#!/usr/bin/env python
"""Full pre-registered battery for the RVOL-ORB study.

Pre-reg: <data-root>/fp-data/studies/2026-08-25-rvol-orb-prereg.md
Runs: primary cell, monotonicity, entry A vs B, friction sensitivity,
year-by-year, IS/OOS split, and the 7-instrument replication set.
"""
import numpy as np
import pandas as pd
from scipy.stats import spearmanr

import rvol_orb as R

SYMS = ["NQ", "ES", "RTY", "YM", "CL", "GC", "ZN", "6E"]
OOS_START = 2020


def t(a):
    return R.tstat(a)


def line(k, n, m, tt, hit, extra=""):
    print(f"{k:>22} {n:>6} {m:>10.3f} {tt:>7.2f} {hit:>6.1f} {extra}")


def main():
    cache = {}
    print("=" * 92)
    print("PRIMARY  —  NQ · 5m OR · Entry B (confirm) · Q5 · 1.0 pt friction · pass bar t > 3.0")
    print("=" * 92)

    df = R.run("NQ", confirm=True, cost_mult=1.0)
    cache[("NQ", True, 1.0)] = df
    q5 = df[df["q"] == 5]["net"].to_numpy()
    print(f"  n = {len(q5)}   mean net = {q5.mean():+.3f} pt   t = {t(q5):+.2f}   "
          f"hit = {100*(q5>0).mean():.1f}%   total = {q5.sum():+.1f} pt")
    verdict = "PASS" if (t(q5) > 3.0 and q5.mean() > 0) else "FAIL"
    print(f"  >>> PRIMARY {verdict}")

    print("\n" + "=" * 92)
    print("SECONDARY 1 — monotonicity across RVOL quintiles (the distinctive Zarattini claim)")
    print("=" * 92)
    print(f"{'bucket':>22} {'n':>6} {'mean_net':>10} {'t':>7} {'hit%':>6}   rvol range")
    means = []
    for q in range(1, 6):
        sub = df[df["q"] == q]
        v = sub["net"].to_numpy()
        means.append(v.mean())
        line(f"Q{q}", len(v), v.mean(), t(v), 100 * (v > 0).mean(),
             f"  {sub['rvol'].min():.2f}–{sub['rvol'].max():.2f}")
    rho = spearmanr(list(range(5)), means).statistic
    print(f"  Spearman rho(Q1->Q5) = {rho:+.3f}   (Zarattini on stocks: strictly increasing)")

    print("\n" + "=" * 92)
    print("SECONDARY 2 — entry rule: A (naive touch) vs B (desk confirm)")
    print("=" * 92)
    print(f"{'cell':>22} {'n':>6} {'mean_net':>10} {'t':>7} {'hit%':>6}")
    for confirm, nm in [(False, "A naive"), (True, "B confirm")]:
        d = cache.get(("NQ", confirm, 1.0))
        if d is None:
            d = R.run("NQ", confirm=confirm, cost_mult=1.0)
        cache[("NQ", confirm, 1.0)] = d
        for scope, sel in [("all", d), ("Q5", d[d["q"] == 5])]:
            v = sel["net"].to_numpy()
            line(f"{nm} · {scope}", len(v), v.mean(), t(v), 100 * (v > 0).mean())

    print("\n" + "=" * 92)
    print("SECONDARY 3 — friction sensitivity (NQ, Entry B, base cost 1.0 pt)")
    print("=" * 92)
    print(f"{'cell':>22} {'n':>6} {'mean_net':>10} {'t':>7} {'hit%':>6}")
    for mult, tag in [(0.0, "0.0 pt (gross)"), (0.5, "0.5 pt"), (1.0, "1.0 pt"), (2.0, "2.0 pt")]:
        d = cache.get(("NQ", True, mult))
        if d is None:
            d = R.run("NQ", confirm=True, cost_mult=mult)
        cache[("NQ", True, mult)] = d
        for scope, sel in [("all", d), ("Q5", d[d["q"] == 5])]:
            v = sel["net"].to_numpy()
            line(f"{tag} · {scope}", len(v), v.mean(), t(v), 100 * (v > 0).mean())

    print("\n" + "=" * 92)
    print("SECONDARY 4 — year-by-year, NQ Q5 (Mesfin's failure mode: one year masking flat ones)")
    print("=" * 92)
    d = cache[("NQ", True, 1.0)]
    sub = d[d["q"] == 5]
    pos = 0
    yrs = sorted(sub["year"].unique())
    print(f"{'year':>22} {'n':>6} {'mean_net':>10} {'total':>10}")
    for y in yrs:
        v = sub[sub["year"] == y]["net"].to_numpy()
        if len(v) == 0:
            continue
        pos += v.mean() > 0
        line(str(y), len(v), v.mean(), float("nan"), float("nan"), f"{v.sum():>10.1f}")
    print(f"  years with positive mean: {pos}/{len(yrs)}")
    is_ = sub[sub["year"] < OOS_START]["net"].to_numpy()
    oos = sub[sub["year"] >= OOS_START]["net"].to_numpy()
    print(f"  IS  (<{OOS_START}): n={len(is_):4d} mean={is_.mean():+.3f} t={t(is_):+.2f}")
    print(f"  OOS (>={OOS_START}): n={len(oos):4d} mean={oos.mean():+.3f} t={t(oos):+.2f}")

    print("\n" + "=" * 92)
    print("SECONDARY 5 — replication set, each instrument's own Q5, Entry B, desk friction")
    print("=" * 92)
    print(f"{'symbol':>22} {'n':>6} {'mean_net':>10} {'t':>7} {'hit%':>6}   p99 RVOL")
    repl = 0
    for s in SYMS:
        try:
            ds = R.run(s, confirm=True, cost_mult=1.0)
        except Exception as e:
            print(f"{s:>22}  ERROR {e}")
            continue
        if ds is None or len(ds) == 0:
            print(f"{s:>22}  no data")
            continue
        v = ds[ds["q"] == 5]["net"].to_numpy()
        tt = t(v)
        if s != "NQ" and tt > 2.0:
            repl += 1
        p99 = np.percentile(ds["rvol"].to_numpy(), 99)
        line(s, len(v), v.mean(), tt, 100 * (v > 0).mean(), f"  {p99:.2f}")
    print(f"  instruments replicating at t > 2.0 (excl. NQ): {repl}/7  (pre-reg needs >= 4)")

    print("\n" + "=" * 92)
    print("VERDICT vs pre-registered falsification criteria")
    print("=" * 92)
    c1 = t(q5) > 3.0 and q5.mean() > 0
    c2 = rho > 0
    c3 = repl >= 4
    print(f"  1. primary t > 3.0 and positive .......... {'PASS' if c1 else 'FAIL'}  (t={t(q5):+.2f})")
    print(f"  2. monotonicity rho > 0 .................. {'PASS' if c2 else 'FAIL'}  (rho={rho:+.3f})")
    print(f"  3. >= 4/7 instruments replicate t > 2.0 .. {'PASS' if c3 else 'FAIL'}  ({repl}/7)")
    print(f"\n  H1 = {'CONFIRMED' if (c1 and c2 and c3) else ('CONTESTED' if c1 else 'REFUTED')}")


if __name__ == "__main__":
    main()