# Audit 4: h3n native replay v3 (#692, auditor2 10 Oct)
Code /root/tm/replay/h3n_nat2/h3n.py read; trades.csv re-counted by me.
- Docstring: FIXED ("prior net buys"). OK.
- TP/stop latency: FIXED. TP/stop sell takes the first swap at trigger+5 s (lines 80-83), pull checked up to that fill, TP no longer capped at 5. OK. Arm C 2.51x (n80, 8 zeros) is the honest exit number.
- Rule row recount (reserve>=0.1, fill<=10 s, arm A): n 80, 78 time exits, 1 pull, 1 quiet. Matches testmgr2 (4.53x, 2 zeros).
- New flag: the arm A time-exit net_x values cluster at 4.92-5.04 (middle 10 of 80 sorted: 4.92..5.04). Time exit (not TP) landing on ~5x again and again means the swap price at +595 s sits at a plateau, i.e. a deterministic or operator-set price level. Possible: price is quoted off a pool that is already drained, or the last swap is a fixed-price operator trade. If outsiders cannot sell at that price, 4.5x is a mirage; tell why it clusters.
## Verdict: replay arithmetic OK, still NOT VERIFIED
Open: (1) explain the 5x plateau at +595 s (sample 3 pools: reserves at +595 s from chain, can a 0.04 ETH sell get ~5x?); (2) Ben $5 sell test (sellability); (3) hand rebuild of 3 trades from chain; (4) replay vs forward overlap.