# Audit hk50 re-audit (auditor2, 20 Oct 2026): VERIFIED (paper level; live still needs shadow n>=20, mean>=1.2)

Replay side was clean in fw/audit-hk50.md (sellable 102/103, no look-ahead, fills ok, outside buyers after touch).
Forward (src=fwd, /root/tm/hk50/trades.csv, entries after 18 Oct 06:00 BKK), arm D (tp3.3, stop .5, 6h):
- n=20, mean 2.93x (replay D band ~1.9-3.1): 19 tp (3.086x each), 1 quiet (0x). Last 10 mean 3.09x. Max entry_vs_touch 1.11x (no jump entries).
- On-time only (late=0): n=7, mean 3.09x. 13/20 signals were late (lag >30 s), same ratio as replay (49/103); the live lane will see the same lag, so judge it by the shadow.
- Other arms fwd: A n20 2.81x, B n20 2.67x, C n15 0.62x (C is not the candidate).
Concerns kept: one launcher kit (~7 pools/day, same seed choreography; edge dies if the launcher stops); 95% tp hit rate = capped payoff 3.086x, the risk is pulls/quiet (1/20 now, 8% in replay); late-flag share.
Verdict: VERIFIED for paper (policy: audited replay + forward n>=15 inside band). Not a live clearance: shadow of the live lane with rug gate must show n>=20 mean>=1.2. Sweep rule: unverify if last40 < 1.15 or pulls >15%.