import pandas as pd
d=pd.read_csv('results/stream_inventory_raw.csv')
d['st']=pd.to_datetime(d.start_ist.str[:19]); d['en']=pd.to_datetime(d.end_ist_est.str[:19]); d['date']=d.st.dt.date
import pickle
P=pickle.load(open('/workspace/research/flow-forensics-2026-10/recov/data/opt_panels.pkl','rb'))
td=sorted(set(pd.Timestamp(k[0]).date() for k in P)); td=[x for x in td if pd.Timestamp('2026-05-06').date()<=x<=pd.Timestamp('2026-10-23').date()]  # trading days = days with real Nifty option bars
# coverage of market hours 09:15-15:30 per date (union of streams)
cov={}
for dt_,g in d.dropna(subset=['st']).groupby('date'):
    mins=set()
    for _,r in g.iterrows():
        a=max(r.st,pd.Timestamp(f'{dt_} 09:15')); b=min(r.en,pd.Timestamp(f'{dt_} 15:30'))
        if b>a: mins|=set(pd.date_range(a.floor('min'),b,freq='min'))
    cov[dt_]=len(mins)/375*100
d['market_hours_coverage_pct_day']=d.date.map(cov).round(0)
d['trading_day']=d.date.isin(set(td))
d.drop(columns=['st','en']).to_csv('results/stream_inventory.csv',index=False)
days=pd.DataFrame({'date':td}); days['streams']=days.date.map(d.groupby('date').size()).fillna(0).astype(int); days['coverage_pct']=days.date.map(cov).fillna(0).round(0)
days['flag']=days.apply(lambda r:'NO_STREAM (possible deleted/not streamed)' if r.streams==0 else ('PARTIAL' if r.coverage_pct<80 else ''),axis=1)
days.to_csv('results/trading_day_coverage.csv',index=False)
print('streams',len(d),'errors',d.error.notna().sum() if 'error' in d else 0,'trading days',len(td),'last data day',td[-1])
print(days.flag.value_counts()); print(days[days.flag!=''].to_string())
print(d[d.error.notna()][['id','title','error']] if 'error' in d else '')
print('non-trading-day streams',d[~d.trading_day][['id','date','duration_s']].to_string())