# a1-2246: v4chase pulls (analyst1, 13 Oct)
Data: /root/tm/v4chase/trades.csv arm B (n87, 27 pulls = 31%, mean 1.06) + /opt/sw/v4tape liq/swaps.
- Every one of the 27 pulls is a full removal of the SEED LP by the same address 0x58daec31 (the address that seeds all 671 A1 pools). It is not an outside LP: it is the launcher pulling its own seed 22-1470 s after first print (median ~250 s, entry is at ~100-180 s).
- Pull is NOT predictable from entry-time features: seconds since first print, prints, buys/sells, net ETH in pool, distinct senders (all ~5, router-level, and one sender = 94% of prints in every pool, so the ">=5 senders" gate is meaningless), pull vs no-pull medians are within noise (e.g. net ETH 8.0 vs 7.7, dt 107 vs 148).
- Shorter hold does not fix it (arm B, tp2/stop0.6, max hold): 120 s 0.90, 300 s 0.94, 600 s 1.03, 1200 s 1.08, 3600 s 1.09. Pull timing is too spread to dodge.
- Verdict: replay 1.49 assumed no pulls (cutoff 3 days, early pools); with pulls = 0 the edge is ~1.06 before real fees/failed sells, i.e. not money. Does not meet kill rule yet (n87, upper 80% bound >1.05), but I expect forward B < 1.15. Recommend: let it run to decide-by 17 Oct, retire if forward B (n>=15) < 1.15. No live lane, no rug_gate possible because the pull is the seeder's own habit.
- Only possible gate left: pools where 0x58da does NOT pull (it pulls ~31% by pool; need a per-pool tell, none found). Not worth a paper slot.