# Apply the train-fixed rules (results/fixed_rule.json) to all periods; breakdowns; random same-minute trade benchmark; plain-cross benchmark.
import numpy as np,pandas as pd,sys,os,json
sys.path.insert(0,os.path.dirname(__file__)); from cwlib import *; from stats import variants,split_mask
DD=load(); days=sorted(DD); E=pd.read_pickle(f'{C}/results/events.pkl'); T=pd.read_pickle(f'{C}/results/trades_all.pkl'); T=T[T.status=='ok']
T=T.merge(E[['date','side','wall','K','hm','dte','wd','spot','exp']],left_on='ev',right_index=True)
pos={d:i for i,d in enumerate(days)}
T['tdte']=[sum(1 for x in days if d<x<=e) if e in pos else np.nan for d,e in zip(T.date,T.exp)]
T['gamma']=np.where(T.tdte==0,'expiry day',np.where(T.tdte==1,'pre-expiry','other'))
T['period']=np.where(T.date<'2026-01-01','train','test')
RU=json.load(open(f'{C}/results/fixed_rule.json')); V=variants(E); rng=np.random.default_rng(5)
out=[]; allrows=[]
for side in ('call','put'):
    r=RU[side]; idx,X=V[(side,r['wall'],r['variant'])]
    t=T[T.ev.isin(idx)&(T.X==X)&(T.instr==r['instr'])&(T.exit==r['exit'])].copy(); t['rule']=f"{side}:{r['wall']}:{r['variant']}:{r['instr']}:{r['exit']}"
    pl=T[(T.side==side)&(T.wall==r['wall'])&(T.X==0)&(T.instr==r['instr'])&(T.exit==r['exit'])].copy(); pl['rule']=f'{side}: plain cross (a), same instrument/exit'
    # random same-minute benchmark: for each rule trade, 20 random other days in same period at same signal minute, ATM option, same exit (spot stop ref = ATM strike)
    rr=[]; sg=1 if side=='call' else -1
    for x in t.itertuples():
        pool=[d for d in days if d!=x.date and ((d<pd.Timestamp('2026-01-01'))==(x.date<pd.Timestamp('2026-01-01')))]
        for d in rng.choice(pool,20,replace=False):
            D=DD[d]; s=x.s
            if D['c'][s]!=D['c'][s]: continue
            K=float(round(D['c'][s]/50)*50); z=trade(d,D,s,sg,K,K,r['exit'])
            if z['status']=='ok': rr.append(dict(date=d,period=x.period,rule=f'{side}: random same-minute (20 per trade)',**z))
    R=pd.DataFrame(rr)
    for nm,g in (('rule',t),('plain',pl),('random',R)):
        for per in ('train','test'):
            h=g[g.period==per]
            if per=='test': h2=h[~h.date.isin(pd.to_datetime(['2026-10-23','2026-10-26']))]
            out.append(dict(side=side,bench=nm,period=per,n=len(h),mean_pts=h.pnl.mean(),median=h.pnl.median(),win=(h.pnl>0).mean()*100,rs_lot_total=h.pnl_rs_lot.sum(),rs_lot_mean=h.pnl_rs_lot.mean(),
                            mean_ex_srcdays=h2.pnl.mean() if per=='test' else np.nan))
    allrows.append(t)
S=pd.DataFrame(out); S.round(2).to_csv(f'{C}/results/rule_vs_bench.csv',index=False)
A=pd.concat(allrows); A.to_csv(f'{C}/results/rule_trades.csv',index=False)
def br(g,key):
    return g.groupby(key).agg(n=('pnl','size'),mean_pts=('pnl','mean'),win=('pnl',lambda x:(x>0).mean()*100),rs_lot=('pnl_rs_lot','sum')).round(2)
A['year']=A.date.dt.year; A['month']=A.date.dt.strftime('%Y-%m')
by=[]
for side,g in A.groupby('side'):
    for key in ('year','month','gamma','wd'):
        b=br(g,key).reset_index().rename(columns={key:'bucket'}); b['side']=side; b['by']=key; by.append(b)
BY=pd.concat(by); BY.to_csv(f'{C}/results/rule_breakdowns.csv',index=False)
W=A.groupby(['side','date']).pnl_rs_lot.sum().reset_index().sort_values('pnl_rs_lot').groupby('side').head(3); W.to_csv(f'{C}/results/rule_worst_days.csv',index=False)
pd.set_option('display.width',250); print(S.round(2).to_string()); print(BY.to_string()); print(W)
# t-test-ish for test mean
from scipy.stats import ttest_1samp,mannwhitneyu
for side,g in A.groupby('side'):
    h=g[g.period=='test'].pnl
    if len(h)>2: print(side,'test mean',h.mean().round(2),'t-test p',ttest_1samp(h,0).pvalue.round(3))