PREREG before code and run: flat regime by the promise failing TWICE in a row (same r* = 1/4, a counter of consecutive unkept model steps, any other step resets it). Stage 1 detectability (T1, T2 against the single flag's 2.48 ratio), stage 2 in the method only if stage 1 passes (W1-W5, stop rule).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-21 13:35:37 +01:00
parent e7e74d71e5
commit 6a8bb2315f
2 changed files with 135 additions and 0 deletions
@@ -0,0 +1,87 @@
+"""Stage 1 of PREREG_flat_twice_2026-09-21.md: does 'the promise failed twice in a row'
+separate the flat families better than a single failure?  Method untouched.
+
+    python -u diag_flat_twice.py
+"""
+from __future__ import annotations
+
+from scipy.optimize._tstutils import _APS_TESTS_DICTS, _CHANDRUPATLA_TESTS_DICTS
+
+from simroot.pure import pure_root
+
+XTOL, MAXFEV, ACC = 1e-12, 200, 1e-6
+R_STAR = 0.25
+CFG = dict(similarity="ratio", scale="derived3", guard="project")
+FLAT = ("fun3", "fun4", "fun5", "fun6")
+
+
+def run_flags(f, a, b, r_true):
+    """(model steps judged, single flags, twice-in-a-row flags) for one run."""
+    evals = []
+
+    def g(t):
+        v = f(t)
+        evals.append((t, v))
+        return v
+    st = {"trace": []}
+    x, _n, conv = pure_root(g, a, b, XTOL, MAXFEV, stats=st, **CFG)
+    if not (conv and abs(float(x) - r_true) <= ACC):
+        return 0, 0, 0
+    order = [e[0] for e in evals]
+    judged = single = twice = run = 0
+    for s in st["trace"]:
+        try:
+            i = order.index(s["x"])
+        except ValueError:
+            run = 0
+            continue
+        if i + 1 >= len(evals):
+            continue
+        x_next, f_next = evals[i + 1]
+        if x_next != s["x"] + s["raw"] or s["fx"] == 0.0:
+            run = 0                       # not the model's own step: the run is broken
+            continue
+        judged += 1
+        r = f_next / s["fx"]
+        if r >= R_STAR:
+            run += 1
+            single += 1
+            twice += run >= 2
+        else:
+            run = 0
+    return judged, single, twice
+
+
+def main():
+    groups = {g: [0, 0, 0] for g in ("APS", "CHA fun1", "CHA flat fun3-6", "CHA fun7",
+                                      "CHA other")}
+    for d in _APS_TESTS_DICTS:
+        a, b = (float(v) for v in d["bracket"])
+        f = (lambda ff, aa: lambda t: ff(t, *aa))(d["f"], d.get("args", ()))
+        for k, v in enumerate(run_flags(f, a, b, float(d["root"]))):
+            groups["APS"][k] += v
+    for d in _CHANDRUPATLA_TESTS_DICTS:
+        a, b = (float(v) for v in d["bracket"])
+        fam = str(d["ID"]).split(".")[0]
+        g = ("CHA flat fun3-6" if fam in FLAT else "CHA fun1" if fam == "fun1"
+             else "CHA fun7" if fam == "fun7" else "CHA other")
+        for k, v in enumerate(run_flags(d["f"], a, b, float(d["root"]))):
+            groups[g][k] += v
+    print(f"{'group':<18} {'model steps':>12} {'single':>8} {'twice':>8}")
+    share = {}
+    for g, (j, s1, s2) in groups.items():
+        share[g] = (100.0 * s1 / max(j, 1), 100.0 * s2 / max(j, 1))
+        print(f"{g:<18} {j:12d} {share[g][0]:7.1f}% {share[g][1]:7.1f}%")
+    flat, aps, f1 = share["CHA flat fun3-6"], share["APS"], share["CHA fun1"]
+    t1 = flat[1] > aps[1] and flat[1] > f1[1]
+    ratio1 = flat[0] / aps[0] if aps[0] else float("inf")
+    ratio2 = flat[1] / aps[1] if aps[1] else float("inf")
+    print(f"\nT1 twice: flat {flat[1]:.1f}% > APS {aps[1]:.1f}% and > fun1 {f1[1]:.1f}%  ->  "
+          f"{'PASS' if t1 else 'FAIL'}")
+    print(f"T2 flat/APS ratio: twice {ratio2:.2f} vs single {ratio1:.2f}  ->  "
+          f"{'PASS' if ratio2 > ratio1 else 'FAIL'}")
+    return share
+
+
+if __name__ == "__main__":
+    main()