# a1: one exit rule for mc5, rug cost priced, holdout 12-14 Oct (task #2425)
Script exr.py (reuses research capture.py: 79 closed live mc5 trades x v3live swap path, 4h hold). Rug = live mult <0.05 or a single-swap drop >90% inside the hold -> exit 0.0 under every rule (13 of 79; 7 train, 6 holdout). Train = bought before 12 Oct (n=56), holdout 12-14 Oct (n=23). Rule picked on train only.

**Rule: no stop for the first 5 min, then stop 0.3 (4h time exit).** (5, 15 and 30 min delays gave identical results.)

| set | base stop0.5 mean / w/o top3 | rule mean / w/o top3 |
|---|---|---|
| train n=56 | 1.80 / 0.67 | 1.93 / 0.80 |
| holdout n=23 | 1.63 / 0.35 | 2.81 / 0.45 |
| all n=79 | 1.75 / 0.82 | 2.19 / 1.27 |

Rug cost is small (about -0.05 to -0.15 on the mean) because stops do not save you from gap rugs either; it does not change the ranking. Plain no-stop is a bit better on mean (2.67 all) but median 0.01 and w/o top3 0.31 on holdout: deeper bleed, rejected.
Caveats: swap-level replay, no slippage, fills at trigger (live stops fill ~0.03 worse, a 0.3 stop is a bigger gap risk in thin pools); the holdout gain rests on 3 giants (w/o top3 0.45 still <1); n=23. Direction yes, not a live change: needs shadow/dry run via exec before Ben sees it. Full table: a1-exit-rule-14oct.txt