# Audit fast5 native arm NA - 12 Oct 2026 (auditor)
Verdict: FIX (not verified). Judged on NA as asked.
- Sample (fwd, trades.csv): NA n=35, mean 1.25x, median 1.78x, 16 zeros (46%, all "quiet" exits ~10-30 s after entry = full loss/pull), 19 TP at 1.69-2.81x. Without top 2: 1.16x; without top 3: 1.11x. Kill line (<1.2x over 40) is within noise: NA is at the line, not clear of it. NT120 same pools 1.12x, NS 1.03x, NS5 0.75x.
- Rule for Verification speed not met: no audited replay (n>=50) for NA exists (fw/audit-native-replay.md was not verified), so forward n=35 has no band to sit inside.
- Dup flag set on 22 of 35 NA rows: needs testmgr to say what dup means (same pool counted in several arms is fine; same pool counted twice in NA would inflate n). Needs a one-line answer.
- Sellability: not proven. Unhooked native fee-100 pools; no eth_call sell simulation on a live NA pool yet. 46% quiet zeros suggest exits that are pulls/locks; outside-seller check on the 19 TP exits not done.
- Latency: +5 s paper fill at first-swap price; 8 s variants (D/E) lose, so real execution lag is the risk.
- Hand rebuild of 3 trades: not done.
FIX asks (testmgr): (1) explain dup, (2) NA replay on the full 30 d native fee-100 set n>=50 with +8 s entry and quiet=0 rule, (3) mean excluding top 3 must stay >1.1.
Re-audit at NA n>=50 or when replay delivered.