#!/usr/bin/env python
"""Regime forensics on late-day momentum: why did 2023/2025 lose? Can a conditioner rescue it?"""
import os
import numpy as np
import pandas as pd

HERE = os.path.dirname(os.path.abspath(__file__))
sess = pd.read_parquet(os.path.join(HERE, "sessions.parquet"))
sess = sess.dropna(subset=["p1500", "p1530", "p1600", "prev_p1600"]).copy()
sess["year"] = sess["date"].dt.year
sess["month"] = sess["date"].dt.month
sess = sess.set_index("date")

r_sig = np.log(sess["p1530"] / sess["p1500"])
label = np.log(sess["p1600"] / sess["p1530"])
pts = sess["p1600"] - sess["p1530"]
r_day = np.log(sess["p1530"] / sess["prev_p1600"])
day_range_pct = (sess["day_hi_1530"] - sess["day_lo_1530"]) / sess["p1530"]
# trailing 20d realized day-range (vol regime proxy, lagged 1 day => known at 15:30)
vol20 = day_range_pct.rolling(20).mean().shift(1)
COST = 1.0

mask = np.abs(r_sig) > 0.003
side = np.sign(r_sig)
d = pd.DataFrame({
    "year": sess["year"], "month": sess["month"],
    "pl_r": side * label, "pl_p": side * pts, "side": side,
    "day_range_pct": day_range_pct, "vol20": vol20,
    "agree": np.sign(r_day) == np.sign(r_sig),
})[mask]

def line(sub, name):
    n = len(sub)
    if n < 20:
        return f"{name:46s} n={n} (too few)"
    m = sub["pl_r"].mean(); t = m / (sub["pl_r"].std(ddof=1) / np.sqrt(n))
    return (f"{name:46s} n={n:4d} {m*1e4:+6.2f}bps t={t:+5.2f} hit={(sub['pl_r']>0).mean():.3f} "
            f"net={sub['pl_p'].mean()-COST:+6.2f}pt")

print("=== 2025 forensics: monthly ===")
y25 = d[d.year == 2025]
print(y25.groupby("month").agg(n=("pl_p","size"), net=("pl_p", lambda x: x.mean()-COST),
                               longs=("side", lambda x: (x>0).sum())).round(1).to_string())
print("\n2025 by side:")
print(line(y25[y25.side > 0], "  2025 longs"))
print(line(y25[y25.side < 0], "  2025 shorts"))
print("\n2023 by side:")
y23 = d[d.year == 2023]
print(line(y23[y23.side > 0], "  2023 longs"))
print(line(y23[y23.side < 0], "  2023 shorts"))

print("\n=== vol-regime conditioning (vol20 = trailing 20d mean day-range%, lagged) ===")
q = d["vol20"].quantile([0.33, 0.67])
lo_v, hi_v = q.iloc[0], q.iloc[1]
print(f"terciles: lo<{lo_v:.3%}, hi>{hi_v:.3%}")
print(line(d[d.vol20 <= lo_v], "low-vol regime"))
print(line(d[(d.vol20 > lo_v) & (d.vol20 <= hi_v)], "mid-vol regime"))
print(line(d[d.vol20 > hi_v], "high-vol regime"))

print("\n=== today's range as conditioner (day_range_pct at 15:30) ===")
for thr in (0.01, 0.015, 0.02):
    print(line(d[d.day_range_pct > thr], f"day range > {thr:.1%}"))
    print(line(d[d.day_range_pct <= thr], f"day range <= {thr:.1%}"))

print("\n=== combined rule candidates ===")
print(line(d[d.agree], "A: momentum agrees with day move"))
print(line(d[d.agree & (d.day_range_pct > 0.015)], "B: agree AND day range>1.5%"))
print(line(d[d.agree & (d.vol20 > lo_v)], "C: agree AND not-low-vol regime"))
print(line(d[(d.day_range_pct > 0.015)], "D: day range>1.5% only"))

print("\n=== candidate rules per-year (rule B and D) ===")
for nm, sub in (("B", d[d.agree & (d.day_range_pct > 0.015)]), ("D", d[d.day_range_pct > 0.015])):
    tbl = sub.groupby("year").agg(n=("pl_p","size"), net=("pl_p", lambda x: x.mean()-COST))
    tbl["net"] = tbl["net"].round(1)
    print(f"rule {nm}: " + "  ".join(f"{y}:{int(r.n)}/{r.net:+.0f}" for y, r in tbl.iterrows()))

print("\n=== rule D shorts vs longs, recent era (2023+) ===")
rec = d[(d.year >= 2023) & (d.day_range_pct > 0.015)]
print(line(rec[rec.side > 0], "2023+ D longs"))
print(line(rec[rec.side < 0], "2023+ D shorts"))