# a1 #1993/#2000: mc5 runner exits with rug model (12 Oct)
Script: work/analyst1/r1/rug.py (live mc5 closed trades n=51 x v3live swap path; PULL events from mc5.db log = exit at 0.05x; 24 of 51 pools had a PULL).

| rule (4h time exit) | mean | ex-top3 | ex-top5 |
|---|---|---|---|
| base stop .5 (5% slip) | 1.93 | 0.68 | 0.42 |
| stop .3 | 1.90 | 0.66 | 0.28 |
| stop .15 | 2.34 | 0.82 | 0.29 |
| stop .1 (10% slip) / no stop | 2.94 | 1.44 | 0.78 |

Pull model and sell-haircut (x0.5 or x0 below 0.2x) change nothing: dead pools are worth ~0 either way, and the no-stop win comes from the 4-6 runners that dip to 0.12-0.4x BEFORE peaking (peaks 8-17 h later: 35x, 22x, 35x, 12x dipped to .14/.12/.16/.37).
Bootstrap (trades resampled): no-stop minus base = +0.94 median, p10 +0.17, P(>0)=94%.

**But not robust in time:** first 25 trades (6-8 Oct) base 2.47 vs nostop 4.63; last 26 trades (9-11 Oct) base 1.41 vs nostop 1.30 (no gain, ex-top2 .45 vs .34). All the gain is four whales in the first half. Median trade goes 0.45 -> 0.05 (39 of 51 end <0.1x), so it is a pure lottery on rare whales; the 4h time stop also cuts runners whose peak is at 8-17 h.
Also cannot see: whether outside wallets could sell at 0.1-0.2x in deep dips (live fills would be worse than swap-level).

Verdict: promising direction, not live-ready. Cheap to prove forward: paper arm on the SAME mc5_pat entries with stop 0.1 (or none) + PULL exit + hold 8-12h, paired against the live stop .5. Needs ~15 forward runners-days; decide by n>=40 trades. Sent to testmgr. No live change.