# Audit batch 11 Oct (auditor3): lv, wlink1, top1, v3ride, v3m1, caps, wide
Verdict: NONE verified. Triage from trades.csv (net_x). Full eth_call sell sim not run on v3 pools: v3 transfer-gate probe (v3g.py) still inconclusive, Danish-launcher-family lock risk (see audit-p204) unproven for these pools.

## Cross-cutting finding: the five v3 tests are one sample, not five
Winners (net_x>=2): lv 6, v3ride 6, v3m1 10, caps 7, wide 12 pools; union only 13 pools. 5/6 of lv winners are also v3ride winners, v3m1 contains all 6 of lv's. Median is 0.28-0.46 everywhere, win rate 13-20%. Means ride on 6-10 runner pools. Agreement between these tests is NOT confirmation.
Rows start 6 Oct = replay; forward-only is small. Every test needs the report to split replay vs forward (rule: forward decides).

| test | main arm | n | mean | median | 80% CI | verdict |
|---|---|---|---|---|---|---|
| lv | A | 38 (fwd 6) | 1.83 | 0.46 | 1.16-2.51 | FIX: split forward; hold to fwd n=30. Same winners as v3ride/v3m1 |
| wlink1 | A cap<150k | 72 (fwd 16) | 2.06 replay, 1.31 fwd | 1.02 / 0.87 | fwd ex-top3 mean 0.74 | FIX: replay shrinks to 1.31 fwd; hold to fwd A n=40. csv has no arm column: add it. Look-ahead OK in design (6 h lag, infra excluded) but winner list includes v3live signers; sellability unproven |
| top1 | log | 36 | 1.11 | 0.48 | 0.49-1.72 (1 pool = 33% of sum) | RETIRE: read-only log, no entry/exit rule, nothing that can make money; gap not measurable |
| v3ride | ALL | 26 | 1.31 | 0.28 | 0.84-1.78 | RETIRE as duplicate of v3m1 C5 (same entry, same pools) |
| v3m1 | C5 | 45 | 1.27 | 0.28 | 0.92-1.62 | FIX/hold: C10 1.81 (1.17-2.44) is the better arm but 8 wins; need forward n>=30 and sell proof at 5x-10x. A3 0.78 and B 0.94 below 1 |
| caps | A | 43 | 1.11 | 0.45 | 0.81-1.42 | FIX: no edge over baseline C 0.91 (diff within noise), thresholds fitted in-sample; forward 0 |
| wide | B (tp20) | 69 | 1.88 | 0.27 | 1.16-2.61 | FIX: B vs C baseline 0.97 looks real but only 8 wins in the same 12 pools; sellability at 20x unproven; forward-only split missing |

## What would make a PASS
1. Forward-only closed n>=30 per test with lower 80% bound >=1.0 (v3m1 C10, wide B, wlink1 A are the candidates).
2. One buy-then-sell eth_simulateV1 proof per pool family (v3 launcher): not built; I need a v3 pool where buy via router then sell from same EOA works.
3. Hand rebuild of 3 fills from chain: not done (no time-boxed data), do at PASS candidacy.