# Audit a1mc + fast5 (auditor, 10 Oct 2026) - #629

Data: /root/tm/a1mc/trades.csv (BASE n=27), /root/tm/fast5/trades.csv. 5-check audit NOT run in full: sample too small, so verdict is NOT VERIFIED for both (re-audit at the sample below).

## a1mc (A1-LATEBUYERS / NOREPEAT / ENS6)
- BASE n=27, mean 1.31x, but 25/27 are <1x: the mean is two winners (12.0x, 12.7x). Without the top 2 the mean is 0.43x; without the top 1 it is 0.87x. A fat-tail result on 2 coins is a bug until proven otherwise.
- LATEB/ENS6/NOREPEAT are re-weightings of the same 27 trades (same 2 winners), so they are not independent evidence. Their own kill rule says report at n>=150 / 60.
- Open checks: fills vs real v3live depth, outside sellers near exit, kit overlap with live mc5, deployer table starts 6 Oct. Entry is mc5-style +300 s, so it is a candidate duplicate of live mc5, not a new test.
- Verdict: NOT VERIFIED. Re-audit at n>=60 BASE with >=5 winners >=3x; then run the 5 checks.

## fast5 (enter 5 s after first swap)
- Arm A (the headline arm, WETH reserve>=1) n=2 (kill rule needs 40). B/Z5 n=2. Nothing to audit yet.
- Native arms: NA n=22 mean 1.11x (12 zeros), NS 0.70x, NS5 0.68x: below the 1.2x kill line already; native tp3 discard rule (<1.0 over 60) is heading that way.
- Skips: 79 of ~209 pools skipped for reserve, 67 native late>2.5s; lag filter makes live fill at +5 s unproven (latency, sell fill, sellability untested; no Ben sell test for unhooked WETH fee-100).
- Verdict: NOT VERIFIED, too few trades. Re-audit at arm A n>=40.