{
  "schema": "latent_mining_bio_site_summary_v2",
  "generated": "2026-06-04",
  "claim_boundary": "This artifact stratifies the locked biology subset as cross-locus regulatory triage. It does not claim broad biological transfer, wet-lab validity, cross-oracle validity, or broad task-family coverage.",
  "release_manifest": {
    "schema": "latent_mining_v1_public_bundle_manifest",
    "status": "pass",
    "task_count": 165,
    "scan_finding_count": 0,
    "copied_file_count": 2490,
    "task_file_count": 2475,
    "included_roots": [
      "benchmark_audit",
      "benchmark_corpus_v1",
      "launch_readiness",
      "manifest",
      "tasks"
    ],
    "excluded_roots": [
      "internal construction artifacts",
      "candidate pool internals",
      "hidden verifier files",
      "grading tests",
      "solutions",
      "raw calibration traces"
    ],
    "classification": "release_safe_public_bundle_manifest"
  },
  "reconciliation": {
    "expected_agent_count": 6,
    "expected_run_count": 990,
    "expected_task_count": 165,
    "observed_run_count": 990,
    "observed_task_count": 165,
    "outcome_total": 990,
    "public_bundle_scope": "release-safe aggregate table, not raw traces",
    "static_audit_status": "static_audit_passed",
    "status": "pass",
    "unique_task_agent_pair_count": 990
  },
  "outcomes": {
    "scientific_search_failure": 626,
    "solve": 112,
    "valid_below_threshold_near_miss": 252
  },
  "assay_categories": [
    {
      "axis": "assay_context",
      "biology_system_count": 2,
      "capability_axis": "reason over perturbations expected to change regulatory accessibility",
      "failure_rate": 0.6,
      "locus_count": 2,
      "near_miss_rate": 0.2,
      "outcomes": {
        "scientific_search_failure": 108,
        "solve": 36,
        "valid_below_threshold_near_miss": 36
      },
      "representative_loci": [
        "fto_irx3_adipose",
        "il2ra_tcell"
      ],
      "run_count": 180,
      "solve_rate": 0.2,
      "task_category": "chromatin-accessibility triage",
      "task_count": 30,
      "value": "ATAC",
      "workflow_grounding": "prioritizing noncoding edits for open-chromatin regulatory follow-up",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "assay_context",
      "biology_system_count": 2,
      "capability_axis": "reason over promoter/enhancer perturbations tied to transcription-start output",
      "failure_rate": 0.5167,
      "locus_count": 2,
      "near_miss_rate": 0.4278,
      "outcomes": {
        "scientific_search_failure": 93,
        "solve": 10,
        "valid_below_threshold_near_miss": 77
      },
      "representative_loci": [
        "pcsk9_liver",
        "tert_promoter"
      ],
      "run_count": 180,
      "solve_rate": 0.0556,
      "task_category": "promoter-output triage",
      "task_count": 30,
      "value": "CAGE",
      "workflow_grounding": "prioritizing promoter-proximal regulatory interventions",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "assay_context",
      "biology_system_count": 4,
      "capability_axis": "reason over motif, TF-bound, histone, and regulatory-element evidence",
      "failure_rate": 0.65,
      "locus_count": 4,
      "near_miss_rate": 0.225,
      "outcomes": {
        "scientific_search_failure": 234,
        "solve": 45,
        "valid_below_threshold_near_miss": 81
      },
      "representative_loci": [
        "bcl11a_erythroid",
        "gata1_chrx",
        "hbb_lcr",
        "myc_8q24"
      ],
      "run_count": 360,
      "solve_rate": 0.125,
      "task_category": "binding/regulatory-element triage",
      "task_count": 60,
      "value": "ChIP",
      "workflow_grounding": "prioritizing edits around TF or chromatin regulatory evidence",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "assay_context",
      "biology_system_count": 3,
      "capability_axis": "reason over public sequence evidence for downstream transcript-level effects",
      "failure_rate": 0.7074,
      "locus_count": 3,
      "near_miss_rate": 0.2148,
      "outcomes": {
        "scientific_search_failure": 191,
        "solve": 21,
        "valid_below_threshold_near_miss": 58
      },
      "representative_loci": [
        "cftr_epithelial",
        "ins_islet",
        "lct_mcm6"
      ],
      "run_count": 270,
      "solve_rate": 0.0778,
      "task_category": "transcript-output triage",
      "task_count": 45,
      "value": "RNA_SEQ",
      "workflow_grounding": "prioritizing noncoding edits for regulatory-expression effect",
      "cohort_id": "current_six_agent_990"
    }
  ],
  "biology_systems": [
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "immune-cell regulatory intervention triage",
      "failure_rate": 0.6,
      "locus_count": 1,
      "near_miss_rate": 0.2,
      "outcomes": {
        "scientific_search_failure": 54,
        "solve": 18,
        "valid_below_threshold_near_miss": 18
      },
      "representative_loci": [
        "il2ra_tcell"
      ],
      "run_count": 90,
      "solve_rate": 0.2,
      "task_category": "autoimmunity",
      "task_count": 15,
      "value": "autoimmunity",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "oncogenic regulatory-control triage",
      "failure_rate": 0.7667,
      "locus_count": 1,
      "near_miss_rate": 0.1,
      "outcomes": {
        "scientific_search_failure": 69,
        "solve": 12,
        "valid_below_threshold_near_miss": 9
      },
      "representative_loci": [
        "myc_8q24"
      ],
      "run_count": 90,
      "solve_rate": 0.1333,
      "task_category": "cancer proliferation",
      "task_count": 15,
      "value": "cancer proliferation",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "promoter and telomere-regulatory triage",
      "failure_rate": 0.3667,
      "locus_count": 1,
      "near_miss_rate": 0.5667,
      "outcomes": {
        "scientific_search_failure": 33,
        "solve": 6,
        "valid_below_threshold_near_miss": 51
      },
      "representative_loci": [
        "tert_promoter"
      ],
      "run_count": 90,
      "solve_rate": 0.0667,
      "task_category": "cancer/aging",
      "task_count": 15,
      "value": "cancer/aging",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "metabolic gene-output triage",
      "failure_rate": 0.6667,
      "locus_count": 1,
      "near_miss_rate": 0.2889,
      "outcomes": {
        "scientific_search_failure": 60,
        "solve": 4,
        "valid_below_threshold_near_miss": 26
      },
      "representative_loci": [
        "pcsk9_liver"
      ],
      "run_count": 90,
      "solve_rate": 0.0444,
      "task_category": "cholesterol metabolism",
      "task_count": 15,
      "value": "cholesterol metabolism",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "epithelial transport regulatory triage",
      "failure_rate": 0.6889,
      "locus_count": 1,
      "near_miss_rate": 0.2778,
      "outcomes": {
        "scientific_search_failure": 62,
        "solve": 3,
        "valid_below_threshold_near_miss": 25
      },
      "representative_loci": [
        "cftr_epithelial"
      ],
      "run_count": 90,
      "solve_rate": 0.0333,
      "task_category": "cystic fibrosis/epithelial transport",
      "task_count": 15,
      "value": "cystic fibrosis/epithelial transport",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "endocrine-cell regulatory-output triage",
      "failure_rate": 0.7444,
      "locus_count": 1,
      "near_miss_rate": 0.1889,
      "outcomes": {
        "scientific_search_failure": 67,
        "solve": 6,
        "valid_below_threshold_near_miss": 17
      },
      "representative_loci": [
        "ins_islet"
      ],
      "run_count": 90,
      "solve_rate": 0.0667,
      "task_category": "diabetes/endocrine",
      "task_count": 15,
      "value": "diabetes/endocrine",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "blood-lineage regulatory-element triage",
      "failure_rate": 0.6222,
      "locus_count": 1,
      "near_miss_rate": 0.3,
      "outcomes": {
        "scientific_search_failure": 56,
        "solve": 7,
        "valid_below_threshold_near_miss": 27
      },
      "representative_loci": [
        "gata1_chrx"
      ],
      "run_count": 90,
      "solve_rate": 0.0778,
      "task_category": "hematopoiesis",
      "task_count": 15,
      "value": "hematopoiesis",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "erythroid enhancer and silencing triage",
      "failure_rate": 0.6444,
      "locus_count": 1,
      "near_miss_rate": 0.2778,
      "outcomes": {
        "scientific_search_failure": 58,
        "solve": 7,
        "valid_below_threshold_near_miss": 25
      },
      "representative_loci": [
        "bcl11a_erythroid"
      ],
      "run_count": 90,
      "solve_rate": 0.0778,
      "task_category": "hemoglobin switching",
      "task_count": 15,
      "value": "hemoglobin switching",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "locus-control-region regulatory triage",
      "failure_rate": 0.5667,
      "locus_count": 1,
      "near_miss_rate": 0.2222,
      "outcomes": {
        "scientific_search_failure": 51,
        "solve": 19,
        "valid_below_threshold_near_miss": 20
      },
      "representative_loci": [
        "hbb_lcr"
      ],
      "run_count": 90,
      "solve_rate": 0.2111,
      "task_category": "hemoglobinopathy",
      "task_count": 15,
      "value": "hemoglobinopathy",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "developmental digestive regulatory triage",
      "failure_rate": 0.6889,
      "locus_count": 1,
      "near_miss_rate": 0.1778,
      "outcomes": {
        "scientific_search_failure": 62,
        "solve": 12,
        "valid_below_threshold_near_miss": 16
      },
      "representative_loci": [
        "lct_mcm6"
      ],
      "run_count": 90,
      "solve_rate": 0.1333,
      "task_category": "lactase regulation/digestion",
      "task_count": 15,
      "value": "lactase regulation/digestion",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "biology_system",
      "biology_system_count": 1,
      "capability_axis": "metabolic enhancer regulatory triage",
      "failure_rate": 0.6,
      "locus_count": 1,
      "near_miss_rate": 0.2,
      "outcomes": {
        "scientific_search_failure": 54,
        "solve": 18,
        "valid_below_threshold_near_miss": 18
      },
      "representative_loci": [
        "fto_irx3_adipose"
      ],
      "run_count": 90,
      "solve_rate": 0.2,
      "task_category": "obesity/metabolism",
      "task_count": 15,
      "value": "obesity/metabolism",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    }
  ],
  "loci": [
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6444,
      "locus_count": 1,
      "near_miss_rate": 0.2778,
      "outcomes": {
        "scientific_search_failure": 58,
        "solve": 7,
        "valid_below_threshold_near_miss": 25
      },
      "representative_loci": [
        "bcl11a_erythroid"
      ],
      "run_count": 90,
      "solve_rate": 0.0778,
      "task_category": "bcl11a_erythroid",
      "task_count": 15,
      "value": "bcl11a_erythroid",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6889,
      "locus_count": 1,
      "near_miss_rate": 0.2778,
      "outcomes": {
        "scientific_search_failure": 62,
        "solve": 3,
        "valid_below_threshold_near_miss": 25
      },
      "representative_loci": [
        "cftr_epithelial"
      ],
      "run_count": 90,
      "solve_rate": 0.0333,
      "task_category": "cftr_epithelial",
      "task_count": 15,
      "value": "cftr_epithelial",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6,
      "locus_count": 1,
      "near_miss_rate": 0.2,
      "outcomes": {
        "scientific_search_failure": 54,
        "solve": 18,
        "valid_below_threshold_near_miss": 18
      },
      "representative_loci": [
        "fto_irx3_adipose"
      ],
      "run_count": 90,
      "solve_rate": 0.2,
      "task_category": "fto_irx3_adipose",
      "task_count": 15,
      "value": "fto_irx3_adipose",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6222,
      "locus_count": 1,
      "near_miss_rate": 0.3,
      "outcomes": {
        "scientific_search_failure": 56,
        "solve": 7,
        "valid_below_threshold_near_miss": 27
      },
      "representative_loci": [
        "gata1_chrx"
      ],
      "run_count": 90,
      "solve_rate": 0.0778,
      "task_category": "gata1_chrx",
      "task_count": 15,
      "value": "gata1_chrx",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.5667,
      "locus_count": 1,
      "near_miss_rate": 0.2222,
      "outcomes": {
        "scientific_search_failure": 51,
        "solve": 19,
        "valid_below_threshold_near_miss": 20
      },
      "representative_loci": [
        "hbb_lcr"
      ],
      "run_count": 90,
      "solve_rate": 0.2111,
      "task_category": "hbb_lcr",
      "task_count": 15,
      "value": "hbb_lcr",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6,
      "locus_count": 1,
      "near_miss_rate": 0.2,
      "outcomes": {
        "scientific_search_failure": 54,
        "solve": 18,
        "valid_below_threshold_near_miss": 18
      },
      "representative_loci": [
        "il2ra_tcell"
      ],
      "run_count": 90,
      "solve_rate": 0.2,
      "task_category": "il2ra_tcell",
      "task_count": 15,
      "value": "il2ra_tcell",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.7444,
      "locus_count": 1,
      "near_miss_rate": 0.1889,
      "outcomes": {
        "scientific_search_failure": 67,
        "solve": 6,
        "valid_below_threshold_near_miss": 17
      },
      "representative_loci": [
        "ins_islet"
      ],
      "run_count": 90,
      "solve_rate": 0.0667,
      "task_category": "ins_islet",
      "task_count": 15,
      "value": "ins_islet",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6889,
      "locus_count": 1,
      "near_miss_rate": 0.1778,
      "outcomes": {
        "scientific_search_failure": 62,
        "solve": 12,
        "valid_below_threshold_near_miss": 16
      },
      "representative_loci": [
        "lct_mcm6"
      ],
      "run_count": 90,
      "solve_rate": 0.1333,
      "task_category": "lct_mcm6",
      "task_count": 15,
      "value": "lct_mcm6",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.7667,
      "locus_count": 1,
      "near_miss_rate": 0.1,
      "outcomes": {
        "scientific_search_failure": 69,
        "solve": 12,
        "valid_below_threshold_near_miss": 9
      },
      "representative_loci": [
        "myc_8q24"
      ],
      "run_count": 90,
      "solve_rate": 0.1333,
      "task_category": "myc_8q24",
      "task_count": 15,
      "value": "myc_8q24",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.6667,
      "locus_count": 1,
      "near_miss_rate": 0.2889,
      "outcomes": {
        "scientific_search_failure": 60,
        "solve": 4,
        "valid_below_threshold_near_miss": 26
      },
      "representative_loci": [
        "pcsk9_liver"
      ],
      "run_count": 90,
      "solve_rate": 0.0444,
      "task_category": "pcsk9_liver",
      "task_count": 15,
      "value": "pcsk9_liver",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    },
    {
      "axis": "locus",
      "biology_system_count": 1,
      "capability_axis": "regulatory triage",
      "failure_rate": 0.3667,
      "locus_count": 1,
      "near_miss_rate": 0.5667,
      "outcomes": {
        "scientific_search_failure": 33,
        "solve": 6,
        "valid_below_threshold_near_miss": 51
      },
      "representative_loci": [
        "tert_promoter"
      ],
      "run_count": 90,
      "solve_rate": 0.0667,
      "task_category": "tert_promoter",
      "task_count": 15,
      "value": "tert_promoter",
      "workflow_grounding": "cross-locus regulatory intervention triage",
      "cohort_id": "current_six_agent_990"
    }
  ],
  "near_miss_review_counts": {
    "good_biological_near_miss": 32,
    "mixed_good_near_miss_with_proxy_attraction": 153,
    "public_proxy_overfit_near_miss": 55,
    "uncertain": 8,
    "outcome_near_process_poor": 4
  },
  "trace_quality": {
    "raw_trace_policy": "raw solver traces and grading diagnostics remain internal evidence",
    "row_count": 990,
    "public_interface": {
      "errors": 0,
      "runner_failures": 0,
      "timeouts": 0,
      "valid": 990
    },
    "scope": "990 retained runs; timeout and validity counts exclude rerun attempts."
  },
  "legacy_three_cli_495": {
    "cohort_id": "legacy_three_cli_495",
    "agents": [
      "codex_gpt55",
      "claude_opus47",
      "claude_sonnet46"
    ],
    "run_count": 495,
    "task_count": 165,
    "outcomes": {
      "solve": 53,
      "valid_below_threshold_near_miss": 138,
      "scientific_search_failure": 304
    },
    "source": "Unchanged frozen release bundle: benchmark_corpus_v1/bio_task_category_stratification.json and behavior_codebook.md.",
    "assay_categories": [
      {
        "axis": "assay_context",
        "biology_system_count": 2,
        "capability_axis": "reason over perturbations expected to change regulatory accessibility",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5889,
        "locus_count": 2,
        "near_miss_rate": 0.2222,
        "outcomes": {
          "scientific_search_failure": 53,
          "solve": 17,
          "valid_below_threshold_near_miss": 20
        },
        "representative_loci": [
          "fto_irx3_adipose",
          "il2ra_tcell"
        ],
        "run_count": 90,
        "solve_rate": 0.1889,
        "task_category": "chromatin-accessibility triage",
        "task_count": 30,
        "value": "ATAC",
        "workflow_grounding": "prioritizing noncoding edits for open-chromatin regulatory follow-up"
      },
      {
        "axis": "assay_context",
        "biology_system_count": 2,
        "capability_axis": "reason over promoter/enhancer perturbations tied to transcription-start output",
        "dominant_failure_mode": "near-miss behavior suggests partial evidence integration but weak final contrast",
        "failure_rate": 0.4333,
        "locus_count": 2,
        "near_miss_rate": 0.5,
        "outcomes": {
          "scientific_search_failure": 39,
          "solve": 6,
          "valid_below_threshold_near_miss": 45
        },
        "representative_loci": [
          "pcsk9_liver",
          "tert_promoter"
        ],
        "run_count": 90,
        "solve_rate": 0.0667,
        "task_category": "promoter-output triage",
        "task_count": 30,
        "value": "CAGE",
        "workflow_grounding": "prioritizing promoter-proximal regulatory interventions"
      },
      {
        "axis": "assay_context",
        "biology_system_count": 4,
        "capability_axis": "reason over motif, TF-bound, histone, and regulatory-element evidence",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.65,
        "locus_count": 4,
        "near_miss_rate": 0.2444,
        "outcomes": {
          "scientific_search_failure": 117,
          "solve": 19,
          "valid_below_threshold_near_miss": 44
        },
        "representative_loci": [
          "bcl11a_erythroid",
          "gata1_chrx",
          "hbb_lcr",
          "myc_8q24"
        ],
        "run_count": 180,
        "solve_rate": 0.1056,
        "task_category": "binding/regulatory-element triage",
        "task_count": 60,
        "value": "ChIP",
        "workflow_grounding": "prioritizing edits around TF or chromatin regulatory evidence"
      },
      {
        "axis": "assay_context",
        "biology_system_count": 3,
        "capability_axis": "reason over public sequence evidence for downstream transcript-level effects",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7037,
        "locus_count": 3,
        "near_miss_rate": 0.2148,
        "outcomes": {
          "scientific_search_failure": 95,
          "solve": 11,
          "valid_below_threshold_near_miss": 29
        },
        "representative_loci": [
          "cftr_epithelial",
          "ins_islet",
          "lct_mcm6"
        ],
        "run_count": 135,
        "solve_rate": 0.0815,
        "task_category": "transcript-output triage",
        "task_count": 45,
        "value": "RNA_SEQ",
        "workflow_grounding": "prioritizing noncoding edits for regulatory-expression effect"
      }
    ],
    "biology_systems": [
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "immune-cell regulatory intervention triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6,
        "locus_count": 1,
        "near_miss_rate": 0.2667,
        "outcomes": {
          "scientific_search_failure": 27,
          "solve": 6,
          "valid_below_threshold_near_miss": 12
        },
        "representative_loci": [
          "il2ra_tcell"
        ],
        "run_count": 45,
        "solve_rate": 0.1333,
        "task_category": "autoimmunity",
        "task_count": 15,
        "value": "autoimmunity",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "oncogenic regulatory-control triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7333,
        "locus_count": 1,
        "near_miss_rate": 0.1333,
        "outcomes": {
          "scientific_search_failure": 33,
          "solve": 6,
          "valid_below_threshold_near_miss": 6
        },
        "representative_loci": [
          "myc_8q24"
        ],
        "run_count": 45,
        "solve_rate": 0.1333,
        "task_category": "cancer proliferation",
        "task_count": 15,
        "value": "cancer proliferation",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "promoter and telomere-regulatory triage",
        "dominant_failure_mode": "near-miss behavior suggests partial evidence integration but weak final contrast",
        "failure_rate": 0.2889,
        "locus_count": 1,
        "near_miss_rate": 0.6222,
        "outcomes": {
          "scientific_search_failure": 13,
          "solve": 4,
          "valid_below_threshold_near_miss": 28
        },
        "representative_loci": [
          "tert_promoter"
        ],
        "run_count": 45,
        "solve_rate": 0.0889,
        "task_category": "cancer/aging",
        "task_count": 15,
        "value": "cancer/aging",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "metabolic gene-output triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5778,
        "locus_count": 1,
        "near_miss_rate": 0.3778,
        "outcomes": {
          "scientific_search_failure": 26,
          "solve": 2,
          "valid_below_threshold_near_miss": 17
        },
        "representative_loci": [
          "pcsk9_liver"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "cholesterol metabolism",
        "task_count": 15,
        "value": "cholesterol metabolism",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "epithelial transport regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7111,
        "locus_count": 1,
        "near_miss_rate": 0.2444,
        "outcomes": {
          "scientific_search_failure": 32,
          "solve": 2,
          "valid_below_threshold_near_miss": 11
        },
        "representative_loci": [
          "cftr_epithelial"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "cystic fibrosis/epithelial transport",
        "task_count": 15,
        "value": "cystic fibrosis/epithelial transport",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "endocrine-cell regulatory-output triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6889,
        "locus_count": 1,
        "near_miss_rate": 0.2222,
        "outcomes": {
          "scientific_search_failure": 31,
          "solve": 4,
          "valid_below_threshold_near_miss": 10
        },
        "representative_loci": [
          "ins_islet"
        ],
        "run_count": 45,
        "solve_rate": 0.0889,
        "task_category": "diabetes/endocrine",
        "task_count": 15,
        "value": "diabetes/endocrine",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "blood-lineage regulatory-element triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6889,
        "locus_count": 1,
        "near_miss_rate": 0.2667,
        "outcomes": {
          "scientific_search_failure": 31,
          "solve": 2,
          "valid_below_threshold_near_miss": 12
        },
        "representative_loci": [
          "gata1_chrx"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "hematopoiesis",
        "task_count": 15,
        "value": "hematopoiesis",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "erythroid enhancer and silencing triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6222,
        "locus_count": 1,
        "near_miss_rate": 0.3333,
        "outcomes": {
          "scientific_search_failure": 28,
          "solve": 2,
          "valid_below_threshold_near_miss": 15
        },
        "representative_loci": [
          "bcl11a_erythroid"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "hemoglobin switching",
        "task_count": 15,
        "value": "hemoglobin switching",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "locus-control-region regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5556,
        "locus_count": 1,
        "near_miss_rate": 0.2444,
        "outcomes": {
          "scientific_search_failure": 25,
          "solve": 9,
          "valid_below_threshold_near_miss": 11
        },
        "representative_loci": [
          "hbb_lcr"
        ],
        "run_count": 45,
        "solve_rate": 0.2,
        "task_category": "hemoglobinopathy",
        "task_count": 15,
        "value": "hemoglobinopathy",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "developmental digestive regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7111,
        "locus_count": 1,
        "near_miss_rate": 0.1778,
        "outcomes": {
          "scientific_search_failure": 32,
          "solve": 5,
          "valid_below_threshold_near_miss": 8
        },
        "representative_loci": [
          "lct_mcm6"
        ],
        "run_count": 45,
        "solve_rate": 0.1111,
        "task_category": "lactase regulation/digestion",
        "task_count": 15,
        "value": "lactase regulation/digestion",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "biology_system",
        "biology_system_count": 1,
        "capability_axis": "metabolic enhancer regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5778,
        "locus_count": 1,
        "near_miss_rate": 0.1778,
        "outcomes": {
          "scientific_search_failure": 26,
          "solve": 11,
          "valid_below_threshold_near_miss": 8
        },
        "representative_loci": [
          "fto_irx3_adipose"
        ],
        "run_count": 45,
        "solve_rate": 0.2444,
        "task_category": "obesity/metabolism",
        "task_count": 15,
        "value": "obesity/metabolism",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      }
    ],
    "loci": [
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6222,
        "locus_count": 1,
        "near_miss_rate": 0.3333,
        "outcomes": {
          "scientific_search_failure": 28,
          "solve": 2,
          "valid_below_threshold_near_miss": 15
        },
        "representative_loci": [
          "bcl11a_erythroid"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "bcl11a_erythroid",
        "task_count": 15,
        "value": "bcl11a_erythroid",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7111,
        "locus_count": 1,
        "near_miss_rate": 0.2444,
        "outcomes": {
          "scientific_search_failure": 32,
          "solve": 2,
          "valid_below_threshold_near_miss": 11
        },
        "representative_loci": [
          "cftr_epithelial"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "cftr_epithelial",
        "task_count": 15,
        "value": "cftr_epithelial",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5778,
        "locus_count": 1,
        "near_miss_rate": 0.1778,
        "outcomes": {
          "scientific_search_failure": 26,
          "solve": 11,
          "valid_below_threshold_near_miss": 8
        },
        "representative_loci": [
          "fto_irx3_adipose"
        ],
        "run_count": 45,
        "solve_rate": 0.2444,
        "task_category": "fto_irx3_adipose",
        "task_count": 15,
        "value": "fto_irx3_adipose",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6889,
        "locus_count": 1,
        "near_miss_rate": 0.2667,
        "outcomes": {
          "scientific_search_failure": 31,
          "solve": 2,
          "valid_below_threshold_near_miss": 12
        },
        "representative_loci": [
          "gata1_chrx"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "gata1_chrx",
        "task_count": 15,
        "value": "gata1_chrx",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5556,
        "locus_count": 1,
        "near_miss_rate": 0.2444,
        "outcomes": {
          "scientific_search_failure": 25,
          "solve": 9,
          "valid_below_threshold_near_miss": 11
        },
        "representative_loci": [
          "hbb_lcr"
        ],
        "run_count": 45,
        "solve_rate": 0.2,
        "task_category": "hbb_lcr",
        "task_count": 15,
        "value": "hbb_lcr",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6,
        "locus_count": 1,
        "near_miss_rate": 0.2667,
        "outcomes": {
          "scientific_search_failure": 27,
          "solve": 6,
          "valid_below_threshold_near_miss": 12
        },
        "representative_loci": [
          "il2ra_tcell"
        ],
        "run_count": 45,
        "solve_rate": 0.1333,
        "task_category": "il2ra_tcell",
        "task_count": 15,
        "value": "il2ra_tcell",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.6889,
        "locus_count": 1,
        "near_miss_rate": 0.2222,
        "outcomes": {
          "scientific_search_failure": 31,
          "solve": 4,
          "valid_below_threshold_near_miss": 10
        },
        "representative_loci": [
          "ins_islet"
        ],
        "run_count": 45,
        "solve_rate": 0.0889,
        "task_category": "ins_islet",
        "task_count": 15,
        "value": "ins_islet",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7111,
        "locus_count": 1,
        "near_miss_rate": 0.1778,
        "outcomes": {
          "scientific_search_failure": 32,
          "solve": 5,
          "valid_below_threshold_near_miss": 8
        },
        "representative_loci": [
          "lct_mcm6"
        ],
        "run_count": 45,
        "solve_rate": 0.1111,
        "task_category": "lct_mcm6",
        "task_count": 15,
        "value": "lct_mcm6",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.7333,
        "locus_count": 1,
        "near_miss_rate": 0.1333,
        "outcomes": {
          "scientific_search_failure": 33,
          "solve": 6,
          "valid_below_threshold_near_miss": 6
        },
        "representative_loci": [
          "myc_8q24"
        ],
        "run_count": 45,
        "solve_rate": 0.1333,
        "task_category": "myc_8q24",
        "task_count": 15,
        "value": "myc_8q24",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "public-proxy attraction or single-channel overweighting dominates",
        "failure_rate": 0.5778,
        "locus_count": 1,
        "near_miss_rate": 0.3778,
        "outcomes": {
          "scientific_search_failure": 26,
          "solve": 2,
          "valid_below_threshold_near_miss": 17
        },
        "representative_loci": [
          "pcsk9_liver"
        ],
        "run_count": 45,
        "solve_rate": 0.0444,
        "task_category": "pcsk9_liver",
        "task_count": 15,
        "value": "pcsk9_liver",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      },
      {
        "axis": "locus",
        "biology_system_count": 1,
        "capability_axis": "regulatory triage",
        "dominant_failure_mode": "near-miss behavior suggests partial evidence integration but weak final contrast",
        "failure_rate": 0.2889,
        "locus_count": 1,
        "near_miss_rate": 0.6222,
        "outcomes": {
          "scientific_search_failure": 13,
          "solve": 4,
          "valid_below_threshold_near_miss": 28
        },
        "representative_loci": [
          "tert_promoter"
        ],
        "run_count": 45,
        "solve_rate": 0.0889,
        "task_category": "tert_promoter",
        "task_count": 15,
        "value": "tert_promoter",
        "workflow_grounding": "cross-locus regulatory intervention triage"
      }
    ],
    "trace_quality": {
      "behavior_codebook_hits": {
        "Deciding: selected one edit under uncertainty": 96,
        "Evaluate: checked public constraints/schema": 393,
        "Grounding: inspected public task files/tools": 178,
        "Reasoning: integrated public feature channels": 495,
        "Reflection: recognized hidden/proxy uncertainty": 251
      },
      "failure_pattern_hits": {
        "biological_mechanism_overfit": 109,
        "public_proxy_attractor": 72,
        "repair_target_drift": 172,
        "single_channel_overweighting": 178
      },
      "tool_log_status_counts": {
        "absent_no_tool_use_detected": 323,
        "capture_missing": 172
      },
      "row_count": 495
    },
    "near_miss_process_targets": {
      "review_count": 138,
      "process_reward_target_counts": {
        "make proxy uncertainty explicit": 87,
        "penalize single/composite public-proxy attraction": 2,
        "require sharper rejected-alternative contrast": 36,
        "reward multi-channel evidence integration": 105,
        "reward public biological mechanism grounding": 127
      },
      "decision": {
        "reason": "The near-misses are valid process-level learning signal. The current priority is preserving task validity and using traces to hillclimb search behavior, not weakening the hidden objective or public gates.",
        "tune_task_objectives_to_recover_nonzero_solve_rate": false
      }
    }
  },
  "reconciled": "2026-09-08",
  "cohort_id": "current_six_agent_990",
  "aggregate_provenance": {
    "source": "Published assets/data.js six-agent aggregates; paper tables report model and assay solve rates.",
    "legacy_source": "Frozen public release bundle retains the earlier 495-run three-CLI calibration.",
    "scope": "This reconciliation uses already-public aggregates; no private rows or hidden grading values were exported."
  },
  "models": [
    {
      "id": "codex_gpt55",
      "name": "Codex GPT-5.5",
      "org": "OpenAI",
      "cls": "openai",
      "abbr": "O",
      "status": "complete",
      "solve": 22,
      "near": 55,
      "fail": 88,
      "runs": 165,
      "rate": 0.133,
      "combined_rate": 0.467
    },
    {
      "id": "kimi_k26",
      "name": "Kimi K2.6",
      "org": "Moonshot AI",
      "cls": "moonshot",
      "abbr": "M",
      "status": "complete",
      "solve": 24,
      "near": 38,
      "fail": 103,
      "runs": 165,
      "rate": 0.145,
      "combined_rate": 0.376
    },
    {
      "id": "glm_51",
      "name": "GLM 5.1",
      "org": "Z.ai",
      "cls": "zai",
      "abbr": "Z",
      "status": "complete",
      "solve": 16,
      "near": 45,
      "fail": 104,
      "runs": 165,
      "rate": 0.097,
      "combined_rate": 0.37
    },
    {
      "id": "claude_opus47",
      "name": "Claude Opus 4.7",
      "org": "Anthropic",
      "cls": "anthropic",
      "abbr": "A",
      "status": "complete",
      "solve": 16,
      "near": 43,
      "fail": 106,
      "runs": 165,
      "rate": 0.097,
      "combined_rate": 0.358
    },
    {
      "id": "claude_sonnet46",
      "name": "Claude Sonnet 4.6",
      "org": "Anthropic",
      "cls": "anthropic",
      "abbr": "A",
      "status": "complete",
      "solve": 15,
      "near": 40,
      "fail": 110,
      "runs": 165,
      "rate": 0.091,
      "combined_rate": 0.333
    },
    {
      "id": "gemini_31_pro",
      "name": "Gemini 3.1 Pro",
      "org": "Google",
      "cls": "google",
      "abbr": "G",
      "status": "complete",
      "solve": 19,
      "near": 31,
      "fail": 115,
      "runs": 165,
      "rate": 0.115,
      "combined_rate": 0.303
    }
  ],
  "evaluation_protocol": {
    "dates": "2026-05-31/2026-06-04",
    "grading": "cached model-backed regulatory-effect objective",
    "candidates_per_task": 10,
    "random_choice_strict_solve_baseline": "not reported; candidate count alone does not determine threshold-clearing probability",
    "retention": "Manual reruns of timed-out or interface-invalid attempts; last valid attempt retained.",
    "api_output_token_cap": 131072,
    "api_tool_call_budget": 64,
    "cli_budget": "native harness budgets",
    "sampling": "Gemini and GLM temperature 0; other provider defaults not recorded.",
    "kimi_routing": "121 native Moonshot API runs; 44 Hugging Face Router fallback fills.",
    "source": "latent-mining-paper.pdf Appendix G"
  },
  "near_miss_review_scope": {
    "cohort": "current_six_agent_990",
    "near_miss_count": 252,
    "three_quality_categories_count": 240,
    "public_scalar_collapse": 189,
    "annotation_note": "Scalar-collapse annotations overlap quality categories. Review uses public rationales, sanitized traces, deterministic features, and quote-backed review, including lower-confidence heuristics.",
    "source": "latent-mining-paper.pdf Appendix D"
  }
}