# Audit 3: h3n native replay v2 (#676, auditor2 10 Oct)
Read /root/tm/replay/h3n_nat2/h3n.py and testmgr2's table.
- Look-ahead: FIXED. R = net flow of swaps with t < entry t (line 53), entry swap excluded. The <0.1 vs >=0.1 split is knowable at entry. OK.
- Impact: applied (line 85, (1-0.04/(R+0.04))^2, R floor 0.01). OK. Docstring line 5 still says "NO reserve impact": stale, fix.
- Caveat: R is net swap flow, not true pool reserve (initial liquidity not counted). Fine as a rule, but call it "prior net buys".
- Exit: arm A exits at first swap price >=5x (tp, capped 5.0), -50% stop, or 4h; a TP trigger at a swap price assumes a fill at that price with no sell delay. Mean 4.99 on 168 is close to the 5x cap, so "no losers" look is by-design capping, not edge proof. The honest read: median 4.94 = most trades hit TP; fill<=10s 4.5x/2 zeros of 81.
- "exit = entry+595" applies to the timed arms only, not arm A (HOLD 14400); the note overstates it.
- Fill delay: on-time (<=10 s) is the real condition; fill>10s mean 3.2, median 1.3. Rule needs fill<=10 s too.
## Verdict: FIX (still not verified)
1. Fix stale docstring; report arm A with rule "prior net buys >=0.1 ETH AND fill<=10 s" (n, mean, median, zeros).
2. Add sell latency at TP (take price at TP swap +5 s, not the TP swap itself).
3. Still open for VERIFIED: sellability (Ben's $5 test), hand rebuild of 3 trades, replay vs forward paper on overlap.