# curated per-day headlines from local archive (market-moving keywords), saved for the report
import json,glob,re,pandas as pd
H={}
for f in sorted(glob.glob('/workspace/market-live-stack/feeds/news/2026-10-*.jsonl')):
    for l in open(f):
        try: j=json.loads(l)
        except: continue
        h=j.get('headline','')
        if h not in H or j.get('ts','')<H[h].get('ts',''): H[h]=j   # rule 2: keep earliest copy of a reposted headline
rows=pd.DataFrame([dict(ts=j.get('ts','')[:16],headline=j.get('headline',''),source=j.get('source',''),impact=j.get('india_impact'),url=j.get('url','')) for j in H.values()])
pat=r'(sensex|nifty|market).*(crash|fall|plunge|tank|slump|rebound|jump|recover|surge|rall)|why is (stock )?market|fii|fpi|jio.*(subscri|bids|demand|opens|closes|crore)|brent|diplomat|saudi|aramco|ceasefire|rbi.*(rate|hike)|finance minister|liquidity|cas (chaos|impact)|q2 result.*(hdfc bank|axis|icici|bajaj finance|nestle|tech mahindra|eternal|reliance)|hdfc bank q2|axis bank q2|icici bank q2'
SESS=['2026-10-12','2026-10-13','2026-10-14','2026-10-15','2026-10-16','2026-10-19','2026-10-21','2026-10-22','2026-10-23']
out=[]
for i,d in enumerate(SESS):
    lo=(pd.Timestamp(SESS[i-1]) if i else pd.Timestamp(d)-pd.Timedelta(days=3)).strftime('%Y-%m-%d')+'T15:30'; hi=d+'T15:30'
    x=rows[(rows.ts>lo)&(rows.ts<=hi)&rows.headline.str.contains(pat,flags=re.I,regex=True)].sort_values('ts')
    x=x[~x.headline.str.contains('SEBI|Notice of Demand|Recovery Certificate',regex=True)]
    x=x.drop_duplicates(subset=x.headline.str.lower().str[:45]) if False else x
    x['session']=d; out.append(x)
o=pd.concat(out)
# rule (2026-10-25 newsfix): ignore headlines whose quoted index move doesn't match the actual Nifty move that day (late reposts)
import sys; sys.path.insert(0,'/workspace/research/flow-forensics-2026-10/scripts'); from news_rules import mismatch
o['move_mismatch']=[mismatch(t.replace('T',' '),h) for t,h in zip(o.ts,o.headline)]
o[o.move_mismatch==True].to_csv('results/news_ignored_move_mismatch.csv',index=False); print('ignored (quoted move mismatch):',(o.move_mismatch==True).sum())
o=o[o.move_mismatch!=True]; o.to_csv('results/news_by_session.csv',index=False); print(o.groupby('session').size())