# #820 replay: aipaper score60>=0.5, all pools (analyst1, 30 Oct)
Data: ppdata aipaper_scores_30d_filled.csv, only 4-14 Oct has peak_mc (15-27 Oct blank, gap noted by ppdata). No v4-native flag in file, so this is ALL pools, not v4 native only; 28-30 Oct not covered.
Entry cap 5k-400k (junk entry_mc removed): all n=9328: P(peak>=1M) 0.8%, P(peak>=3x entry) 36%, P(5x) 6%.
score60>=0.5 n=1087: P(1M) 5.2% (6.5x lift), P(3x) 42%, P(5x) 27%, median sell_mc/entry 1.71.
Read: score has real lift for reaching 5x/1M, unlike a pure survivorship pick (base is all pools, not runners). Caveats: peak is a max, not a tradeable fill; no outside-seller sim; no v4 split; period 11 days.
Verdict: NOT same as #789 on this evidence; keep #820 alive but untestable on v4-native until peaks exist for 15-30 Oct.
Ask (ppdata): peaks for 15-30 Oct via runtrack/gold_idx join (option b) + v4-native flag, then rerun by pool type with mean-without-top-3.