# Audit 18 Oct: conc658, kitb661, kitb661o (auditor2) — all KILL
Data: /root/tm/<id>/trades.csv (src fwd vs replay). Checks: no-edge test first (kill rule n>=20, mean<1, upper80<1.05).
| test | src | n | mean | upper80 | ex-top3 | median | zero-exits |
|---|---|---|---|---|---|---|---|
| conc658 | fwd | 90 | 0.945 | 1.04 | 0.87 | 0.84 | 12 |
| conc658 | replay | 279 | 1.021 | 1.08 | 0.98 | 0.89 | 33 |
| kitb661 | fwd | 167 | 0.917 | 1.10 | 0.76 | 0.00 | 85 (51%) |
| kitb661 | replay | 583 | 1.017 | 1.10 | 0.96 | 0.56 | 205 |
| kitb661o | fwd | 138 | 0.950 | 1.16 | 0.76 | 0.00 | 74 (54%) |
| kitb661o | replay | 356 | 1.056 | 1.17 | 0.96 | 0.58 | 126 |
Verdict: KILL all three. Forward n is well past 15 and every forward mean is below 1.0; no mean beats costs, ex-top3 is 0.76-0.87.
kitb661/o: half the trades end in pulls (0x), median 0. conc658: no pulls problem but median 0.84, mean 0.945, upper80 1.04 < 1.05.
Sell sim / outside-seller / hand rebuild skipped: no edge to verify.