{
  "runs": [
    {
      "ci_run_id": 35856953771,
      "ci_url": "https://github.com/XinyangWuEthz/review-router/actions/runs/35856953771",
      "run_id": "20260923T115155088481Z-human-review-baseline",
      "git_commit": "124b63fe9c8dd1ae7e7e287e87c5f88061e1a8df",
      "git_dirty": false,
      "data_sha256": {
        "train": "bd4084611bd27c939ba98e5e63bc3e5a2c1a4e99477dcba46c829e4c986c429d",
        "test": "c2513ce4abb98c4d1d216e3ca0d4377d57589a0989aa8c06a840509a16c786e8",
        "test_labels": "2a56dcbeba5c05f965a636f56cb5ae972bad60c3b952c239b49be18d7ab70f49"
      },
      "config_sha256": "801efb58457b838a7316359cc89b91670d762379269720fe6d548611f137162d",
      "policy_sha256": "2f8f186c4a0a48213759a2592467bcc73b1995e3902f488abaaa1dcf1fe2fa5a",
      "splits_sha256": "9f86d2a17105abaeff7b3e15bbdf4d983a307a8246b342655558390abb31045c",
      "model_sha256": "63820de571ee75dfe310c1a54c2e8669e573249efeaededd5b451698964a5d34",
      "predictions_sha256": "f9e8d73a52c666083e22f33cb343d02a45ef462c86450198f0ab40fd582a0373",
      "training_source_sha256": {
        "review_router/model.py": "efcb9b4b8c9b61325d054ca17c175416ec3e7e80e8b7e0fe01fba402080a8253",
        "review_router/data.py": "932257ac72a007d6fb31f2bfd400b9504494d46f488fd0562e66142d87431334"
      },
      "seed": 20260922,
      "dependencies": {
        "python": "3.12.14",
        "review_router": "0.1.0",
        "numpy": "2.5.3",
        "pandas": "3.0.6",
        "scikit-learn": "1.9.1",
        "scipy": "1.18.1",
        "pyyaml": "6.0.3"
      },
      "model_config": {
        "max_features": 200000,
        "ngram_max": 2,
        "min_df": 3,
        "C": 4.0,
        "max_iter": 1000,
        "analyzer": "word",
        "char_ngram_min": 2,
        "char_ngram_max": 5,
        "char_max_features": 300000
      },
      "review_workload": {
        "n_total": 63978,
        "n_requires_human_review": 6312,
        "n_priority_review": 2823,
        "n_human_review": 3489,
        "review_fraction": 0.09865891400168808,
        "n_truly_positive": 4234,
        "precision": 0.6707858048162231,
        "precision_ci95": [
          0.6590919898124957,
          0.6822718670918358
        ],
        "high_risk_by_band": {
          "priority_review": {
            "n": 2823,
            "n_high_risk": 714,
            "high_risk_share": 0.25292242295430395
          },
          "human_review": {
            "n": 3489,
            "n_high_risk": 202,
            "high_risk_share": 0.057896245342505016
          },
          "allow": {
            "n": 57666,
            "n_high_risk": 164,
            "high_risk_share": 0.002843963514029064
          }
        },
        "n_high_risk_total": 1080,
        "high_risk_recall_into_queue": 0.8481481481481481,
        "high_risk_recall_into_priority": 0.6611111111111111
      },
      "overload": {
        "severity": {
          "high_risk_arrived": {
            "mean": 217.4,
            "std": 14.235167719419396
          },
          "high_risk_handled": {
            "mean": 196.0,
            "std": 11.40175425099138
          },
          "harm_per_reviewer_hour": {
            "mean": 72.61875,
            "std": 1.5718967443824037
          }
        },
        "priority": {
          "high_risk_arrived": {
            "mean": 217.4,
            "std": 14.235167719419396
          },
          "high_risk_handled": {
            "mean": 196.0,
            "std": 11.40175425099138
          },
          "harm_per_reviewer_hour": {
            "mean": 72.56875,
            "std": 1.567417262569224
          }
        }
      }
    },
    {
      "ci_run_id": 35857528118,
      "ci_url": "https://github.com/XinyangWuEthz/review-router/actions/runs/35857528118",
      "run_id": "20260923T115815829404Z-human-review-baseline",
      "git_commit": "210d9857220b23d175d4aff5a54ab221ad86280b",
      "git_dirty": false,
      "data_sha256": {
        "train": "bd4084611bd27c939ba98e5e63bc3e5a2c1a4e99477dcba46c829e4c986c429d",
        "test": "c2513ce4abb98c4d1d216e3ca0d4377d57589a0989aa8c06a840509a16c786e8",
        "test_labels": "2a56dcbeba5c05f965a636f56cb5ae972bad60c3b952c239b49be18d7ab70f49"
      },
      "config_sha256": "801efb58457b838a7316359cc89b91670d762379269720fe6d548611f137162d",
      "policy_sha256": "3df259614b78297f94c1465c767c8386ef9ca13ea43242e2f72c7c47ded27b2e",
      "splits_sha256": "9f86d2a17105abaeff7b3e15bbdf4d983a307a8246b342655558390abb31045c",
      "model_sha256": "ac31ea513bf55ebc2500edf88e68f69c4e49f5e4b73d1c9a6f1a84fc5b378eef",
      "predictions_sha256": "438029e7d44f3b333240b3c536e97d5ecdb1349009851abf85e7c18a3e111f09",
      "training_source_sha256": {
        "review_router/model.py": "efcb9b4b8c9b61325d054ca17c175416ec3e7e80e8b7e0fe01fba402080a8253",
        "review_router/data.py": "932257ac72a007d6fb31f2bfd400b9504494d46f488fd0562e66142d87431334"
      },
      "seed": 20260922,
      "dependencies": {
        "python": "3.12.14",
        "review_router": "0.1.0",
        "numpy": "2.5.3",
        "pandas": "3.0.6",
        "scikit-learn": "1.9.1",
        "scipy": "1.18.1",
        "pyyaml": "6.0.3"
      },
      "model_config": {
        "max_features": 200000,
        "ngram_max": 2,
        "min_df": 3,
        "C": 4.0,
        "max_iter": 1000,
        "analyzer": "word",
        "char_ngram_min": 2,
        "char_ngram_max": 5,
        "char_max_features": 300000
      },
      "review_workload": {
        "n_total": 63978,
        "n_requires_human_review": 6312,
        "n_priority_review": 2823,
        "n_human_review": 3489,
        "review_fraction": 0.09865891400168808,
        "n_truly_positive": 4235,
        "precision": 0.6709442332065906,
        "precision_ci95": [
          0.6592517416596075,
          0.6824287793049322
        ],
        "high_risk_by_band": {
          "priority_review": {
            "n": 2823,
            "n_high_risk": 714,
            "high_risk_share": 0.25292242295430395
          },
          "human_review": {
            "n": 3489,
            "n_high_risk": 202,
            "high_risk_share": 0.057896245342505016
          },
          "allow": {
            "n": 57666,
            "n_high_risk": 164,
            "high_risk_share": 0.002843963514029064
          }
        },
        "n_high_risk_total": 1080,
        "high_risk_recall_into_queue": 0.8481481481481481,
        "high_risk_recall_into_priority": 0.6611111111111111
      },
      "overload": {
        "severity": {
          "high_risk_arrived": {
            "mean": 200.8,
            "std": 14.372195378577345
          },
          "high_risk_handled": {
            "mean": 184.6,
            "std": 10.011992808627062
          },
          "harm_per_reviewer_hour": {
            "mean": 71.04375,
            "std": 2.089276938321007
          }
        },
        "priority": {
          "high_risk_arrived": {
            "mean": 200.8,
            "std": 14.372195378577345
          },
          "high_risk_handled": {
            "mean": 184.2,
            "std": 10.087616170334794
          },
          "harm_per_reviewer_hour": {
            "mean": 70.9,
            "std": 2.1447064193031173
          }
        }
      }
    }
  ],
  "same_inputs": {
    "data_sha256": true,
    "config_sha256": true,
    "splits_sha256": true,
    "seed": true,
    "dependencies": true,
    "model_config": true,
    "training_source_sha256": true
  },
  "vocabulary": {
    "sizes": [
      200000,
      200000
    ],
    "common_terms": 199961,
    "exclusive_counts": [
      39,
      39
    ],
    "exclusive_terms": [
      [
        "00 04",
        "00 50",
        "be bully",
        "community school",
        "community ve",
        "community your",
        "compacted",
        "companies as",
        "companies but",
        "companies not",
        "ga was",
        "ga1 ve",
        "gaba",
        "gabrielf",
        "gadaffi",
        "gaddafi is",
        "gaelic and",
        "mathematical model",
        "only tangentially",
        "only term",
        "only third",
        "only times",
        "only tool",
        "only tried",
        "only unsourced",
        "only visual",
        "only warned",
        "slit up",
        "slit your",
        "slop",
        "slope arrests",
        "slope of",
        "zoran",
        "zuck have",
        "zuck just",
        "zuck people",
        "\u03b5\u03bc\u03c0\u03c1\u03bf\u03c2",
        "\u03ba\u03c5\u03c0\u03b1\u03c4\u03b6\u03b7\u03b4\u03b5\u03c2",
        "\u03ba\u03c5\u03c0\u03b1\u03c4\u03b6\u03b7\u03b4\u03b5\u03c2 and"
      ],
      [
        "and byzantine",
        "and calculate",
        "and callous",
        "and cameron",
        "be hip",
        "be historian",
        "be honored",
        "be hope",
        "connections if",
        "guatemalans",
        "gud",
        "guerre",
        "guess all",
        "guess could",
        "guess from",
        "guess how",
        "mark which",
        "markedly different",
        "markers in",
        "market as",
        "market research",
        "market share",
        "marketing speak",
        "slog through",
        "sloppy ill",
        "slovenia which",
        "slow so",
        "slowed and",
        "slower than",
        "slowly being",
        "slowly but",
        "slowly working",
        "slung",
        "slur is",
        "slur on",
        "slur somewhat",
        "smack ya",
        "smackbot has",
        "smackbot problem"
      ]
    ],
    "common_terms_max_abs_idf_difference": 4.440892098500626e-16,
    "training_rows_inspected": 95743,
    "swapped_term_frequency_counts": {
      "3": 78
    },
    "swapped_document_frequency_counts": {
      "3": 78
    }
  },
  "test_prediction_changes": {
    "n_rows": 63978,
    "max_abs_probability_difference": 0.10877756391996685,
    "mean_abs_probability_difference": 1.2697284693307496e-05,
    "n_rows_with_any_exact_score_difference": 63978,
    "n_rows_changing_final_band": 2,
    "n_rows_changing_admission": 2,
    "admission_changes": [
      {
        "id": "0eda54b0169500d1",
        "first_tier": "allow",
        "second_tier": "human_review"
      },
      {
        "id": "e1f4cc855f011b71",
        "first_tier": "human_review",
        "second_tier": "allow"
      }
    ]
  },
  "inspection_environment": {
    "python": "3.13.7",
    "dependencies": {
      "numpy": "2.5.3",
      "scikit-learn": "1.9.1",
      "pandas": "3.0.6"
    },
    "pickle_version_warnings": [],
    "note": "Only trusted project CI pickles were loaded. Saved vocabularies/IDFs were inspected; classifiers were not invoked. Cross-version estimator loading is unsupported by sklearn; all numerical score comparisons use saved CSV predictions."
  },
  "conclusion": "All checked data, split, seed, model-setting, dependency and training-source inputs match. The vocabularies share 199961 terms and have 39 and 39 exclusive terms. Every swapped term has raw term frequency 3 and document frequency 3, supporting tied-term selection at the feature cap. Platform-dependent ordering of ties is a suspected mechanism; the specific CPU or library cause has not been established. Retraining did not reproduce exact saved probability values. Because the training implementation is unchanged, this exposes a pre-existing reproducibility limitation. Changes in admitted row identities can also change which comments index-based simulation sampling draws. Compare orderings and cumulative/segment bands within the same saved run; cross-run metric differences alone do not isolate a policy effect."
}