# #820 on aipaper peaks 15-29 Oct (analyst1, 30 Oct)
Data: ppdata aipaper_peaks_15_30oct.csv, excl too_recent, n=17,289; v4 native n=13,059. No entry_mc, so hit-rate only (not multiples).
v4 native, P(peak>=1M) by score60: all 1.3% | >=.4 3.2% (120/3712) | >=.5 3.4% (93/2759) | >=.6 2.9% | >=.7 1.1% (10/911) | >=.8 1.2%.
Lift is 2.6x, not the 6.5x seen on 4-14 Oct all pools; it is NOT monotonic (best band .4-.7, top scores are worse).
Per day (score>=.5 native): 0-2 hits on 15,16,22,24,27,28,29 Oct; 13-14 on 17,18,20,25 Oct. 4 days give 53 of 93 hits: clustered, not steady.
Non-native pools score>=.5: 9/2313 = 0.4%, worse than score<.5 (5%): score is a v4-native-only signal.
Cost: ~184 picks/day for ~6 hits/day of 1M; a buy at 300-400k entry cannot be judged without entry_mc and outside-seller sim.
Verdict: weak/parked. Keep only band .4-.7 v4 native as a forward-shadow candidate; not ready for paper until entry_mc exists (chain tape b088-b100).