"""Detectability check for PREREG_empty_step_2026-09-21.md: empty step (E) and repeating
pattern with the period chosen from the data (R).  Method untouched - states come from the
behaviour-neutral trace hook.

    python -u diag_empty_step.py
"""
from __future__ import annotations

from scipy.optimize._tstutils import _APS_TESTS_DICTS, _CHANDRUPATLA_TESTS_DICTS

from diag_contraction_similarity import CONFIGS, FLAT, auc
from simroot.pure import pure_root

XTOL, MAXFEV, ACC = 1e-12, 200, 1e-6


def ratio_sim(a, b):
    hi, lo = max(a, b), min(a, b)
    return lo / hi if hi > 0.0 else 1.0


def features(f, a, b, r, cfg):
    """(S_E, S_R) at every traced state; S_R is None until two repeats fit the history."""
    st = {"trace": []}
    x, _n, conv = pure_root(f, a, b, XTOL, MAXFEV, stats=st, **cfg)
    if not (conv and abs(float(x) - r) <= ACC):
        return []
    out, rhos = [], []
    for s in st["trace"]:
        if s["xp2"] is None or s["fx"] == 0.0 or s["fp"] == 0.0:
            continue
        rho = abs(s["fx"] / s["fp"])
        if not (rho == rho) or rho == float("inf"):
            continue
        rhos.append(rho)
        s_e = ratio_sim(abs(s["fx"]), abs(s["fp"]))
        s_r = None
        k = len(rhos)
        best = None
        for p in range(2, k // 2 + 1):          # at least two repeats must fit the history
            sims = [ratio_sim(rhos[j], rhos[j - p]) for j in range(p, k)]
            v = sum(sims) / len(sims)
            if best is None or v > best:
                best = v
        s_r = best
        out.append((s_e, s_r))
    return out


def main():
    out = {}
    for cname, cfg in CONFIGS.items():
        groups = {"APS": [], "CHA fun1": [], "CHA flat fun3-6": [], "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", ()))
            groups["APS"] += features(f, a, b, float(d["root"]), cfg)
        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 other")
            groups[g] += features(d["f"], a, b, float(d["root"]), cfg)
        print(f"\n== {cname}")
        print(f"   {'group':<18} {'states':>7} {'E: S_E > 1/2':>13} {'R: S_R > 1/2':>13}"
              f" {'(R defined)':>12}")
        res = {"rates": {}, "auc": {}}
        for g, v in groups.items():
            n = max(len(v), 1)
            vr = [x[1] for x in v if x[1] is not None]
            e_rate = 100.0 * sum(x[0] > 0.5 for x in v) / n
            r_rate = 100.0 * sum(y > 0.5 for y in vr) / max(len(vr), 1)
            res["rates"][g] = (len(v), e_rate, len(vr), r_rate)
            print(f"   {g:<18} {len(v):7d} {e_rate:12.1f}% {r_rate:12.1f}% {len(vr):12d}")
        flat = groups["CHA flat fun3-6"]
        print("   AUC, flat states against:           E        R")
        for g in ("APS", "CHA fun1", "CHA other"):
            ae = auc([x[0] for x in flat], [x[0] for x in groups[g]])
            ar = auc([x[1] for x in flat if x[1] is not None],
                     [x[1] for x in groups[g] if x[1] is not None])
            res["auc"][g] = (ae, ar)
            print(f"     {g:<30} {ae:.3f}    {ar:.3f}")
        out[cname] = res
    print("\nbars: m>=2 fires 0.2% canonical / 5.8% flat; successive-ratio similarity AUC"
          " 0.28-0.42")
    return out


if __name__ == "__main__":
    main()