# Copy-exit replay on wlink / wlink1 (research #2137, 12 Oct)
Rule: sell when the signal winner wallets sell (fill at first print >=10 s after their sell). Same entries, fees (3% + 5% slip), arm A cap<150k, v3live prints. Code: research/copyexit/run.py (reuses wlink.py entry logic).
"pull-hit" = liquidity fell below 50% of entry liq before our exit. "pull-adj" = pull-hit trades counted as 0.

| test | exit | n | mean | lo80 | w/o top3 | pull-hit | pull-adj mean |
|---|---|---|---|---|---|---|---|
| wlink (K=3) | tp5/stop.5/4h (today) | 151 | 2.23 | 2.04 | 2.18 | 58 | 1.82 |
| wlink | copy: first signal wallet sells | 151 | 1.54 | 1.41 | 1.43 | 27 | 1.32 |
| wlink | copy: half of signal wallets sell | 151 | 1.66 | 1.51 | 1.52 | 36 | 1.36 |
| wlink | tp5 + stop + copy | 151 | 1.52 | 1.40 | 1.45 | 23 | 1.34 |
| wlink1 (K=1) | tp5/stop.5/4h | 85 | 1.90 | 1.65 | 1.80 | 39 | 1.46 |
| wlink1 | copy first sell | 85 | 1.44 | 1.23 | 1.23 | 22 | 1.21 |
| wlink1 | tp5 + stop + copy | 85 | 1.47 | 1.32 | 1.36 | 21 | 1.20 |

Mean peak of these entries 5.8x (wlink); copy-exit keeps 26% of peak, tp5/stop keeps 38%.
Finding: winner wallets are flippers: they sell early (copy exit fires in ~70% of trades, well before 5x), so we give up the runners. They do sell before pulls more often than we do (pull-hit halves), but the pull-adjusted mean is still clearly lower than today's exit (1.3 vs 1.8; 1.2 vs 1.5). No overlay variant beats the current exit, and the gap (0.5x) is far outside the lo80 bands.
Verdict: do NOT paper-test as exit. Possible narrow use only: none found. Not handed to testmgr2. Caveat: pull-hit is a liquidity-drop proxy, not a confirmed rug.