{
  "schema_version": 2,
  "device": "cpu",
  "accelerator": "CPU",
  "seeds": [
    42,
    43,
    44
  ],
  "code_commit": "4cce1ab4d701df8daaed9a8abe8671f3eef9d168",
  "source_sha256": {
    "scripts/run_google_emnlp25_components.py": "bb336482f0927c7d1164d93534309a78d8258acaaab9f90549ef397fdf12a0d9",
    "docs/research-audits/google-emnlp25-closure/execution-spec.json": "6fa9f60245d7bad1ff087b0fd387360c32597912c2a004f4d7adfee536a9f1bf",
    "scripts/oct18_frozen_sources.py": "da08355049c1f6b5184ee659643c9be9decedf2041ef6a87a3dce2e4582e385a",
    "src/auto_research/agent_research/dmast.py": "9bd1d38a133cac57e2cac822a5f8050b0208196c1a5dd011fe67e588823444d3",
    "src/auto_research/agent_research/finance_harness_protocol.py": "4799418d649ce1793e763c2445366d19e3f909b6f924369d9a03804e51a25cb4",
    "src/auto_research/agent_research/sopbench_protocol.py": "f8645a19302ba354354866f5bf7bf6342a3342bee38d8d69cfd1402826398a7d",
    "src/auto_research/evaluation/xlgobench.py": "7ff09bf3d807f4c4d47728960b014f32161c9d5595ad2258edef25a48ec1bdb6",
    "src/auto_research/foundation_models/combinatorial_watermark.py": "ee1ceba5cd67ad81d4dbee5b0746e00ac15596466fd0099ab1bf0661a8829668",
    "src/auto_research/foundation_models/interpolated_null_calibration.py": "15da6e0c838ee16eec2ff13708b811699d82f5c6759d8e8da49f3c5561ce7452",
    "src/auto_research/foundation_models/mechanistic_etm.py": "4a0a27f1aa28ed0ad5333324333c1db712ddb6bfdf522e470b0692a16502ac83",
    "src/auto_research/foundation_models/taskpress.py": "8ac4db9204ccac1a29145e3c1fb57cae7fa02ece30b4325a76aa6e06750a65e9",
    "src/auto_research/multimodal/cue2narrate_drl.py": "15f35e0ec9e88684aa93f07418459768e8d010755d6837f2cd3d5dd84e6f9e71",
    "src/auto_research/post_training/benchmark_self_bias.py": "2f4aa1a0dde3c03a018344b286a897d328cf31b78e170ff463f15c9eee252282",
    "src/auto_research/post_training/polaris_hri.py": "4b7f7c7e0537c3483646fb9ea8f68c3af3c859781d87427a307d90d88f9a163a",
    "src/auto_research/post_training/probabilistic_autorater.py": "76ae4cbb9bdf55e54ae30528d24d81868f3e1b02eed1d62a814d883f55f5cabe",
    "src/auto_research/post_training/rm_memorization_protocol.py": "a91fcc7b7d2c0a5773f1e1c53a8886e54426a2777588672f8212c7a2c961f5f8",
    "src/auto_research/post_training/scar_shapley_credit.py": "363c7f0c6abeb59b5809b33031f82e4dd769c4c68face3abd738696958698850"
  },
  "components": {
    "scar-shapley-credit": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 5.551115123125783e-17,
            "return_preservation_error": 0.0,
            "coalitions": 16,
            "players": 4,
            "model_calls": 0
          },
          "checks": {
            "exact_shapley_matches_independent_game": true,
            "dummy_player_zero": true,
            "signed_contributions_not_clamped": true,
            "efficiency_preserves_undiscounted_return": true,
            "span_end_only_not_uniform_spread": true,
            "alpha_zero_is_sparse_reward": true,
            "frozen_reward_model_table": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 5.551115123125783e-17,
            "return_preservation_error": 2.220446049250313e-16,
            "coalitions": 16,
            "players": 4,
            "model_calls": 0
          },
          "checks": {
            "exact_shapley_matches_independent_game": true,
            "dummy_player_zero": true,
            "signed_contributions_not_clamped": true,
            "efficiency_preserves_undiscounted_return": true,
            "span_end_only_not_uniform_spread": true,
            "alpha_zero_is_sparse_reward": true,
            "frozen_reward_model_table": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 5.551115123125783e-17,
            "return_preservation_error": 0.0,
            "coalitions": 16,
            "players": 4,
            "model_calls": 0
          },
          "checks": {
            "exact_shapley_matches_independent_game": true,
            "dummy_player_zero": true,
            "signed_contributions_not_clamped": true,
            "efficiency_preserves_undiscounted_return": true,
            "span_end_only_not_uniform_spread": true,
            "alpha_zero_is_sparse_reward": true,
            "frozen_reward_model_table": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "probabilistic-autorater": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "metrics": {
            "analytic_loss_before": -0.6820574101645932,
            "analytic_loss_after": -0.682589852908247,
            "formula_abs_error": 0.0,
            "gradient_max_abs_error": 3.469446951953614e-18,
            "model_calls": 0
          },
          "checks": {
            "proper_reward_matches_expected_brier": true,
            "analytic_gradient_matches": true,
            "analytic_optimizer_update_decreases_loss": true,
            "unparseable_penalties_exact": true,
            "invalid_probability_zero_gradient": true,
            "strict_tag_parser_rejects_duplicates": true,
            "overconfident_wrong_log_penalty": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "metrics": {
            "analytic_loss_before": -0.8069839927363702,
            "analytic_loss_after": -0.8070663296674313,
            "formula_abs_error": 0.0,
            "gradient_max_abs_error": 3.469446951953614e-18,
            "model_calls": 0
          },
          "checks": {
            "proper_reward_matches_expected_brier": true,
            "analytic_gradient_matches": true,
            "analytic_optimizer_update_decreases_loss": true,
            "unparseable_penalties_exact": true,
            "invalid_probability_zero_gradient": true,
            "strict_tag_parser_rejects_duplicates": true,
            "overconfident_wrong_log_penalty": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "metrics": {
            "analytic_loss_before": -0.7752344801003366,
            "analytic_loss_after": -0.7754797184372453,
            "formula_abs_error": 1.1102230246251565e-16,
            "gradient_max_abs_error": 3.469446951953614e-18,
            "model_calls": 0
          },
          "checks": {
            "proper_reward_matches_expected_brier": true,
            "analytic_gradient_matches": true,
            "analytic_optimizer_update_decreases_loss": true,
            "unparseable_penalties_exact": true,
            "invalid_probability_zero_gradient": true,
            "strict_tag_parser_rejects_duplicates": true,
            "overconfident_wrong_log_penalty": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "rm-memorization-protocol": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 0.0,
            "pairs": 4,
            "observed_runs": 25,
            "mean_counterfactual_memorization": -0.04936955632767136,
            "model_calls": 0
          },
          "checks": {
            "conditional_probability_means_match": true,
            "cm_is_tm_minus_tg": true,
            "user_level_leakage_rejected": true,
            "joint_run_permutation_invariant": true,
            "frozen_external_model_scores": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 1.1102230246251565e-16,
            "pairs": 4,
            "observed_runs": 25,
            "mean_counterfactual_memorization": -0.11892493846410102,
            "model_calls": 0
          },
          "checks": {
            "conditional_probability_means_match": true,
            "cm_is_tm_minus_tg": true,
            "user_level_leakage_rejected": true,
            "joint_run_permutation_invariant": true,
            "frozen_external_model_scores": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 1.1102230246251565e-16,
            "pairs": 4,
            "observed_runs": 25,
            "mean_counterfactual_memorization": 0.07091150466444614,
            "model_calls": 0
          },
          "checks": {
            "conditional_probability_means_match": true,
            "cm_is_tm_minus_tg": true,
            "user_level_leakage_rejected": true,
            "joint_run_permutation_invariant": true,
            "frozen_external_model_scores": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "benchmark-self-bias": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 0.0,
            "mean_relative_self_bias": -1.5,
            "segments": 7,
            "models": 3,
            "model_calls": 0
          },
          "checks": {
            "leave_self_out_formula_matches": true,
            "negative_means_self_preference": true,
            "model_role_joint_permutation_equivariant": true,
            "positive_score_affine_invariant": true,
            "explicit_average_ties": true,
            "metricx_direction_explicit": true,
            "not_absolute_quality_estimate": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 0.0,
            "mean_relative_self_bias": -1.5,
            "segments": 7,
            "models": 3,
            "model_calls": 0
          },
          "checks": {
            "leave_self_out_formula_matches": true,
            "negative_means_self_preference": true,
            "model_role_joint_permutation_equivariant": true,
            "positive_score_affine_invariant": true,
            "explicit_average_ties": true,
            "metricx_direction_explicit": true,
            "not_absolute_quality_estimate": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "metrics": {
            "formula_max_abs_error": 0.0,
            "mean_relative_self_bias": -1.5,
            "segments": 7,
            "models": 3,
            "model_calls": 0
          },
          "checks": {
            "leave_self_out_formula_matches": true,
            "negative_means_self_preference": true,
            "model_role_joint_permutation_equivariant": true,
            "positive_score_affine_invariant": true,
            "explicit_average_ties": true,
            "metricx_direction_explicit": true,
            "not_absolute_quality_estimate": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "dmast": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "loss": 0.005321757503511502,
            "gradient_reference_error": 3.469446951953614e-18,
            "formula_reference_error": 1.5612511283791264e-17,
            "parameter_delta": 0.0023947898446033218,
            "valid_tokens": 35,
            "eligible_trajectories": 6,
            "optimizer_steps": 1
          },
          "checks": {
            "raw_pooled_role_statistics": true,
            "zero_raw_variance_excluded": true,
            "clean_attack_modes_separated": true,
            "gradient_matches_reference": true,
            "padding_and_excluded_zero_gradient": true,
            "frozen_rollout_reference_rewards": true,
            "real_shared_policy_update": true,
            "zero_importance_branch_finite_zero_gradient": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "loss": 0.005321757503511496,
            "gradient_reference_error": 3.469446951953614e-18,
            "formula_reference_error": 3.469446951953614e-18,
            "parameter_delta": 0.001781377153675355,
            "valid_tokens": 35,
            "eligible_trajectories": 6,
            "optimizer_steps": 1
          },
          "checks": {
            "raw_pooled_role_statistics": true,
            "zero_raw_variance_excluded": true,
            "clean_attack_modes_separated": true,
            "gradient_matches_reference": true,
            "padding_and_excluded_zero_gradient": true,
            "frozen_rollout_reference_rewards": true,
            "real_shared_policy_update": true,
            "zero_importance_branch_finite_zero_gradient": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "loss": 0.005321757503511496,
            "gradient_reference_error": 3.469446951953614e-18,
            "formula_reference_error": 3.469446951953614e-18,
            "parameter_delta": 0.002731507804682134,
            "valid_tokens": 35,
            "eligible_trajectories": 6,
            "optimizer_steps": 1
          },
          "checks": {
            "raw_pooled_role_statistics": true,
            "zero_raw_variance_excluded": true,
            "clean_attack_modes_separated": true,
            "gradient_matches_reference": true,
            "padding_and_excluded_zero_gradient": true,
            "frozen_rollout_reference_rewards": true,
            "real_shared_policy_update": true,
            "zero_importance_branch_finite_zero_gradient": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "sopbench-protocol": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "fixture_cases": 4,
            "observed_pass_cases": 1,
            "false_accepts": 0,
            "model_calls": 0
          },
          "checks": {
            "three_dimensions_conjoined": true,
            "chain_order_enforced": true,
            "wrong_request_helper_rejected": true,
            "oracle_refutes_claimed_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "fixture_cases": 4,
            "observed_pass_cases": 1,
            "false_accepts": 0,
            "model_calls": 0
          },
          "checks": {
            "three_dimensions_conjoined": true,
            "chain_order_enforced": true,
            "wrong_request_helper_rejected": true,
            "oracle_refutes_claimed_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "fixture_cases": 4,
            "observed_pass_cases": 1,
            "false_accepts": 0,
            "model_calls": 0
          },
          "checks": {
            "three_dimensions_conjoined": true,
            "chain_order_enforced": true,
            "wrong_request_helper_rejected": true,
            "oracle_refutes_claimed_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "finance-harness-protocol": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "documents": 4,
            "visible_documents": 2,
            "question_equal_score": 0.5,
            "pooled_item_negative_control": 0.25,
            "model_calls": 0
          },
          "checks": {
            "search_cutoff_inclusive_fail_closed": true,
            "read_cannot_bypass_cutoff": true,
            "question_equal_not_item_equal": true,
            "live_web_audit_invalidates_submission": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "documents": 4,
            "visible_documents": 2,
            "question_equal_score": 0.5,
            "pooled_item_negative_control": 0.25,
            "model_calls": 0
          },
          "checks": {
            "search_cutoff_inclusive_fail_closed": true,
            "read_cannot_bypass_cutoff": true,
            "question_equal_not_item_equal": true,
            "live_web_audit_invalidates_submission": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "documents": 4,
            "visible_documents": 2,
            "question_equal_score": 0.5,
            "pooled_item_negative_control": 0.25,
            "model_calls": 0
          },
          "checks": {
            "search_cutoff_inclusive_fail_closed": true,
            "read_cannot_bypass_cutoff": true,
            "question_equal_not_item_equal": true,
            "live_web_audit_invalidates_submission": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "cue2narrate-drl": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "formula_error": 8.881784197001252e-16,
            "gradient_error": 5.551115123125783e-17,
            "negative_gradient_norm": 0.02243025852940085,
            "fixture_loss_before": 5.518487899970988,
            "fixture_loss_after": 5.477329315904614,
            "positive_tokens": 9,
            "negative_tokens": 18,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "equations5_8_reference": true,
            "both_current_policy_gradients_match": true,
            "prompt_padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "formula_error": 0.0,
            "gradient_error": 5.551115123125783e-17,
            "negative_gradient_norm": 0.05527159223620806,
            "fixture_loss_before": 7.947262190270929,
            "fixture_loss_after": 7.886555786619788,
            "positive_tokens": 9,
            "negative_tokens": 18,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "equations5_8_reference": true,
            "both_current_policy_gradients_match": true,
            "prompt_padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "formula_error": 0.0,
            "gradient_error": 5.551115123125783e-17,
            "negative_gradient_norm": 0.07487438089560484,
            "fixture_loss_before": 9.685957186179243,
            "fixture_loss_after": 9.615204842075261,
            "positive_tokens": 9,
            "negative_tokens": 18,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "equations5_8_reference": true,
            "both_current_policy_gradients_match": true,
            "prompt_padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "polaris-hri": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "advantage_reference_error": 4.440892098500626e-16,
            "loss_reference_error": 1.942890293094024e-16,
            "reference_gradient_norm": 0.0687195845865246,
            "fixture_loss_before": -0.1698670325171836,
            "fixture_loss_after": -0.17033381107863382,
            "policy_rollouts_per_group": 5,
            "references_per_group": 1,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq5_policy_only_sample_statistics": true,
            "zero_step_reference_warmup": true,
            "reference_uses_same_clipped_surrogate": true,
            "old_and_advantages_frozen": true,
            "padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "advantage_reference_error": 4.440892098500626e-16,
            "loss_reference_error": 1.3877787807814457e-16,
            "reference_gradient_norm": 0.06976512274955257,
            "fixture_loss_before": -0.17881792521714995,
            "fixture_loss_after": -0.17928858509636142,
            "policy_rollouts_per_group": 5,
            "references_per_group": 1,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq5_policy_only_sample_statistics": true,
            "zero_step_reference_warmup": true,
            "reference_uses_same_clipped_surrogate": true,
            "old_and_advantages_frozen": true,
            "padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "advantage_reference_error": 4.440892098500626e-16,
            "loss_reference_error": 1.6653345369377348e-16,
            "reference_gradient_norm": 0.06703217929258426,
            "fixture_loss_before": -0.1706437805591568,
            "fixture_loss_after": -0.1711109879288819,
            "policy_rollouts_per_group": 5,
            "references_per_group": 1,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq5_policy_only_sample_statistics": true,
            "zero_step_reference_warmup": true,
            "reference_uses_same_clipped_surrogate": true,
            "old_and_advantages_frozen": true,
            "padding_zero_gradient": true,
            "real_optimizer_step_decreases_fixture": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "taskpress": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "score_reference_error": 0.0,
            "additive_negative_control_gap": 1.8376675552848294,
            "original_tokens": 11,
            "context_tokens": 8,
            "guide_tokens": 3,
            "retained_tokens_per_head": 4,
            "kv_heads": 2,
            "value_scale_reference_error": 0.0,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq1_3_independent_reference": true,
            "per_head_topk_and_original_positions": true,
            "guide_tokens_discarded": true,
            "actual_kv_gather": true,
            "multiplicative_not_additive": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "score_reference_error": 0.0,
            "additive_negative_control_gap": 1.8477973703557136,
            "original_tokens": 11,
            "context_tokens": 8,
            "guide_tokens": 3,
            "retained_tokens_per_head": 4,
            "kv_heads": 2,
            "value_scale_reference_error": 0.0,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq1_3_independent_reference": true,
            "per_head_topk_and_original_positions": true,
            "guide_tokens_discarded": true,
            "actual_kv_gather": true,
            "multiplicative_not_additive": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "score_reference_error": 0.0,
            "additive_negative_control_gap": 1.869071894882368,
            "original_tokens": 11,
            "context_tokens": 8,
            "guide_tokens": 3,
            "retained_tokens_per_head": 4,
            "kv_heads": 2,
            "value_scale_reference_error": 0.0,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq1_3_independent_reference": true,
            "per_head_topk_and_original_positions": true,
            "guide_tokens_discarded": true,
            "actual_kv_gather": true,
            "multiplicative_not_additive": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "combinatorial-watermark": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "global_reference_error": 0.0,
            "local_reference_error": 0.0,
            "clean_global_score": 1.0,
            "edited_global_score": 0.9344262295081968,
            "flagged_interior_positions": 7,
            "target_tag_mass_gain": 0.4527122670734478,
            "tokens": 64,
            "window": 4,
            "pattern_period": 8,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq4_global_matches_phase_oracle": true,
            "eq5_local_matches_covering_window_oracle": true,
            "cyclic_phase_invariance": true,
            "clean_hard_pattern_no_interior_false_alarm": true,
            "single_edit_flags_its_interior_position": true,
            "logit_shift_increases_target_tag_mass": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "global_reference_error": 0.0,
            "local_reference_error": 0.0,
            "clean_global_score": 1.0,
            "edited_global_score": 0.9344262295081968,
            "flagged_interior_positions": 7,
            "target_tag_mass_gain": 0.44825000262980125,
            "tokens": 64,
            "window": 4,
            "pattern_period": 8,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq4_global_matches_phase_oracle": true,
            "eq5_local_matches_covering_window_oracle": true,
            "cyclic_phase_invariance": true,
            "clean_hard_pattern_no_interior_false_alarm": true,
            "single_edit_flags_its_interior_position": true,
            "logit_shift_increases_target_tag_mass": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "global_reference_error": 0.0,
            "local_reference_error": 0.0,
            "clean_global_score": 1.0,
            "edited_global_score": 0.9344262295081968,
            "flagged_interior_positions": 7,
            "target_tag_mass_gain": 0.4610471979072531,
            "tokens": 64,
            "window": 4,
            "pattern_period": 8,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq4_global_matches_phase_oracle": true,
            "eq5_local_matches_covering_window_oracle": true,
            "cyclic_phase_invariance": true,
            "clean_hard_pattern_no_interior_false_alarm": true,
            "single_edit_flags_its_interior_position": true,
            "logit_shift_increases_target_tag_mass": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "interpolated-null-calibration": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "decomposition_error": 8.326672684688674e-17,
            "gradient_error": 0.0,
            "endpoint_error": 0.0,
            "bare_null_negative_control_gap": 0.9814010535412249,
            "queries": 3,
            "documents_per_query": 6,
            "passes": 3,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq6_eq7_decomposition": true,
            "exact_three_pass_gradient": true,
            "zero_and_one_endpoints": true,
            "lambda0_is_not_no_calibration": true,
            "instruction_footprint_can_change_ranking": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "decomposition_error": 1.1102230246251565e-16,
            "gradient_error": 0.0,
            "endpoint_error": 0.0,
            "bare_null_negative_control_gap": 0.9934973542330007,
            "queries": 3,
            "documents_per_query": 6,
            "passes": 3,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq6_eq7_decomposition": true,
            "exact_three_pass_gradient": true,
            "zero_and_one_endpoints": true,
            "lambda0_is_not_no_calibration": true,
            "instruction_footprint_can_change_ranking": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "decomposition_error": 1.1102230246251565e-16,
            "gradient_error": 0.0,
            "endpoint_error": 0.0,
            "bare_null_negative_control_gap": 0.9664086363981846,
            "queries": 3,
            "documents_per_query": 6,
            "passes": 3,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq6_eq7_decomposition": true,
            "exact_three_pass_gradient": true,
            "zero_and_one_endpoints": true,
            "lambda0_is_not_no_calibration": true,
            "instruction_footprint_can_change_ranking": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "xlgobench": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "scorer_positive_controls": 5,
            "scorer_wrong_controls_accepted": 0,
            "blind_fixture_model_correct": 0,
            "model_payloads_without_gold": 5,
            "task_families": 5,
            "pretrained_model_calls": 0,
            "human_validated_translation_templates_executed": 0
          },
          "checks": {
            "all_five_exact_algorithmic_scorers": true,
            "probabilistic_top8_order_invariant": true,
            "no_gold_in_model_payload": true,
            "callback_cannot_mutate_scorer_case": true,
            "wrong_model_not_reported_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "scorer_positive_controls": 5,
            "scorer_wrong_controls_accepted": 0,
            "blind_fixture_model_correct": 0,
            "model_payloads_without_gold": 5,
            "task_families": 5,
            "pretrained_model_calls": 0,
            "human_validated_translation_templates_executed": 0
          },
          "checks": {
            "all_five_exact_algorithmic_scorers": true,
            "probabilistic_top8_order_invariant": true,
            "no_gold_in_model_payload": true,
            "callback_cannot_mutate_scorer_case": true,
            "wrong_model_not_reported_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "scorer_positive_controls": 5,
            "scorer_wrong_controls_accepted": 0,
            "blind_fixture_model_correct": 0,
            "model_payloads_without_gold": 5,
            "task_families": 5,
            "pretrained_model_calls": 0,
            "human_validated_translation_templates_executed": 0
          },
          "checks": {
            "all_five_exact_algorithmic_scorers": true,
            "probabilistic_top8_order_invariant": true,
            "no_gold_in_model_payload": true,
            "callback_cannot_mutate_scorer_case": true,
            "wrong_model_not_reported_success": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "mechanistic-etm": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "binomial_reference_error": 1.7763568394002505e-15,
            "topic_gradient_norm": 2.7934405937960913,
            "fixture_loss_before": 20.800021921856455,
            "fixture_loss_after": 20.766294384496234,
            "steering_projection_error": 8.881784197001252e-16,
            "steering_perpendicular_error": 4.440892098500626e-16,
            "encoder_decoder_swap_negative_control_gap": 0.016370650502114892,
            "documents": 4,
            "sae_features": 7,
            "topics": 3,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq12_binomial_likelihood_reference": true,
            "frozen_sae_matrices": true,
            "eq11_uses_encoder_not_decoder": true,
            "real_topic_and_variational_update": true,
            "eq16_unit_topic_direction": true,
            "eq19_replaces_parallel_preserves_perpendicular": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "binomial_reference_error": 1.7763568394002505e-15,
            "topic_gradient_norm": 4.583914243907599,
            "fixture_loss_before": 14.673811008174933,
            "fixture_loss_after": 14.65655847343407,
            "steering_projection_error": 8.881784197001252e-16,
            "steering_perpendicular_error": 4.440892098500626e-16,
            "encoder_decoder_swap_negative_control_gap": 0.030617640707011173,
            "documents": 4,
            "sae_features": 7,
            "topics": 3,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq12_binomial_likelihood_reference": true,
            "frozen_sae_matrices": true,
            "eq11_uses_encoder_not_decoder": true,
            "real_topic_and_variational_update": true,
            "eq16_unit_topic_direction": true,
            "eq19_replaces_parallel_preserves_perpendicular": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "binomial_reference_error": 0.0,
            "topic_gradient_norm": 1.9077225337556702,
            "fixture_loss_before": 17.557194336955188,
            "fixture_loss_after": 17.522493755567762,
            "steering_projection_error": 4.440892098500626e-16,
            "steering_perpendicular_error": 5.551115123125783e-16,
            "encoder_decoder_swap_negative_control_gap": 0.02975313221229703,
            "documents": 4,
            "sae_features": 7,
            "topics": 3,
            "optimizer_steps": 1,
            "pretrained_model_calls": 0
          },
          "checks": {
            "eq12_binomial_likelihood_reference": true,
            "frozen_sae_matrices": true,
            "eq11_uses_encoder_not_decoder": true,
            "real_topic_and_variational_update": true,
            "eq16_unit_topic_direction": true,
            "eq19_replaces_parallel_preserves_perpendicular": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    }
  },
  "elapsed_seconds": 1.9950369582511485,
  "evaluation_tier": "l1_mechanism",
  "diagnostic_only": true,
  "formal_comparison": false,
  "promotion_eligible": false,
  "evolve_integrated": false,
  "pretrained_model_calls": 0,
  "full_experiments_executed": false
}
