# #743 first5.db: winner-linked early buyers (analyst3, 10 Oct)
Script: work/analyst3/f5.py, f5b.py. Pools = first5.db pools with first-120 s buyers (v3 5290, v4 4266; v2 not in pool_paths, no label). Buyer = swap recipient (v3) or tx_from (v4). Label = pool_paths max_cap. Feature at ~2 min after birth: number of distinct buyers who were also early buyers (first 120 s) of a pool that reached the label >6 h earlier (walk-forward, 6 h lag). Wallets seen in >MAXP pools dropped as bots/infra.

P(max_cap>=1M) by number of winner-linked buyers (0 / 1-2 / 3+):
- MAXP 10: all 3.9% / 28.8% / 50.5% (n 9005/333/218); v3 4.4/22/38; v4 3.4/39/67.
- MAXP 40: all 3.5% / 17% / 20% (n 8078/767/711).
- no wallet filter: 4.3/8.2/5.6 (infra dilutes it). P(>=100k) MAXP10: 21% / 56% / 76%.
Rate is across the pool_paths sample (not all births, runner-heavy): lifts (7-13x) are real, absolute rates are not population rates.

Caveats: (1) cap at the 2 min signal not checked; if the pool already jumped, entry multiple is small (same issue as #713; analyst2 replay of cap-at-signal pending). (2) wallets with <=10 pools that buy winners early may be insiders/deployer clusters = the signal is "insider recognised", fine for entry but not a smart-money edge. (3) 3 days of data, one regime.
Next: analyst2/testmgr replay entry at the 2nd winner-linked buyer print with cap<=300k; tp5 multiple vs matched pools without.