# Pre-specified 'heavy buying at the low' signals + controls; outcomes on spot (no filling, cap 15:14)
import pandas as pd,numpy as np,itertools
P=pd.read_parquet('data/minute_panel_v.parquet')
P=P[P.hm<='15:14'].copy()
P['tm']=P.timestamp.dt.hour*60+P.timestamp.dt.minute
ROWS=[]
XS=[150,200,250]; VS=[3,5]
for d,g in P.groupby('date'):
    g=g.reset_index(drop=True)
    if g.close.isna().mean()>0.05: continue   # sessions with >5% missing minutes skipped (data gap)
    o,h,l,c=g.open.values,g.high.values,g.low.values,g.close.values
    fvr=g.fvr.values; foi=g.foi.values; tm=g.tm.values; n=len(g)
    rh=np.fmax.accumulate(np.nan_to_num(h,nan=-1e9)); rl=np.fmin.accumulate(np.nan_to_num(l,nan=1e9))
    rh_i=np.array([int(np.argmax(np.nan_to_num(h[:i+1],nan=-1e9))) for i in range(n)])
    rl_i=np.array([int(np.argmin(np.nan_to_num(l[:i+1],nan=1e9))) for i in range(n)])
    fv3=pd.Series(fvr).rolling(3,min_periods=3).mean().values
    doi3=foi-np.r_[[np.nan]*3,foi[:-3]]
    done={}
    for t in range(1,n):
        if not (9*60+20<=tm[t]<=14*60+14): continue
        if not np.isfinite(c[t]) or not np.isfinite(c[t-1]): continue
        dd=rh[t]-rl[t]
        if dd<150 or rl_i[t]<rh_i[t] or t-rl_i[t]>3: continue
        rng=h[t]-l[t]; shape=(rng>0) and (c[t]>=l[t]+0.6*rng) and (c[t]>c[t-1]) and (c[t]-rl[t]<=25)
        at_low=(c[t]-rl[t]<=25)
        fut=dict(fvr1=fvr[t],fvr3=fv3[t],doi3=doi3[t])
        # outcomes
        H={k:min(t+k,n-1) for k in (15,30,60)}
        out={}
        for k,j in H.items():
            seg=h[t+1:j+1]; out[f'mfe{k}']=np.nanmax(seg)-c[t] if len(seg) and np.isfinite(seg).any() else np.nan
            segl=l[t+1:j+1]; out[f'mae{k}']=np.nanmin(segl)-c[t] if len(segl) and np.isfinite(segl).any() else np.nan
        segl=l[t+1:min(t+61,n)]; out['hold60']=bool(np.nanmin(segl)>=rl[t]) if np.isfinite(segl).any() else np.nan
        segl=l[t+1:]; out['hold_day']=bool(np.nanmin(segl)>=rl[t]) if np.isfinite(segl).any() else np.nan
        out['to_1514']=c[n-1]-c[t] if np.isfinite(c[n-1]) else np.nan
        for X in XS:
            if dd<X: continue
            # price-only reversal bar (all years), first per day
            if shape and ('P',X) not in done: done[('P',X)]=1; ROWS.append(dict(date=d,hm=g.hm[t],kind='price_bar',X=X,V=0,win='-',oi='-',dd=dd,entry=c[t],low=rl[t],**fut,**out))
            if at_low and ('A',X) not in done: done[('A',X)]=1; ROWS.append(dict(date=d,hm=g.hm[t],kind='at_low_any',X=X,V=0,win='-',oi='-',dd=dd,entry=c[t],low=rl[t],**fut,**out))
            if not np.isfinite(fvr[t]): continue
            # control: price bar with quiet futures volume
            if shape and fvr[t]<1.5 and (np.isfinite(fv3[t]) and fv3[t]<1.5) and ('C',X) not in done:
                done[('C',X)]=1; ROWS.append(dict(date=d,hm=g.hm[t],kind='control_quiet',X=X,V=0,win='-',oi='-',dd=dd,entry=c[t],low=rl[t],**fut,**out))
            for V,win,oif in itertools.product(VS,['1bar','3bar'],['any','oi_down']):
                v=fvr[t] if win=='1bar' else fv3[t]
                if not shape or not np.isfinite(v) or v<V: continue
                if oif=='oi_down' and not (np.isfinite(doi3[t]) and doi3[t]<0): continue
                key=('S',X,V,win,oif)
                if key in done: continue
                done[key]=1; ROWS.append(dict(date=d,hm=g.hm[t],kind='signal',X=X,V=V,win=win,oi=oif,dd=dd,entry=c[t],low=rl[t],**fut,**out))
S=pd.DataFrame(ROWS); S['year']=pd.to_datetime(S.date).dt.year
S.to_csv('results/p2_events.csv',index=False)
print(S.groupby(['kind','X','V','win','oi']).size().to_string())