# Mechanical call extraction from caption episodes (rules fixed before scoring; see README_youtube.md)
import re,sys,datetime as dt,pandas as pd
sys.path.insert(0,'scripts'); from subs2txt import load
from extract_calls import ACT,SL,TG,CT,CS,first
from episodes import PLAN,STK
SIDEC=re.compile(r'कॉल|सीई'); SIDEP=re.compile(r'पुट|पीई')
def calls(vid,start,gap=120):
    L=[(t,x) for t,x in load(f'data/subs/{vid}.hi-orig.json3') if 'गोल्ड' not in x and 'क्रूड' not in x and 'बिटकॉइन' not in x]
    st=dt.datetime.fromisoformat(start)
    hits=[(t,x) for t,x in L if ACT.search(x) or PLAN.search(x) or any(r.search(x) for r in SL+TG)]
    eps=[];cur=None
    for t,x in hits:
        if cur and t-cur[-1][0]<=gap: cur.append((t,x))
        else: cur=[(t,x)]; eps.append(cur)
    out=[]
    for e in eps:
        acts=[(t,x) for t,x in e if ACT.search(x) and not re.search(r'नहीं|मत ',x)]
        txt=' '.join(x for _,x in e); sl,slt=first(SL,txt); tg,tgt=first(TG,txt)
        if acts: te=acts[0][0]; tier='ACT'
        elif PLAN.search(txt) and (sl or tg): te=e[-1][0]; tier='PLAN'
        else: continue
        w=' '.join(x for t,x in L if e[0][0]-60<=t<=te+30); nc=len(SIDEC.findall(w)); npp=len(SIDEP.findall(w))
        side='CE' if nc>npp else ('PE' if npp>nc else None)
        stk=STK.search(w); post=[(tt,xx) for tt,xx in L if te<tt<=te+2400]; cl=None
        for tt,xx in post:
            if CT.search(xx): cl=('T',tt); break
            if CS.search(xx): cl=('SL',tt); break
        out.append(dict(vid=vid,date=st.strftime('%Y-%m-%d'),entry_off_s=round(te),entry_ist=(st+dt.timedelta(seconds=te)).strftime('%Y-%m-%d %H:%M:%S'),tier=tier,side=side,n_ce=nc,n_pe=npp,spoken_strike=stk.group(1).replace(',','') if stk else None,sl_pts=sl,tg_pts=tg,claim=cl[0] if cl else None,claim_ist=(st+dt.timedelta(seconds=cl[1])).strftime('%H:%M:%S') if cl else None,ep_start=(st+dt.timedelta(seconds=e[0][0])).strftime('%H:%M:%S'),evidence=(acts[0][1] if acts else e[-1][1])[:100],sl_text=slt,tg_text=tgt))
    return out
if __name__=='__main__':
    inv=pd.read_csv('results/stream_inventory.csv').set_index('id'); o=[]
    for v in sys.argv[1:]: o+=calls(v,inv.loc[v,'start_ist'])
    d=pd.DataFrame(o); print(d[['vid','entry_ist','tier','side','spoken_strike','sl_pts','tg_pts','claim','evidence']].to_string())