PREREG_defect_contra_flat outcome (smoke, all 199 problems). Goal (below Chandrupatla on both) NOT reached by any cell: best CHA 2537 vs 2441. Best total: defect+flatanalogmem APS 2530 / CHA 2561 = 5091 < Chandrupatla 5228 (first time), C1-C3 hold; ablation: defect earns APS +284 (costs CHA 9), remembered analogy earns CHA +649 (costs APS 7). contra earned without defect (APS 2806 -> 2785, CHA 3220 -> 3198; with the remembered analogy 2792 / 2537, the best CHA), redundant with defect (defect+contra = defect). fun9: 105 -> 81 (Chandrupatla 55). Remaining CHA gap 120: flat +88, fun1 +20, fun9 +26. Server regression x2: 1116 passed, 1 failed = the paper-number check.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-21 15:57:15 +01:00
parent 3c92d610d6
commit 3d9287ec2c
6 changed files with 205 additions and 12 deletions
@@ -0,0 +1,64 @@
+"""Verdicts of PREREG_defect_contra_flat_2026-09-21 (C1-C4, the goal) from a results file.
+
+    python -u analyze_defcontra.py [sweep_defcontra_smoke.json]
+"""
+from __future__ import annotations
+
+import json
+import sys
+
+from analyze_extrap import h2h, sign_p
+from where_we_lose import per_family
+
+CH_APS, CH_CHA = 2787, 2441
+
+
+def main(path="sweep_defcontra_smoke.json"):
+    rows = json.load(open(path, encoding="utf8"))
+    cha = {r["label"]: r for r in json.load(open("sweep_full.json", encoding="utf8"))}["chandrupatla"]
+    cell = {(r["spec"][9], r["spec"][5]): r for r in rows if r["spec"][3] == "project"}
+    for ext, name in (("none", "proposed rule"), ("pattern", "proposed rule + G2")):
+        base = cell[(ext, "none")]
+        b_aps, b_cha = base["APS"]["total"], base["CHA"]["total"]
+        print(f"\n== {name}: baseline APS {b_aps}  CHA {b_cha}")
+        print(f"   {'prestep':<30} {'APS':>5} {'CHA':>5} {'both':>5}  C1 C2 C3  goal  "
+              f"vs Chandrupatla APS / CHA (w/t/l)")
+        best = None
+        for (e, pre), r in cell.items():
+            if e != ext:
+                continue
+            aps, chv = r["APS"]["total"], r["CHA"]["total"]
+            c1, c2 = chv < b_cha, aps <= b_aps
+            c3 = r["APS"]["over"] + r["CHA"]["over"] == 0 and \
+                r["APS"]["solved"] + r["CHA"]["solved"] >= 193
+            goal = aps < CH_APS and chv < CH_CHA
+            wa, ta, la = h2h(r, cha, "APS")
+            wc, tc, lc = h2h(r, cha, "CHA")
+            print(f"   {pre:<30} {aps:5d} {chv:5d} {aps + chv:5d}  "
+                  f"{'✓' if c1 else '✗'}  {'✓' if c2 else '✗'}  {'✓' if c3 else '✗'}   "
+                  f"{'YES' if goal else ' no'}   {wa}/{ta}/{la} p={sign_p(wa, la):.2g}  |  "
+                  f"{wc}/{tc}/{lc}")
+            if c1 and c2 and c3 and (best is None or aps + chv < best[1]):
+                best = (pre, aps + chv)
+        if best is None:
+            print("   no cell passes C1-C3")
+            continue
+        pre = best[0]
+        print(f"   best accepted cell: {pre}  (both {best[1]})")
+        parts = pre.split("+")
+        r = cell[(ext, pre)]
+        for p in parts:
+            rest = "+".join(q for q in parts if q != p) or "none"
+            if (ext, rest) not in cell:
+                continue
+            o = cell[(ext, rest)]
+            print(f"     C4 ablation - remove {p:<14}: APS {o['APS']['total'] - r['APS']['total']:+d}"
+                  f"  CHA {o['CHA']['total'] - r['CHA']['total']:+d}   (positive = the component earns it)")
+        a = per_family(r, "CHA")
+        c = per_family(cha, "CHA")
+        print("     CHA per family (ours/Chandrupatla): "
+              + "  ".join(f"{k} {a[k]}/{c[k]}" for k in sorted(c)))
+
+
+if __name__ == "__main__":
+    main(*sys.argv[1:2])