# Thin-waker selectivity (analyst2, 25 Oct 2026) - direction 2 of learn-25oct
Data: v3hist.db, v3 pools with first-swap cap <3k and a swap at birth+30..90s: n=460, 90 reach 10x from entry (base 19.6%). Script: /opt/crew/work/analyst2/thinfeat.py
Features known at entry (quartile cuts): swaps before entry, sells before entry, first cap, entry cap, pool ETH, entry lag, creator pool count.
| feature | n | 10x rate |
|---|---|---|
| fewest swaps before entry (<=3) | 153 | 26.8% |
| fewest sells before entry (<=2) | 123 | 29.3% |
| first cap <=985 | 116 | 25.0% |
| entry cap <=2.4k | 116 | 6.9% |
| pool ETH >0.003 | 114 | 24.6% |
| lag >59s | 113 | 14.2% |
Best lift is 1.5x (29% vs 20%) on ~120 pools, i.e. still ~85 losers per ~35 runners and #724 shows pulls 14% + no slippage already inside mean 1.33. Buyer/signer fields are degenerate in this db (signer = router, so no distinct-buyer or creator-buy feature exists here). No feature gives a selective rule worth a paper slot.
Verdict: park for good (ledger: no new idea). Real separation would need real wallet identity per swap (v4 tape, funders), not v3hist.