# Audit res60 + round60 (auditor2, 16 Oct 2026)
Source: /root/tm/<id>/trades.csv, all src=fwd, all hook=0x0 (unhooked, so no hook skim; sellability by pool type OK in principle, depth proven only on winners).

## res60 (arm R, tp5 sl0.5 4h) -> KILL
- fwd n60, mean 1.07x. Its own kill rule: mean <1.1 at n60 -> met. 44/60 trades are 0.
- Exits: quiet 42 (counted 0; 2.05x if sold at last price), tp 13 (4.85x), stop 3 (0.41x), pull 2 (0).
- Sim expected 1.46 (n=356); forward 1.07 = far below, and cutoffs (3.6-6.2 ETH) came from quartiles of the same sample (selection risk). No look-ahead issue found, simply no edge.
- Action: testmgr2 retire res60 (registry retired + ledger killed).

## round60 (arm Q, tp5 no stop 20min) -> FIX / unverified
- fwd n63, mean 1.47x (kill line 1.0 passed). First half 1.17x, second half 1.77x (noisy, improving).
- Exits: quiet 37 (0, no sell depth: cap_sell=0; 1.40x if last-price sold), tp 11 (4.85x), time 13 (3.04x), pull 2 (0).
- Only 24/63 trades are winners and all of them have outside depth (cap_sell median ~2.3e5 vs buy ~5e4-6.5e4). 13/13 time exits >=3x average is odd for a 20 min exit: need per-trade check of price path.
- Open before VERIFIED: (1) 59% quiet exits cannot be told apart from silent pulls (v4tape has no ModifyLiquidity). Ask for liq removal on the 37 quiet pools. (2) Hand rebuild of 3 fills from chain (rb.py) not yet done. (3) No audited replay n>=50 (sim 1.53 n=263 is in-sample threshold). (4) Forward window is 7 days of history, check feed lag: late=0 on all rows, verify the flag works.
- Not ready for shadow/live. Re-audit when testmgr2 delivers (1) and (3), or at fwd n90.