# Audit v3ce (auditor3, 18 Oct BKK) — VERIFIED
Arm E. Replay n286 mean 1.50x (lo80 1.24, up80 1.75). Forward n40: mean 3.07x, lo80 2.12, ex-top1 2.71, ex-top3 2.11 (lo80 1.41), last20 3.56x. 20 pulls (0x), 20 trail exits, 0 time exits.
1. Sellability: v3g transfer-gate sell sim 21/21 pass earlier (audit-v3ce-cr-kb-18oct.md).
2. Outside sellers: for all 20 trail exits (v3live.db swaps, sells by non-entry signers within -30..+60 s of the exit swap at >=0.9x exit price): 20/20 have >=1, all >=0.2 ETH sold (e.g. 34 sells/2.5 ETH, 22/5.9 ETH, 14/5.2 ETH). Script: v3ce_chk.py.
3. No look-ahead: entry = first swap >=15 s after trigger within 60 s (script v3ce.py); trigger uses only swaps up to itself; pull flag before entry excluded.
4. Fill rebuild: independently re-derived entry price and trail exit from swaps for all 20 trail trades: 0 mismatches (net_x = exit px/entry px x 0.9605 cost). Top 6: 17.35x, 14.0x, 13.7x, 12.2x, 9.3x, 8.6x all reproduced.
5. Replay vs forward: forward ABOVE the replay band (3.07 vs 1.75), on the conservative side; pull share 50% fwd vs 59% replay. Not a bug: same code, same data, rebuild exact. Cause is a strong runner stretch, not a leak.
Verdict: VERIFIED. Size/expect from replay 1.50x, not 3x. Wins are fat-tailed (median 0.31x, 20/40 zero), so watch last30 vs 1.0 in the daily drift sweep.