"""Offline Decimal audit of original aggregate records and complete later queries.
Requires the archived complete cohort. --write regenerates the public projection; default checks it.
"""
from pathlib import Path
from decimal import Decimal,getcontext
import collections,datetime as dt,hashlib,json,statistics
getcontext().prec=50
import argparse,gzip,urllib.parse
REPO=Path(__file__).resolve().parents[3]
INPUT=Path(__file__).with_name('2026-10-23-listing-review-inputs.json.gz')
archive=json.loads(gzip.decompress(INPUT.read_bytes()))
original=archive['original']
by_symbol={c['symbol']:c for c in archive['queries']}
DAY=86400000;D=Decimal
def rows_for(c,kind):
 rows=[];previous=None
 for env in c[kind]:
  assert hashlib.sha256(env['body'].encode()).hexdigest()==env['bodySha256']
  params=urllib.parse.parse_qs(urllib.parse.urlparse(env['url']).query)
  page=json.loads(env['body']);assert isinstance(page,list)
  assert params['symbol']==[c['symbol']]
  if previous is not None:assert int(params['startTime'][0])==previous+1
  assert all(r['symbol']==c['symbol'] and int(params['startTime'][0])<=r['fundingTime']<=int(params['endTime'][0]) for r in page)
  assert all(page[i]['fundingTime']<page[i+1]['fundingTime'] for i in range(len(page)-1))
  rows+=page
  if page:previous=page[-1]['fundingTime']
 assert len(rows)==len({r['fundingTime'] for r in rows})
 return rows
def rate_stats(rows,anchor):
 result={}
 for name,lo,hi in [('day 1',0,DAY),('days 2-3',DAY,3*DAY),('days 4-7',3*DAY,7*DAY),('days 8-14',7*DAY,14*DAY),('days 15-30',14*DAY,30*DAY),('first 7 days',0,7*DAY),('late',30*DAY,90*DAY)]:
  sel=[r for r in rows if anchor+lo<=r['fundingTime']<anchor+hi]
  if not sel:continue
  values=[D(r['fundingRate']) for r in sel];gaps=[sel[i]['fundingTime']-sel[i-1]['fundingTime'] for i in range(1,len(sel))]
  result[name]={'n':len(values),'absoluteSum':str(sum(abs(v) for v in values)),'absoluteMean':str(sum(abs(v) for v in values)/len(values)),'negativeN':sum(v<0 for v in values),'distinctN':len(set(values)),'gapHours':sorted({round(g/3600000,5) for g in gaps}),'first':sel[0]['fundingTime'],'last':sel[-1]['fundingTime']}
 return result
query={'results':[]};bars={'results':[]}
for c in archive['queries']:
 real=rows_for(c,'real');placebo=rows_for(c,'placebo');ob=c['onboard']
 query['results'].append({'symbol':c['symbol'],'onboard':ob,'real':rate_stats(real,ob),'first1000':rate_stats(real[:1000],ob),'placebo':rate_stats(placebo,ob+200*DAY)})
 for name in ['sessions','firstReturned']:assert hashlib.sha256(c[name]['body'].encode()).hexdigest()==c[name]['bodySha256']
 sessions=json.loads(c['sessions']['body']);first=json.loads(c['firstReturned']['body'])
 assert all(ob//DAY*DAY<=r[0]<ob+7*DAY for r in sessions)
 bars['results'].append({'symbol':c['symbol'],'quoteVolumeUsdt':str(sum(D(r[7]) for r in sessions)) if sessions else None,'originalSessionQuoteVolume':archive['archivedSessionVolumes'][c['symbol']],'returnedLagMinutes':(first[0][0]-ob)/60000 if first else None,'originalRecordedLagMinutes':archive['archivedLagMinutes'][c['symbol']]})
assert len(query['results'])==len(bars['results'])==514
old={c['sym']:c for c in original['contracts']};bar={c['symbol']:c for c in bars['results']}
def summary(values):
 s=sorted(values);n=len(s)
 if not n:return {'n':0}
 def q(p):
  at=D(n-1)*D(p);i=int(at);return s[i]+(s[min(i+1,n-1)]-s[i])*(at-i)
 return {'n':n,'median':str(statistics.median(s)),'q1':str(q('.25')),'q3':str(q('.75')),'min':str(s[0]),'max':str(s[-1]),'aboveOne':sum(x>1 for x in s),'equalOne':sum(x==1 for x in s)}
def mean(s):return D(s['absoluteSum'])/s['n']
def ratios(stats):
 late=stats.get('late')
 if not late or late['n']<30 or mean(late)==0:return {}
 return {k:mean(v)/mean(late) for k,v in stats.items() if k!='late'}
def load_rows(c,kind):
 return rows_for(by_symbol[c['symbol']],kind)
def cadence(rs,ob,lo,hi):
 ts=[r['fundingTime'] for r in rs if ob+lo<=r['fundingTime']<ob+hi]
 gaps=[ts[i]-ts[i-1] for i in range(1,len(ts))]
 if not gaps:return None
 m=statistics.median(gaps);nominal=min([3600000,14400000,28800000],key=lambda h:abs(h-m))
 return {'medianHours':m/3600000,'nominalHours':nominal/3600000,'allInteriorGapsWithin3s':all(abs(g-nominal)<=3000 for g in gaps),'gapCount':len(gaps),'maxGapHours':max(gaps)/3600000}
windows=['day 1','days 2-3','days 4-7','days 8-14','days 15-30','first 7 days']
contracts=[];changed=[];placebo_changed=[];reconciliation=[];rawcount=0;rawplacebo=0
for c in query['results']:
 sym=c['symbol'];o=old[sym];rr=load_rows(c,'real');pr=load_rows(c,'placebo');rawcount+=len(rr);rawplacebo+=len(pr)
 assert len(rr)==len({r['fundingTime'] for r in rr})
 for k,st in c['first1000'].items():
  expected=o['late'] if k=='late' else o['win'].get(k)
  if expected is None or st['n']!=expected['n'] or abs(mean(st)-D(str(expected['abs_mean'])))>D('1e-14'):
   reconciliation.append({'symbol':sym,'window':k,'original':expected,'laterFirst1000':st})
 r=ratios(c['real']);p=ratios(c.get('placebo',{}));r0=ratios(c['first1000'])
 if len(pr)>1000:
  pa=c['onboard']+200*DAY
  full_base=[D(x['fundingRate']) for x in pr if pa+30*DAY<=x['fundingTime']<pa+90*DAY]
  first_base=[D(x['fundingRate']) for x in pr[:1000] if pa+30*DAY<=x['fundingTime']<pa+90*DAY]
  checks={}
  for name,days,key in [('day 1',1,'placebo_d1'),('first 7 days',7,'placebo_w1')]:
   early_values=[D(x['fundingRate']) for x in pr[:1000] if pa<=x['fundingTime']<pa+days*DAY]
   if early_values and first_base and sum(abs(x) for x in first_base)>0:
    first_ratio=(sum(abs(x) for x in early_values)/len(early_values))/(sum(abs(x) for x in first_base)/len(first_base))
    checks[name]={'originalRatio':o.get(key),'laterFirstPageRatio':str(first_ratio),'completeRatio':str(p.get(name))}
  if len(full_base)!=len(first_base):placebo_changed.append({'symbol':sym,'firstPageBaselineN':len(first_base),'completeBaselineN':len(full_base),'ratios':checks})
 early=cadence(rr,c['onboard'],0,7*DAY);late=cadence(rr,c['onboard'],30*DAY,90*DAY);all90=cadence(rr,c['onboard'],0,90*DAY)
 if c['real']['late']['n']!=c['first1000']['late']['n']:
  changed.append({'symbol':sym,'originalLateN':c['first1000']['late']['n'],'completeLateN':c['real']['late']['n'],'originalLateMean':str(mean(c['first1000']['late'])),'completeLateMean':str(mean(c['real']['late']))})
 first7=[x for x in rr if c['onboard']<=x['fundingTime']<c['onboard']+7*DAY]
 freq=collections.Counter(D(x['fundingRate']) for x in first7)
 mode=max(freq.values()) if freq else 0;modes=[str(v) for v,n in freq.items() if n==mode]
 contracts.append({'symbol':sym,'onboard':c['onboard'],'ratios':{k:str(v) for k,v in r.items()},'placeboRatios':{k:str(v) for k,v in p.items()},'stats':c['real'],'placeboStats':c.get('placebo',{}),'earlyCadence':early,'lateCadence':late,'whole90Cadence':all90,'modalWeekRates':modes,'weekDistinctN':len(freq),'firstReturnedLagMinutes':bar[sym]['returnedLagMinutes'],'archivedLagMinutes':bar[sym]['originalRecordedLagMinutes'],'sessionQuoteVolumeUsdt':bar[sym]['quoteVolumeUsdt'],'archivedSessionQuoteVolumeUsdt':str(bar[sym]['originalSessionQuoteVolume'])})
out={'sourceKind':'Later historical requery of the fixed original514-contract cohort; not contemporary raw capture.','originalCapturedAt':original['captured_utc'],'originalReportedCensus':{k:v for k,v in original.items() if k!='contracts'},'methods':{'windows':'UTC half-open elapsed intervals from archived onboard timestamp: [0,1),[1,3),[3,7),[7,14),[14,30),[0,7)days; baseline[30,90). Placebo offset200days, baseline[230,290).','quantity':'Mean absolute fractional funding rate PER SETTLEMENT; not variance, cumulative cash flows or an equal-time rate across variable intervals.','quartiles':'Linear interpolation at(n-1)*p; median averages middle pair.','completeQuery':'Ascending API pages, inclusive request end, next start=lastTimestamp+1; exact half-open analysis filter.','sessionVolume':'Whole UTC daily sessions intersecting[onboard,onboard+7days), including portions outside exact elapsed week. Quote currencyUSDT, no historical USD conversion. No returned bars is missing, not zero.','lag':'Earliest returned5mbar in query beginning3daysbeforeonboard, not established first executed trade.','stableCadenceSensitivity':'Every observed interior90day gap within3000ms of one nominal1/4/8hour median-derived gap; no guarantee about unobserved history or exchange rule changes.','dependency':'Cohort observations overlap calendar periods and underlying market factors; no independent or causal sample guarantee.'},'firstPageReconciliationDifferences':reconciliation,'paginationCorrections':changed,'rateRecords':{'real':rawcount,'placebo':rawplacebo},'windows':{w:summary([D(c['ratios'][w]) for c in contracts if w in c['ratios']]) for w in windows},'placebo':{},'contracts':contracts}
for w in ['day 1','first 7 days']:
 pairs=[(D(c['ratios'][w]),D(c['placeboRatios'][w])) for c in contracts if w in c['ratios'] and w in c['placeboRatios']]
 out['placebo'][w]={'real':summary([a for a,b in pairs]),'placebo':summary([b for a,b in pairs]),'realGreater':sum(a>b for a,b in pairs),'equal':sum(a==b for a,b in pairs)}
out['placeboPaginationCorrections']=placebo_changed
week=[c for c in contracts if 'first 7 days' in c['ratios']]
vol=[c for c in week if c['sessionQuoteVolumeUsdt'] is not None]
cut=statistics.median(D(c['sessionQuoteVolumeUsdt']) for c in vol)
hi=[D(c['ratios']['first 7 days']) for c in vol if D(c['sessionQuoteVolumeUsdt'])>=cut];lo=[D(c['ratios']['first 7 days']) for c in vol if D(c['sessionQuoteVolumeUsdt'])<cut]
out['volumeSplit']={'thresholdUsdt':str(cut),'included':len(vol),'missingSymbols':[c['symbol'] for c in week if c['sessionQuoteVolumeUsdt'] is None],'high':summary(hi),'low':summary(lo),'pairComparisons':len(hi)*len(lo),'highGreater':sum(a>b for a in hi for b in lo),'equal':sum(a==b for a in hi for b in lo),'meaning':'Descriptive activity association; not order-book depth or evidence refuting liquidity mechanisms.'}
out['signs']={'contracts':len(week),'earlyNegativeN':sum(c['stats']['first 7 days']['negativeN'] for c in week),'earlyN':sum(c['stats']['first 7 days']['n'] for c in week),'lateNegativeN':sum(c['stats']['late']['negativeN'] for c in week),'lateN':sum(c['stats']['late']['n'] for c in week),'moreNegative':sum(D(c['stats']['first 7 days']['negativeN'])/c['stats']['first 7 days']['n']>D(c['stats']['late']['negativeN'])/c['stats']['late']['n'] for c in week),'lessNegative':sum(D(c['stats']['first 7 days']['negativeN'])/c['stats']['first 7 days']['n']<D(c['stats']['late']['negativeN'])/c['stats']['late']['n'] for c in week),'equal':sum(D(c['stats']['first 7 days']['negativeN'])/c['stats']['first 7 days']['n']==D(c['stats']['late']['negativeN'])/c['stats']['late']['n'] for c in week)}
out['sensitivity']={}
groups={
 'allObserved90dayGapsConsistent':[c for c in contracts if c['whole90Cadence'] and c['whole90Cadence']['allInteriorGapsWithin3s']],
 'similarEarlyLateMedianGap':[c for c in week if c['earlyCadence'] and c['lateCadence'] and abs(c['earlyCadence']['medianHours']-c['lateCadence']['medianHours'])<.5],
 'differentEarlyLateMedianGap':[c for c in week if c['earlyCadence'] and c['lateCadence'] and abs(c['earlyCadence']['medianHours']-c['lateCadence']['medianHours'])>=.5],
 'archivedLagWithin6h':[c for c in contracts if abs(c['archivedLagMinutes'])<=360],
 'twoOrFewerWeekValues':[c for c in week if c['weekDistinctN']<=2],
 'moreThanTwoWeekValues':[c for c in week if c['weekDistinctN']>2],
 'listed2019Through2021':[c for c in contracts if dt.datetime.fromtimestamp(c['onboard']/1000,dt.timezone.utc).year<=2021],
 'listed2024Onward':[c for c in contracts if dt.datetime.fromtimestamp(c['onboard']/1000,dt.timezone.utc).year>=2024]
}
for name,group in groups.items():out['sensitivity'][name]={'cohortN':len(group),**{w:summary([D(c['ratios'][w]) for c in group if w in c['ratios']]) for w in ['day 1','first 7 days']}}
hour=archive['hourlyRows'];venues=['Binance','Okx','Bybit','Bitget'];tol=D('1e-9');anchor=D('.0001')
out['hourlyArchive']={'n':len(hour),'firstUtc':hour[0]['time'],'lastUtc':hour[-1]['time'],'strictToleranceFraction':str(tol),'counts':{v:sum(abs(D(str(r['fundingRate'+v]))-anchor)<tol for r in hour) for v in venues},'simultaneous':sum(all(abs(D(str(r['fundingRate'+v]))-anchor)<tol for v in venues) for r in hour),'latestPercent':{v:str(D(str(hour[-1]['fundingRate'+v]))*100) for v in venues},'meaning':'Archived hourly averages of indicative8h rates within tolerance; not exact equality or separately observed settlement payments.'}
sent=next(c for c in query['results'] if c['symbol']=='SENTUSDT');sr=load_rows(sent,'real');groups=collections.defaultdict(list);transitions=[];prior=None
for i in range(1,len(sr)):
 gap=(sr[i]['fundingTime']-sr[i-1]['fundingTime'])/3600000;clock=round(gap)
 assert abs(gap-clock)*3600<=3
 if prior!=clock:transitions.append({'endingTimestamp':sr[i]['fundingTime'],'utc':dt.datetime.fromtimestamp(sr[i]['fundingTime']/1000,dt.timezone.utc).isoformat(),'elapsedDays':(sr[i]['fundingTime']-sent['onboard'])/DAY,'previousObservedGapHours':clock});prior=clock
 if sent['onboard']+30*DAY<=sr[i]['fundingTime']<sent['onboard']+90*DAY:groups[clock].append(D(sr[i]['fundingRate']))
out['sentClockCheck']={'observedTransitions':transitions,'lateRatesByPreviousObservedGap':{str(h):{'n':len(v),'absoluteMean':str(sum(abs(x) for x in v)/len(v))} for h,v in groups.items()},'interpretation':'First observed1hgap occurs aroundday76, not31. Changing interval is not the sole demonstrated cause of the low ratio: later4hrecords also have a different mean magnitude. Not proof of exact exchange-policy effective time.'}
out['inputArchiveSha256']=hashlib.sha256(INPUT.read_bytes()).hexdigest()
out['originalSourceHashes']=archive['originalSources']
out['sourceKinds']=archive['sourceKinds']
out['requeryStart']=min(e['retrievedAt'] for c in archive['queries'] for e in [*c['real'],*c['placebo'],c['sessions'],c['firstReturned']])
out['requeryEnd']=max(e['retrievedAt'] for c in archive['queries'] for e in [*c['real'],*c['placebo'],c['sessions'],c['firstReturned']])
parser=argparse.ArgumentParser(description=__doc__);parser.add_argument('--write',action='store_true');args=parser.parse_args()
target=REPO/'products/athenum-analytics/apps/marketing/public/research/listing-funding-review-20261023.json'
if args.write:target.write_text(json.dumps(out,ensure_ascii=False,indent=2)+'\n')
else:
 published=json.loads(target.read_text());assert published==out,'Public calculation projection differs'
print(json.dumps({k:v for k,v in out.items() if k not in ['contracts','methods']},indent=2))
