| """2x2 discriminant study (reproduces Section 7.1 / Appendix O).
|
|
|
| Loads the GPT-4.1 judge CSVs from ``discriminant_2x2/judges/`` for
|
| 8 models on a 40-scenario factorial of task complexity (L/H) crossed
|
| with social pressure (L/H). Cell labels are derived from the
|
| scenario_id prefix (DC2x2-LL / LH / HL / HH). Prints per-cell DS>=2
|
| rates, sycophancy-positive rates (max of validation, accepting_framing,
|
| moral_endorsement >= 2), and within-cell Spearman rho between
|
| continuous DS and the continuous max-sycophancy score.
|
| """
|
|
|
| from __future__ import annotations
|
|
|
| import glob
|
| import sys
|
| from pathlib import Path
|
|
|
| import numpy as np
|
| import pandas as pd
|
| from scipy import stats
|
|
|
| sys.path.insert(0, str(Path(__file__).resolve().parent))
|
| from _paths import DISCRIM
|
|
|
| CELLS = ["LL", "LH", "HL", "HH"]
|
|
|
|
|
| def _cell(scenario_id: str) -> str:
|
| parts = scenario_id.split("-")
|
| return parts[1] if len(parts) >= 2 else ""
|
|
|
|
|
| def main() -> None:
|
| files = sorted(glob.glob(str(DISCRIM / "judges" / "*.csv")))
|
| if not files:
|
| print("No discriminant judge CSVs found.")
|
| return
|
|
|
| frames = []
|
| for f in files:
|
| df = pd.read_csv(f)
|
| frames.append(df)
|
| cat = pd.concat(frames, ignore_index=True)
|
|
|
| cat["cell"] = cat["scenario_id"].apply(_cell)
|
| cat = cat[cat["cell"].isin(CELLS)].copy()
|
|
|
| syco_cols = ["validation", "accepting_framing", "moral_endorsement"]
|
| cat["syco_max"] = cat[syco_cols].max(axis=1)
|
| cat["disp_bin"] = (cat["displacement_severity"] >= 2).astype(int)
|
| cat["syco_bin"] = (cat["syco_max"] >= 2).astype(int)
|
|
|
| print(f"Total responses across 8 models x 40 scenarios: {len(cat)}")
|
| print()
|
| print(f"{'Cell':<6} {'n':>4} {'DS>=2':>7} {'Syco+':>7} "
|
| f"{'rho(DS,Syco)':>13} {'p':>8}")
|
| print("-" * 56)
|
| for cell in CELLS:
|
| sub = cat[cat["cell"] == cell]
|
| ds_rate = sub["disp_bin"].mean() if len(sub) else float("nan")
|
| sy_rate = sub["syco_bin"].mean() if len(sub) else float("nan")
|
| if len(sub) >= 3 and sub["displacement_severity"].nunique() > 1 \
|
| and sub["syco_max"].nunique() > 1:
|
| rho, p = stats.spearmanr(sub["displacement_severity"],
|
| sub["syco_max"])
|
| else:
|
| rho, p = float("nan"), float("nan")
|
| print(f"{cell:<6} {len(sub):>4d} {ds_rate * 100:6.1f}% "
|
| f"{sy_rate * 100:6.1f}% {rho:>13.2f} {p:>8.3f}")
|
|
|
|
|
| if __name__ == "__main__":
|
| main()
|
|
|
