# Audit s777 re-check at fwd n15 (#4112, auditor2, 27 Oct) - verdict: FIX (not verified)
Forward on-time arm A n15 mean 1.58x (median 1.56). Mean looks fine, but it is carried by the hooked pools:
| class | fwd n | fwd mean | replay n | replay mean |
|---|---|---|---|---|
| A1 (no hook) | 10 | 1.34x (4 quiet zeros, 4 tp) | 419 | 1.16x (80% band of 10 trades roughly 0.7-1.7) |
| H3000 (hook ...0cc, code hash 3030765c4f) | 5 | 2.06x | 90 | 1.68x |

Sellability (check 1): the 5 forward H3000 pools use the same hook code as the h3n/h3k 0xcc pools. Hook storage slot 4 = registry 0xdf4848...a856ae, the one that skimmed outsiders to 0.01% (fw/audit-v4-sellsim.md). Now: registry fe9fbb80(fresh address) = 0 and the pool deployer = 0 (not listed); 12 of 12 sampled sellers on each of the 5 pools are listed. So the H3000 trades (the best ones, 2.8x tp) are not sellable by us. Counting them at ~0, fwd n15 mean falls to 0.9-1.0x.
The old v4sim.py cannot replay these exits (public RPC has no historical state at exit blocks, 'EXC result'), so the fresh-address test was done at latest state on registry + hook code: that is enough for the gate, not a price quote.
A1 (unhooked) pools: no hook gate, but token-level transfer gates (d0fd "salg spaerret" type) are not excluded; token code hashes vary (5fcfacb98bd0, 9f03b2d2d10d, 13da8413f7db...). Not simulated. Also 4/10 fwd A1 are quiet/pull zeros.
Not done: hand rebuild of 3 fills, outside-seller check near exit on A1.

Verdict FIX (sent to testmgr2): exclude cls H3000 / any hook ending 0cc from s777 entries (paper must not credit unsellable pools), re-run replay as A1-only (expect ~1.16x, ex-top3 lower), then fwd A1-only n>=15 and re-audit with a token-level sell sim on A1 pools. At A1-only replay <1.05 -> KILL. Decide-by 30 Oct unchanged; registry verified stays false.