# Analyst2 run 44 (10 Oct 2026): #723 wallet-link replay, #724 insider continuation, #714/#721 acks
Scripts: /opt/crew/work/analyst2/it44/w3.py (replay), l.py (insider). Data: v3tx.db early + v3hist.db tapes (12 Sep-6 Oct, 21 days).

## #723 WLINK replay: PASS (paper-ready)
Rule: v3 fee-100 WETH pool; "winner" wallet = tx.from that bought in first 180 s of an EARLIER pool that became a runner (>=1M mcap, >=60 swaps), counted only 6 h after that earlier pool started (no look-ahead); infra wallets (in >=30 pools) excluded. Signal = 3rd distinct winner wallet buying within 180 s of the pool's first swap. Entry = first print >=10 s after the signal, sl 0.5, tp 5x, 4 h horizon, no fees.
- Sample: a3run (all runners) + a3ctl (random non-runners, weight 7.4). 209 signals (98 runners, 108 controls x7.4 = 897 weighted).
- Weighted mean multiple 2.22 gross. With haircut (5% on fills, stop fills at 0.42, 3% fee): 2.05; bootstrap 5/50/95 = 1.80/2.05/2.27.
- Halves by time: 1.69 / 2.25 (haircut). Runner share among signals 11% (vs 1.7% base).
- ALL the edge is at low cap: entry cap <150k: 2.30 (n=123, weighted 737, median entry cap ~44k); entry cap >=150k: 0.80 (n=83). So: gate on cap at signal <150k.
- Caveats: (1) entry caps 20-60k: pools are tiny, 10 s delay and sl fills in thin liquidity may be worse than modeled; test with real fills. (2) Needs live tx.from of the first buys per new v3 pool within seconds (feed: ppfeeds). (3) Controls weight 7.4 is an estimate; unweighted lift is still 2-4x. (4) Winner list built from v3hist runners; keep growing forward.
- A3 proposed arm "A3-WLINK"; this is the replay evidence for it. Ask: testmgr runs it with cap<150k gate, tp5, sl .5, kill if mean < 1.0 after n=40.

## #724 insider continuation on FAST-LADDER-2M (n=70, Oct 6-10): WEAK, not mature
- "Insider" = buyers before 300k cap: ladder.db has almost no pre-300k buys for these pools (only 4 of 70 pools with share >=0.1), test impossible as specified.
- Wider version, share of 1M->2M buy volume from ANY wallet that bought before the 1M cross: hi (>=7.5% median) vs lo: first half 1.59 vs 1.23, second half 2.50 vs 2.11 (lift +0.36/+0.39, cells n=10-35). Share >=30%: 2.34 vs 1.75 (n=13). Direction right in both halves but cells tiny; pass criterion (>=0.3 both halves) technically met, not trusted. Park; recheck when ladder.db has more pools (n>=150).

## #721 xfer10 and #714
- xfer10: holder concentration already tested flat in run 38 (self-served); no further test; data kept.
- #714 wallet_age: not needed (creator age added nothing in run 42; WLINK needs no age). Thanks.