# Audit s795 (auditor2, 28 Oct 20:05 UTC) - NOT VERIFIED (FIX)
- Replay (src=replay, to 03:18 UTC), arm A on-time: n797 mean 1.20x, exits quiet 440 / tp 336 / stop 19. Arm B 1.44x, C 1.35x (all arms include 0x quiet = pull).
- Forward (entries 04:36-19:49 UTC, after replay end, so no overlap): arm A on-time n17 mean 2.16x (ex-top3 2.02x), 14 tp / 3 quiet; late n4 (excluded). Std error 0.29.
- Divergence: forward tp rate 82% vs replay 42%; forward quiet 18% vs replay 55%. With p=0.42, 14+/17 tp is p<0.001. Either a lucky regime (13 of the last 14 trades were tp) or the forward exit/quiet logic differs from replay. Every tp exits at exactly 2.822x (tp3 minus fees) so no variance info: the mean is only a tp-count.
- Checks not done: no entry-time sell simulation on these pools (native unhooked v4: eth_call sell needed), no hand rebuilt fills (3), no outside-seller check. Replay vs forward overlap: none.
## Ask (testmgr2)
1. Explain forward vs replay quiet/tp rate gap (run the forward code on the replay window pools, compare exits pool by pool).
2. Add entry-time sell-sim column per forward row (same as s777 gate).
3. Re-audit at forward on-time n>=25 with those. Kill rule stays: mean<1.0 & upper80<1.05 at n>=20.