# #2138 wlink: rank winner wallets by own hit rate (analyst2, 12 Oct)
Script: /opt/crew/work/analyst2/r2138/w6.py (same replay as a2-wlink-44: v3 fee-100, 3 signal rule, tp5 sl.5 4h, 5% slip + 3% fee, controls x7.4, 12 Sep-6 Oct). Wallet score = later hits / (tries+3), using only pools resolved >6 h before (no look-ahead). "Top 20%" = score quantile among wallets with >=1 hit.
Net multiples, cap<150k (all caps in brackets); n = signals:
| rule | n | mean | halves |
| base K3 (current wlink) | 123 (206) | 2.30 (2.04) | 2.15 / 2.44 |
| base K2 | 166 | 2.32 | 2.28 / 2.36 |
| base K1 | 261 | 2.25 | 2.24 / 2.25 |
| top20 K1 | 137 | 2.25 | 2.28 / 2.21 |
| top20 K2 | 45 (143) | 1.74 (1.29) | 1.90 / 1.62 |
| top20 K3 | 34 (116) | 1.56 (1.20) | 1.53 / 1.58 |
| top20 two within 60 s | 43 (135) | 1.79 (1.32) | 1.96 / 1.66 |
## Verdict: ranking does NOT help; drop it
- Top-20% filtering cuts n by 3x and the mean DROPS (K2 1.74 vs 2.32 base; K1 same as base). Hit-rate rank is mostly noise: most wallets have 1-3 tries, so the score ties/overfits; edge is breadth of the winner pool, not wallet quality.
- Nothing beaten by "two top wallets in 60 s" (1.79). Base K2 = base K3 in mean with 35% more signals (166 vs 123), stable in both halves: the only variant worth a look, it adds flow not edge.
- Pulls as 0x: not measurable in this tape (pool_eth_after set on only 22% of swaps, liq table starts 6 Oct), so the numbers above are pull-free. The live replay (tm-wlink) shows pull-adjusted 1.82 vs 2.23 raw (x0.82); applying that to any row leaves the ordering unchanged.
- Suggestion (low priority, testmgr): shadow arm K2 (2nd winner wallet) cap<150k alongside A3-WLINK to get ~35% more forward flow; no ranking.