{
  "status": "Descriptive historical calibration; not a welfare model or causal estimate",
  "source": {
    "path_description": "local Coda prioritization CSV export",
    "snapshot_modified_at": "2026-09-08T15:38:38.940687+00:00",
    "rows": 325,
    "rated_rows": 238,
    "rows_with_internal_discussion": 296
  },
  "outcomes": {
    "published_evaluations": 57,
    "selected_or_in_evaluation_pipeline": 64,
    "terminal_model_rows": 194,
    "terminal_selected": 64,
    "terminal_deprioritized_or_blocked": 130
  },
  "definitions": {
    "selected_statuses": [
      "04_selected_choose_evaluation_manager",
      "10_seeking_(more)_evaluators",
      "20_awaiting_evaluations",
      "50_published evaluations (on PubPub, by Unjournal)"
    ],
    "published_status": "50_published evaluations (on PubPub, by Unjournal)",
    "negative_status_rule": "Status begins with deprioritized or authors rejected/blocked us",
    "pending_rows": "Excluded from the selection model rather than treated as negative"
  },
  "descriptive": {
    "by_rating": [
      {
        "rating_range": "0-39",
        "n_terminal": 17,
        "selected_n": 0,
        "selected_rate": 0.0
      },
      {
        "rating_range": "40-59",
        "n_terminal": 43,
        "selected_n": 0,
        "selected_rate": 0.0
      },
      {
        "rating_range": "60-79",
        "n_terminal": 77,
        "selected_n": 38,
        "selected_rate": 0.494
      },
      {
        "rating_range": "80-100",
        "n_terminal": 39,
        "selected_n": 19,
        "selected_rate": 0.487
      }
    ],
    "by_cause": [
      {
        "group": "Global health (LMICs)",
        "n_terminal": 31,
        "selected_n": 15,
        "selected_rate": 0.484,
        "mean_human_rating": 67.7,
        "rated_n": 29
      },
      {
        "group": "Dev. econ/gov. (LMICs)",
        "n_terminal": 27,
        "selected_n": 11,
        "selected_rate": 0.407,
        "mean_human_rating": 74.3,
        "rated_n": 25
      },
      {
        "group": "Econ., welfare, misc.",
        "n_terminal": 21,
        "selected_n": 5,
        "selected_rate": 0.238,
        "mean_human_rating": 63.8,
        "rated_n": 19
      },
      {
        "group": "Social impact of tech/AI",
        "n_terminal": 18,
        "selected_n": 3,
        "selected_rate": 0.167,
        "mean_human_rating": 53.2,
        "rated_n": 17
      },
      {
        "group": "Environment",
        "n_terminal": 17,
        "selected_n": 6,
        "selected_rate": 0.353,
        "mean_human_rating": 68.5,
        "rated_n": 15
      },
      {
        "group": "Innovation & meta-science",
        "n_terminal": 17,
        "selected_n": 5,
        "selected_rate": 0.294,
        "mean_human_rating": 68.6,
        "rated_n": 17
      },
      {
        "group": "Animal welfare, markets",
        "n_terminal": 15,
        "selected_n": 5,
        "selected_rate": 0.333,
        "mean_human_rating": 64.5,
        "rated_n": 12
      },
      {
        "group": "Catastrophic & X-risk, LT, forecasting",
        "n_terminal": 15,
        "selected_n": 6,
        "selected_rate": 0.4,
        "mean_human_rating": 64.5,
        "rated_n": 11
      },
      {
        "group": "Attitudes, behaviors, psych, misinfo.",
        "n_terminal": 13,
        "selected_n": 3,
        "selected_rate": 0.231,
        "mean_human_rating": 57.9,
        "rated_n": 13
      },
      {
        "group": "Health (all countries)",
        "n_terminal": 6,
        "selected_n": 1,
        "selected_rate": 0.167,
        "mean_human_rating": 63.4,
        "rated_n": 5
      }
    ],
    "by_publication": [
      {
        "group": "Published",
        "n_terminal": 89,
        "selected_n": 23,
        "selected_rate": 0.258,
        "mean_human_rating": 62.3,
        "rated_n": 83
      },
      {
        "group": "Working paper or preprint",
        "n_terminal": 88,
        "selected_n": 35,
        "selected_rate": 0.398,
        "mean_human_rating": 64.8,
        "rated_n": 82
      },
      {
        "group": "Unknown",
        "n_terminal": 17,
        "selected_n": 6,
        "selected_rate": 0.353,
        "mean_human_rating": 76.6,
        "rated_n": 11
      }
    ],
    "coded_feature_presence": [
      {
        "feature": "lmic_context",
        "n_terminal": 73,
        "selected_n": 32,
        "selected_rate": 0.438
      },
      {
        "feature": "methods_or_measurement",
        "n_terminal": 54,
        "selected_n": 26,
        "selected_rate": 0.481
      },
      {
        "feature": "catastrophic_or_existential_risk",
        "n_terminal": 37,
        "selected_n": 11,
        "selected_rate": 0.297
      },
      {
        "feature": "animal_welfare",
        "n_terminal": 23,
        "selected_n": 8,
        "selected_rate": 0.348
      },
      {
        "feature": "observed_choices_or_behavior",
        "n_terminal": 19,
        "selected_n": 9,
        "selected_rate": 0.474
      },
      {
        "feature": "experimental_or_quasi_experimental",
        "n_terminal": 18,
        "selected_n": 8,
        "selected_rate": 0.444
      },
      {
        "feature": "wealthy_country_context",
        "n_terminal": 17,
        "selected_n": 4,
        "selected_rate": 0.235
      },
      {
        "feature": "synthesis_or_meta_analysis",
        "n_terminal": 13,
        "selected_n": 4,
        "selected_rate": 0.308
      },
      {
        "feature": "hypothetical_preferences",
        "n_terminal": 7,
        "selected_n": 2,
        "selected_rate": 0.286
      }
    ]
  },
  "models": {
    "rating_only_cross_validation": {
      "folds": 5,
      "n": 194,
      "auc": 0.8,
      "brier": 0.174
    },
    "rating_plus_metadata_cross_validation": {
      "folds": 5,
      "n": 194,
      "auc": 0.763,
      "brier": 0.189
    },
    "full_sample_regularized_logit_coefficients": [
      {
        "feature": "rating_missing",
        "log_odds_coefficient": 1.568
      },
      {
        "feature": "cause:Innovation & meta-science",
        "log_odds_coefficient": -1.047
      },
      {
        "feature": "methods_or_measurement",
        "log_odds_coefficient": 0.983
      },
      {
        "feature": "human_rating_10_points",
        "log_odds_coefficient": 0.878
      },
      {
        "feature": "cause:Dev. econ/gov. (LMICs)",
        "log_odds_coefficient": -0.736
      },
      {
        "feature": "synthesis_or_meta_analysis",
        "log_odds_coefficient": -0.588
      },
      {
        "feature": "wealthy_country_context",
        "log_odds_coefficient": -0.56
      },
      {
        "feature": "publication:Working paper or preprint",
        "log_odds_coefficient": 0.515
      },
      {
        "feature": "lmic_context",
        "log_odds_coefficient": 0.503
      },
      {
        "feature": "animal_welfare",
        "log_odds_coefficient": 0.489
      },
      {
        "feature": "cause:Catastrophic & X-risk, LT, forecasting",
        "log_odds_coefficient": 0.431
      },
      {
        "feature": "catastrophic_or_existential_risk",
        "log_odds_coefficient": -0.373
      },
      {
        "feature": "observed_choices_or_behavior",
        "log_odds_coefficient": 0.366
      },
      {
        "feature": "cause:Environment",
        "log_odds_coefficient": -0.312
      },
      {
        "feature": "experimental_or_quasi_experimental",
        "log_odds_coefficient": -0.31
      },
      {
        "feature": "hypothetical_preferences",
        "log_odds_coefficient": -0.298
      },
      {
        "feature": "cause:Econ., welfare, misc.",
        "log_odds_coefficient": -0.279
      },
      {
        "feature": "cause:Social impact of tech/AI",
        "log_odds_coefficient": -0.218
      },
      {
        "feature": "cause:Animal welfare, markets",
        "log_odds_coefficient": -0.149
      },
      {
        "feature": "publication:Published",
        "log_odds_coefficient": 0.087
      }
    ]
  },
  "interpretation_limits": [
    "Human ratings and commissioning decisions combine expected impact, evaluation value, feasibility, author consent, evaluator supply, timing, and management judgment.",
    "The model describes historical choices. It does not recover moral weights or identify the causal effect of a paper characteristic.",
    "Feature coding uses transparent keyword rules over local text, including internal comments, but exports only aggregate counts and coefficients.",
    "Country-income and method labels are incomplete and may be misclassified; titles and comments are not substitutes for structured fields.",
    "Status is right-censored. Pending and unclear rows are excluded, and results can change as the pipeline advances."
  ]
}