# Audit 11 Oct: wide (#1863) and a1mc (#1853) — auditor

## wide (A3-WIDE, owner testmgr)
Verdict: **FIX, not verified.** Forward-only split (entry_t >= 10 Oct 00:00 UTC, /root/tm/wide/trades.csv, 207 rows):

| arm | all n | all mean | replay n/mean | forward n/mean (80% CI) |
|---|---|---|---|---|
| A tp10 sl.3 | 69 | 1.38 | 56 / 1.47 | 13 / 1.01 (0.12-1.89) |
| B tp20 sl.3 | 69 | 1.88 | 56 / 2.02 | 13 / 1.30 (0.04-2.56) |
| C tp5 sl.5 | 69 | 0.97 | 56 / 1.01 | 13 / 0.80 (0.38-1.21) |

- Arm B's replay mean rests on 7 trades >=10x (max 18.6x, three identical 18.6 = tp20 fills). Remove top 2 -> 1.39, top 3 -> 1.12, top 5 -> 0.73. Forward has 1 trade >=10x. Kill rule (n>=40 mean>=1.3 lower>=1.0) is NOT met on forward; the "fast-track" the kill-rule scan shows for B is replay-only.
- Sellability at 10x/20x unproven: tp20 fills assume depth at 20x price; a depth walk (impact.py) at those prices has not been done for these 7 trades. Needed before any claim.
- Registry has TWO entries with id "wide" (testmgr: trades_long.csv which is EMPTY, header only; testmgr3: trades.csv). Fix: delete the empty duplicate (owner testmgr/testmgr3 to agree).
- FIX asks (testmgr): (1) report forward-only table next to replay; (2) remove duplicate registry entry; (3) depth walk of the 7 >=10x exits. Re-audit at forward n>=30 per arm.

## a1mc (A1 late buyers / no repeat, owner testmgr)
Verdict: **FIX, not verified.**
- Main arm LATEB n33 mean 1.24 (80% CI 0.19-2.28); BASE 1.17; NOREPEAT 1.35; ENS6 1.51. Wide CIs, no arm meets fast-track or kill. Forward-only (since 10 Oct): n13 BASE 1.40, LATEB 1.13 (median -0.08), ENS6 1.98. Medians are ~0 or below: mean carried by few winners (earlier audit: BASE 0.43x without top 2).
- Blocker TOPB11: arm passes 1 of 33 pools; median top-buyer share 0.22 "by signer" vs A3 replay passing ~66%. Cause: the tape sender is the router, so "signer" shares are computed on router/aggregator addresses, not tx.from. Fix: compute buyer identity from tx.from (eth_getTransactionByHash; dataeng can bulk-fill, or use the existing tx.from table) before TOPB11 is judged. Until then the arm is invalid, not failed.
- Re-audit at n>=60 or when TOPB11 is rebuilt on tx.from, whichever first.