Add memory-pollution paper figure rendering

scripts/render_mempollution_figures.py

@@ -0,0 +1,111 @@
+"""Render the memory-pollution paper figures.
+
+Figure 1: rule-based exposure-amplification dose-response, one line per
+memory-decay rate, showing the curves coincide (decay does not attenuate
+uniform pollution).
+Figure 2: per-LLM hold-ratio shift at the highest dose, grouped by pollution
+kind, with a zero line; the bars separate susceptible from immune models.
+
+Usage:
+  python scripts/render_mempollution_figures.py --output-dir paper/mempollution/figures
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+from collections import defaultdict
+from pathlib import Path
+
+import matplotlib
+
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+
+ROOT = Path(__file__).resolve().parents[1]
+DET = ROOT / "docs/results/memory_pollution/memory_pollution_dose_response.csv"
+LLM = ROOT / "docs/results/memory_pollution_llm/model_difference.csv"
+
+LLM_LABEL = {
+    "poe:gpt-5.5": "gpt-5.5",
+    "poe:claude-opus-4.7": "claude-opus-4.7",
+    "poe:gemini-3.1-pro": "gemini-3.1-pro",
+    "deepseek:deepseek-v4-pro": "deepseek-v4-pro$^\\dagger$",
+    "glm:glm-5": "glm-5$^\\dagger$",
+    "poe:glm-5": "glm-5 (routed)",
+}
+
+
+def render_decay_invariance(output_dir: Path) -> Path:
+    rows = list(csv.DictReader(DET.open(encoding="utf-8")))
+    metric = "memory_driven_leverage_amplification"
+    by_decay: dict[str, dict[float, float]] = defaultdict(dict)
+    for r in rows:
+        if r["kind"] == "fake_rejections" and r["risk"] == "max-position" and r["outcome"] == metric:
+            by_decay[r["decay"]][float(r["dose"])] = float(r["mean_delta_vs_dose0"])
+    fig, ax = plt.subplots(figsize=(5.4, 3.0))
+    markers = {"0.6": "o", "0.85": "s", "1.0": "^"}
+    for decay in sorted(by_decay, key=float):
+        pts = sorted(by_decay[decay].items())
+        xs = [0.0] + [d for d, _ in pts]
+        ys = [0.0] + [v for _, v in pts]
+        ax.plot(xs, ys, marker=markers.get(decay, "o"), label=f"decay {decay}", linewidth=1.4)
+    ax.set_xlabel("Pollution dose", fontsize=9)
+    ax.set_ylabel("Exposure-amplification shift", fontsize=9)
+    ax.set_title("Decay does not attenuate uniform pollution", fontsize=9)
+    ax.legend(fontsize=8, frameon=False)
+    ax.tick_params(labelsize=8)
+    ax.grid(linewidth=0.3, alpha=0.5)
+    fig.tight_layout()
+    path = output_dir / "decay_invariance.pdf"
+    fig.savefig(path, bbox_inches="tight")
+    plt.close(fig)
+    return path
+
+
+def render_model_shift(output_dir: Path) -> Path:
+    rows = [r for r in csv.DictReader(LLM.open(encoding="utf-8")) if r["outcome"] == "hold_ratio"]
+    shift: dict[str, dict[str, float]] = defaultdict(dict)
+    sig: dict[str, dict[str, bool]] = defaultdict(dict)
+    for r in rows:
+        shift[r["agent"]][r["kind"]] = float(r["mean_delta"])
+        p = r["permutation_p_value"]
+        sig[r["agent"]][r["kind"]] = p not in ("", None) and float(p) < 0.05
+    models = [m for m in LLM_LABEL if m in shift]
+    models.sort(key=lambda m: -max(shift[m].values()))
+    kinds = [("fake_rejections", "phantom rejections", "#4c72b0"), ("fake_violations", "phantom violations", "#c44e52")]
+    fig, ax = plt.subplots(figsize=(6.6, 3.0))
+    width = 0.38
+    for j, (k, label, color) in enumerate(kinds):
+        xs = [i + (j - 0.5) * width for i in range(len(models))]
+        vals = [shift[m].get(k, 0.0) for m in models]
+        bars = ax.bar(xs, vals, width, label=label, color=color)
+        for x, m, b in zip(xs, models, bars):
+            if sig[m].get(k):
+                ax.text(x, b.get_height() + 0.004, "*", ha="center", fontsize=11)
+    ax.axhline(0.0, color="black", linewidth=0.7)
+    ax.set_xticks(range(len(models)), [LLM_LABEL[m] for m in models], rotation=20, ha="right", fontsize=8)
+    ax.set_ylabel("Hold-ratio shift at max dose", fontsize=9)
+    ax.legend(fontsize=8, frameon=False, loc="upper right")
+    ax.tick_params(labelsize=8)
+    ax.set_title("* = $p<0.05$ (sign-flip permutation)", fontsize=8)
+    fig.tight_layout()
+    path = output_dir / "model_shift.pdf"
+    fig.savefig(path, bbox_inches="tight")
+    plt.close(fig)
+    return path
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="Render memory-pollution paper figures.")
+    parser.add_argument("--output-dir", default="paper/mempollution/figures")
+    args = parser.parse_args(argv)
+    output_dir = ROOT / args.output_dir if not Path(args.output_dir).is_absolute() else Path(args.output_dir)
+    output_dir.mkdir(parents=True, exist_ok=True)
+    print("wrote", render_decay_invariance(output_dir))
+    print("wrote", render_model_shift(output_dir))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())