_violations": 0,
      "memory_retention": 0.0,
      "improvement_inheritance": false
    },
    "C": {
      "n_paired_tasks": 200,
      "composite_success_rate": 0.59,
      "composite_ci_95": [
        0.5208,
        0.6558
      ],
      "generalization_macro_rate": 0.75,
      "causal_pass_rate": 0.43,
      "causal_mean_mae": 0.7382,
      "total_evaluations_used": 646311,
      "compute_efficiency_solved_per_10k_evals": 1.8257,
      "llm_tokens_used": 218,
      "governance_violations": 0,
      "memory_retention": 0.5,
      "improvement_inheritance": false
    },
    "D": {
      "n_paired_tasks": 200,
      "composite_success_rate": 0.7,
      "composite_ci_95": [
        0.6332,
        0.7593
      ],
      "generalization_macro_rate": 0.91,
      "causal_pass_rate": 0.49,
      "causal_mean_mae": 0.7067,
      "total_evaluations_used": 385909,
      "compute_efficiency_solved_per_10k_evals": 3.6278,
      "llm_tokens_used": 218,
      "governance_violations": 0,
      "memory_retention": 1.0,
      "improvement_inheritance": true
    }
  },
  "comparisons_composite": {
    "D_vs_A": {
      "test": "exact_mcnemar",
      "n": 200,
      "rate_a": 0.7,
      "rate_b": 0.34,
      "gain": 0.36,
      "a_only": 88,
      "b_only": 16,
      "p_value": 0.0,
      "effect_size": 0.7372,
      "effect_size_type": "cohens_h",
      "confidence_interval_diff": [
        0.2734,
        0.4466
      ]
    },
    "D_vs_B": {
      "test": "exact_mcnemar",
      "n": 200,
      "rate_a": 0.7,
      "rate_b": 0.385,
      "gain": 0.315,
      "a_only": 84,
      "b_only": 21,
      "p_value": 4.16e-10,
      "effect_size": 0.6436,
      "effect_size_type": "cohens_h",
      "confidence_interval_diff": [
        0.2246,
        0.4054
      ]
    },
    "D_vs_C": {
      "test": "exact_mcnemar",
      "n": 200,
      "rate_a": 0.7,
      "rate_b": 0.59,
      "gain": 0.11,
      "a_only": 37,
      "b_only": 15,
      "p_value": 0.003185253448,
      "effect_size": 0.2305,
      "effect_size_type": "cohens_h",
      "confidence_interval_diff": [
        0.041,
        0.179
      ]
    }
  },
  "comparisons_causal_mae": {
    "D_vs_A": {
      "test": "paired_wilcoxon_signed_rank",
      "n": 100,
      "mean_a": 0.7067,
      "mean_b": 0.7229,
      "mean_improvement": 0.0162,
      "wins": 50,
      "losses": 50,
      "ties": 0,
      "p_value": 0.657372132727,
      "sign_test_p": 1.0,
      "effect_size": 0.0469,
      "effect_size_type": "paired_cohens_d",
      "rank_biserial_r": 0.0,
      "confidence_interval_diff": [
        -0.0525,
        0.0849
      ]
    },
    "D_vs_B": {
      "test": "paired_wilcoxon_signed_rank",
      "n": 100,
      "mean_a": 0.7067,
      "mean_b": 0.7114,
      "mean_improvement": 0.0047,
      "wins": 51,
      "losses": 49,
      "ties": 0,
      "p_value": 0.962977552576,
      "sign_test_p": 0.920410762613,
      "effect_size": 0.0119,
      "effect_size_type": "paired_cohens_d",
      "rank_biserial_r": 0.02,
      "confidence_interval_diff": [
        -0.0737,
        0.0831
      ]
    },
    "D_vs_C": {
      "test": "paired_wilcoxon_signed_rank",
      "n": 100,
      "mean_a": 0.7067,
      "mean_b": 0.7382,
      "mean_improvement": 0.0315,
      "wins": 52,
      "losses": 48,
      "ties": 0,
      "p_value": 0.409259210027,
      "sign_test_p": 0.764353434403,
      "effect_size": 0.0852,
      "effect_size_type": "paired_cohens_d",
      "rank_biserial_r": 0.04,
      "confidence_interval_diff": [
        -0.0419,
        0.105
      ]
    }
  },
  "holm_adjusted_p": {
    "D_vs_A": 0.0,
    "D_vs_B": 8.32e-10,
    "D_vs_C": 0.003185253448
  },
  "bh_fdr_q": {
    "D_vs_A": 0.0,
    "D_vs_B": 6.24e-10,
    "D_vs_C": 0.003185253448
  },
  "unification_advantage_verified": true,
  "scientific_verdict": "PASS",
  "integrity_verdict": "VERIFIED",
  "limitations": [
    "Constructed multi-domain benchmark combining integer-rule macro synthesis and 6-lever confounded SCMs.",
    "Matched compute budgets across A/B/C/D; Arm D uses inherited G1+G2 validated improvements from prior governed generations (whose discovery compute is disclosed separately in multi_generation_rsi.json).",
    "Zero external benchmark or third-party validation."
  ]
}