# v3005 audit (auditor3, 13 Oct) - NOT VERIFIED, retire at the 14 Oct decide-by
Forward tp5 (trades.csv, 16 unique pools, spread over 4 days): mean 1.61x, median 0.0x, SE 0.50, 80% band 0.97-2.25.
Outcomes: 9 pulls (56%) = 0x, 2 time exits 2.09x, 5 tp 4.16-4.51x. All 7 survivors are +; no middle.
Verified rule: replay band must contain forward mean. Replay 2.45x is ABOVE the forward upper 80% bound (2.25). Replay pull rate 17% vs forward 56%: replay under-models pulls, so 2.45x is overstated.
Lower 80% bound 0.97 < 1.0: forward does not even prove mean > 1. Test's own kill (tp5 < 1.3x over 30) is not reached at n16, but the fast-kill rule (n>=20 mean<1, upper80<1.05) is not triggered either: mean is positive but unproven.
Still open from before: tx.from/outside money, sell proof on the 16 forward pools not re-run (v3g.py available), pulls are 1 in 2 so a rug gate would be needed live, and 3-10 gated pools/day gives ~1 trade/day.
Money: ~1 trade/day x (1.61-1) x 0.04 ETH = ~0.02 ETH/day at best, with n too small to trust the sign; rug exposure 56%.
Verdict: NOT VERIFIED. Owner testmgr3: keep to 14 Oct as set, retire if forward does not reach n>=30 inside a band with lo80 >= 1.0. Re-audit at n>=30 only if it is still running.