# Audit batch u: h100, h100s, busy60, v3005, fast5 - auditor2, 10 Oct 2026 (items #856-#860)

Four of five were audited in batch t (fw/audit-batch-10oct-t.md); nothing new has changed their blockers. v3005 is new.

## v3005: NOT VERIFIED (replay holds, forward and sellability open)
Replay n218 tp5 2.82x (/root/tm/replay/v3005_30d). Checks with v3hist.db (script /opt/crew/work/auditor2/v3a.py):
- Look-ahead: rule uses data up to entry only (read earlier). OK.
- Cap fills (108/218 tp at 4.851x): crossing swap is always a buy, median 456 s after entry, flow around it is organic-looking (median 67 buys / 33 sells in 20 min, 0 pools without sells). Pool reserve at the crossing: median 1.09 ETH (p10 0.84), always > 0.4. BUT our sell of ~0.2 ETH into ~1.1 ETH is ~15-18% impact, not the 2% modelled. Impact-adjusted (r/(r+v)) tp5 mean = 2.47x (median 3.68x). Still above 1.3x, so the edge survives; the test's number is ~12% optimistic.
- Signer: 77% of swaps in v3hist have signer NULL, so outside-seller / operator checks cannot be done; tx.from needed from dataeng for the 218 pools (entry..exit window).
- Sellability on v3 fee-100 pools: not simulated (WETH-quote UR sim limit, see batch t). v3 has no transfer hook for plain tokens but token-level blocks are unchecked.
- Forward: 0 trades. 37/218 pulled (17%) counted as 0 - fine.
- Ask testmgr3: use depth-aware sell impact (r/(r+v)) instead of flat 2%; post forward n. Re-audit when forward n>=10 or tx.from lands.

## busy60: unchanged, NOT VERIFIED (forward n>=30 open)
Sellsim 6/6 + hand rebuild passed earlier. Forward arm A n=7 mean 1.66x (replay A 1.53x n85, Z 0.98x). Verify automatically on re-audit when A n>=30.

## fast5: NOT VERIFIED
Main arm A n=2 (needs 40, ~14 Oct). Sellability and 5 s latency unproven (WETH-quote sim limit).

## h100 / h100s: NOT VERIFIED
Hook-registry gate unproven (needs real buy->sell on a fresh pool, or dataeng registry write history #851). Forward h100 A n=13 mean 2.24x; h100s A n=2. Need n>=40.