# Audit: h3n / d0fd native 30-day replays (#657), auditor 10 Oct
Source: tm2-native-replay-10oct.md, /root/tm/replay/{h3n,d0fd}_nat/trades.csv, arm A.
- h3n A: n233, mean 6.97x, MEDIAN 7.31x, mean capped at 10x 5.85x, capped at 5x 3.40x, ex-top5 6.02x, zeros 43 (18%), <1x 59. Not outlier-driven: median is above mean. Positive on 17 of 17 days except day 20721 (n1) and 20718 (1.2x).
- d0fd A: n60, mean 3.74x, median 4.68x. Winners are all exactly 4.676 = TP fill at +5x less costs (cap by design, no outliers); 12 zeros (20%). ex-top5 3.66x.
- Looks-too-good flags: h3n mean is 33% above sim (5.2x); median 7.3 is a TP/time artefact worth a look: median 7.3x with 18% zeros means most trades hit a big pump in 600s. Needs hand rebuild of 3 trades (not yet done) and fill/impact check: cap_buy_usd up to 53k looks like pool volume, not our size; reserve impact NOT modelled (per testmgr2).
- Zeros counted as 0: ok. Sellability unproven (Ben $5 sell test #525/#528 still waiting: no MATCH pool, no Ben reply). Outside-seller check not done.
Verdict: NOT VERIFIED (stays "replay-supported"). Open: (1) sellability sell test, (2) hand rebuild of 3 h3n trades from Sync logs, (3) impact at our real size, (4) forward paper agreement.