# a1 #2295: mc5 exit replay with rug cost, one rule (13 Oct 2026)
Rule tested: **stop 0.2x + 4h time exit** (instead of stop 0.5x + 4h), on all 59 closed live mc5 buys (fm_mirror mc5.db x v3live swaps; script work/analyst1/mc5x.py). Fill = min(stop, first swap price at/below stop), so gaps count. Rug-aware: no stop hit and (LP remove >=50% or pool ETH <20% of entry) in 4h -> 0. Holdout: the 59 trades span only 7 days (6-13 Oct), so holdout = newest half (30 trades) judged against the older half.

| set | live | stop 0.5 (sim) | **stop 0.2** | stop 0.1 | no stop |
|---|---|---|---|---|---|
| ALL n=59 mean | 1.63 | 1.74 | **1.87** | 2.83 | 2.76 |
| ALL w/o top 3 | 0.62 | 0.67 | **0.76** | 1.54 | 1.47 |
| older half n=29 | 2.48 | 2.62 | 2.38 | 4.41 | 4.36 |
| older half w/o top 3 | 0.39 | 0.45 | 0.18 | 1.83 | 1.77 |
| **newest half n=30** | 0.82 | 0.90 | **1.37** | 1.29 | 1.22 |
| newest half w/o top 3 | 0.38 | 0.44 | **0.21** | 0.12 | 0.04 |

Rug cost: no-stop ends at ~0 on 48 of 59 trades (median 0.00x). Stop 0.2 leaves 4 near-zero (vs 8 live); stop 0.1 leaves 7.

Verdict: NOT verified, no rule to ship.
- Stop 0.2 beats 0.5 by +0.13x overall, but in the older half it is worse (2.38 vs 2.62), and without the top 3 it is worse in BOTH halves' recent view (newest 0.21 vs 0.44). The gain is a few runners that dipped to 0.3-0.5 and then ran, not a repeatable edge.
- The big numbers (stop 0.1 / no stop, 2.8x) are the 3 giants (35x, 35x, 22x). In the newest half they are gone and everything is 1.2-1.4x with w/o-top-3 near 0.
- Sim caveat: it cannot fill a stop when the pool is drained without a swap, so stop columns are optimistic on rugs (live had 8 near-zero fills, the 0.5 sim shows 0). Real stop 0.2 would be worse than shown.
- Same answer as a1-1993 / a1-2102: exit width does not fix the leak; the leak is entries (shadow last 3 days 0.43x). Existing paper arm (stop.1+PULL+8h) keeps running; read it at forward n>=15 vs newest-half band 1.2-1.4x (w/o top 3 ~0.1). No new arm.