similarity in two more places MEASURED: triple by shape gives APS 2527 — below the published 2806 and below EVERY baseline including chandrupatla 2785 — and far-end by dissimilarity gives CHA 2683 (the ITP target met) with 5511 over both sets (project target 6026 beaten, ITP-oracle 5871 passed). Guarantee intact everywhere, 193/199. Neither mode is inert (38.6% / 26.9% of steps changed). A+B do NOT compose: 2884/2859 is worse than either alone — select, never weight, and never assume composability

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 23:45:55 +01:00
parent 04ee5c89b9
commit 22f19c57f0
10 changed files with 342 additions and 3 deletions
@@ -0,0 +1,54 @@
+"""A2/B2: is the new reading INERT? (PREREG_points_and_direction_2026-09-21 §4)
+
+A mode that changes nothing is not evidence either way, so the share of steps on which the
+new rule actually changed the decision is measured over the whole benchmark, per set.
+
+    python -u measure_inertness.py
+"""
+
+from __future__ import annotations
+
+from simroot.pure import pure_root
+from sweep_full import MAXFEV, XTOL, load_problems
+
+
+def run(mode):
+    kw = {"points": "shape"} if mode == "shape" else {"direction": "dissimilar"}
+    key = "triple_changed" if mode == "shape" else "dir_changed"
+    out = {}
+    for setname in ("APS", "CHA"):
+        steps = changed = solved = n = 0
+        for s, _pid, f, a, b, root in load_problems():
+            if s != setname:
+                continue
+            n += 1
+            st = {}
+            try:
+                x, nfev, ok = pure_root(f, a, b, XTOL, MAXFEV, similarity="ratio",
+                                        scale="derived3", guard="safe", prestep="slope",
+                                        stats=st, **kw)
+            except Exception:
+                continue
+            steps += max(nfev - 2, 0)
+            changed += st.get(key, 0)
+            solved += int(bool(ok) and abs(x - root) <= 1e-6)
+        out[setname] = (changed, steps, 100.0 * changed / steps if steps else float("nan"),
+                        solved, n)
+    return out
+
+
+def main():
+    for mode in ("shape", "dissimilar"):
+        res = run(mode)
+        print(f"\nmode {mode!r}:")
+        for setname, (ch, st, pct, solved, n) in res.items():
+            print(f"   {setname}: changed {ch} of {st} steps = {pct:.1f}%  "
+                  f"(solved {solved}/{n})")
+        pct_all = 100.0 * sum(r[0] for r in res.values()) / max(sum(r[1] for r in res.values()), 1)
+        verdict = "NOT inert" if pct_all >= 5.0 else ("INERT — the measurement is void"
+                                                     if pct_all < 1.0 else "borderline")
+        print(f"   overall {pct_all:.1f}%  -> {verdict} (prereg: >=5% not inert, <1% void)")
+
+
+if __name__ == "__main__":
+    main()