{
  "schema": "biosim.reference_calibration_evidence_v2",
  "status": "experiment_completed_not_production_qualified",
  "result": {
    "status": "completed_fixed_calibration",
    "model_qualified": false,
    "epochs": 3,
    "training_pairs": 3271,
    "validation_pairs": 825,
    "steps": 9813,
    "baseline_validation": {
      "omol25_wb97m_v_def2_tzvpd_full_qm": {
        "records": 825,
        "total_energy_rmse_ev": 0.5092603809316375,
        "force_component_rmse_ev_a": 0.09438214406179507
      },
      "spice2_wb97m_d3bj_def2_tzvppd_full_qm": {
        "records": 825,
        "total_energy_rmse_ev": 0.5031083693755897,
        "force_component_rmse_ev_a": 0.08526145838574653
      }
    },
    "final_validation": {
      "omol25_wb97m_v_def2_tzvpd_full_qm": {
        "records": 825,
        "total_energy_rmse_ev": 0.4178788682715482,
        "force_component_rmse_ev_a": 0.08609551949689541
      },
      "spice2_wb97m_d3bj_def2_tzvppd_full_qm": {
        "records": 825,
        "total_energy_rmse_ev": 0.43044722383405654,
        "force_component_rmse_ev_a": 0.085405349140361
      }
    },
    "backbone_unchanged": true,
    "elapsed_seconds": 782.5941306520253,
    "peak_gpu_allocated_bytes": 233576960,
    "bindings": {
      "protocol_sha256": "a8fc2371d5d7861ec66c5be88c94ea4123221fb3603c7ec1cecbda0f423a0a40",
      "audit_sha256": "c253dd4624d9ddbfd73c8058fbbcb4cdada5f82b2f7a706abaeb216a158a1224"
    },
    "energy_gauge": [
      {
        "reference": "omol25_wb97m_v_def2_tzvpd_full_qm",
        "coefficients_ev": [
          -7.511722089966133,
          -1032.8635766265493,
          -1487.0105700474778,
          -2044.923615866059,
          -2714.727480205714,
          -9285.427760604922,
          -10833.20040786562,
          -12519.608128205953,
          -70040.88857367433,
          -8098.419207489828,
          4.922913904071404
        ],
        "rank": 11,
        "singular_values": [
          1221.8556308818663,
          192.46043907819876,
          111.09831945681542,
          104.68555015887597,
          83.22809891245795,
          51.8205944974085,
          44.88774872500938,
          29.196515701508446,
          13.686146699127802,
          10.371574802780929,
          8.023622697650016
        ],
        "training_records": 3271
      },
      {
        "reference": "spice2_wb97m_d3bj_def2_tzvppd_full_qm",
        "coefficients_ev": [
          -7.649358692466827,
          -1033.5349921759898,
          -1487.7991436086145,
          -2045.822425796207,
          -2715.7331562561412,
          -9287.156665843127,
          -10835.028697969727,
          -12521.537786536366,
          -70044.85467368834,
          -8101.0122811143365,
          4.971276303743153
        ],
        "rank": 11,
        "singular_values": [
          1221.8556308818663,
          192.46043907819876,
          111.09831945681542,
          104.68555015887597,
          83.22809891245795,
          51.8205944974085,
          44.88774872500938,
          29.196515701508446,
          13.686146699127802,
          10.371574802780929,
          8.023622697650016
        ],
        "training_records": 3271
      }
    ],
    "limitations": [
      "Auxiliary full-QM reference calibration only; solution backbone unchanged",
      "No physical rate, spin generalisation or globally unseen chemistry claim",
      "No label averaging, source split changes or validation-based gauge fitting"
    ]
  },
  "independent_audit": {
    "status": "passed_independent_saved_result_audit",
    "pairs": 4096,
    "train_pairs": 3271,
    "validation_pairs": 825,
    "metrics": {
      "omol25_wb97m_v_def2_tzvpd_full_qm": {
        "baseline": {
          "records": 825,
          "total_energy_rmse_ev": 0.5092603809316624,
          "force_component_rmse_ev_a": 0.09438214406179507
        },
        "final": {
          "records": 825,
          "total_energy_rmse_ev": 0.4178788682715516,
          "force_component_rmse_ev_a": 0.08609551949689541
        },
        "energy_error_change_percent": -17.943966599744833,
        "force_error_change_percent": -8.779864716226548,
        "configurations_with_lower_force_rmse": 764
      },
      "spice2_wb97m_d3bj_def2_tzvppd_full_qm": {
        "baseline": {
          "records": 825,
          "total_energy_rmse_ev": 0.5031083693756057,
          "force_component_rmse_ev_a": 0.08526145486777653
        },
        "final": {
          "records": 825,
          "total_energy_rmse_ev": 0.4304472238341342,
          "force_component_rmse_ev_a": 0.08540534619894162
        },
        "energy_error_change_percent": -14.44244420573827,
        "force_error_change_percent": 0.16876480865619659,
        "configurations_with_lower_force_rmse": 297
      }
    },
    "training_only_gauge_max_reproduction_errors_ev": [
      4.3655745685100555e-11,
      6.002665031701326e-11
    ],
    "backbone_tensors_exactly_unchanged": true,
    "checkpoint_completed_steps": 9813,
    "checks": [
      "Native targets reconstructed independently; both method-specific error tables reproduce",
      "Only training rows reproduce the full-rank numerical energy gauges",
      "Complete validation IDs and component disjointness verified",
      "All original backbone tensors exactly match released weights",
      "Checkpoint head weights finite; optimizer recorded all 9813 updates"
    ],
    "sha256": {
      "checker": "1074a90a0d751a522cb6940834e5bce3a75d9fb6dbbe448661d0679dcbf8d1a2",
      "review": "ae08fa17acb2b29b47ccf8dcc74960e50e0cbac581307c172d2067e28c0a268e",
      "checkpoint": "51c2fcc8b5cb0c524ea10957c250dc28e81bd505366463793c823116ede62d02",
      "result_archive": "d348e7fb3c33fea979c8e760d607e95eacb891dbc59aca55fdc749cad9c51a97",
      "cohort_audit": "c253dd4624d9ddbfd73c8058fbbcb4cdada5f82b2f7a706abaeb216a158a1224"
    },
    "elapsed_seconds": 4.960134371183813,
    "model_qualified": false,
    "scope": "Audit of saved outputs and checkpoint; no extra model fitting/inference or rate qualification."
  },
  "source_counts": {
    "paired_geometries": 4096,
    "method_specific_records": 8192,
    "training_pairs": 3271,
    "validation_pairs": 825,
    "linked_molecular_components": 2783
  },
  "provider": {
    "gpu": "RTX 2000 Ada",
    "count": 1,
    "quoted_hourly_rate_usd": 0.24,
    "pod_stop_confirmed": true,
    "provider_state_independently_checked": "EXITED",
    "operator_receipt_sha256": "e6df44bb0deb8fc4a6d6bf757ffd7b51cecdc139b546b0183864606401ed1909"
  },
  "scope": "Paired full-QM PubChem calibration across two reference methods; native labels remain distinct. Existing BMS solution backbone unchanged. No production dynamics or reaction rates validated."
}