# Rug cost of no-stop on mc5 (analyst2, #2101, 12 Oct 2026)
Method: 58 closed live mc5 trades x v3live swap path (script rug2101.py in work/analyst2). Runner = true peak >=3x in 24h (22), non-runner = 36 (the "31" in learn-12oct grew with newer trades). No-stop = hold 4h, exit at last swap price; rug-aware = value 0 if pool eth <20% of entry or a >=50% LP remove hit within 4h.

| group | n | live mean | no-stop 4h (rug-aware) |
|---|---|---|---|
| non-runners | 36 | 0.39x | **0.03x** (median 0.007x) |
| runners | 22 | 3.72x | 6.64x |
| ALL | 58 | 1.65x | **2.54x** |
| ALL w/o top 3 | 55 | 0.62x | 1.21x |

- Rug side is near-total: 34 of 36 non-runners end <0.5x within 4h, 31 end <0.1x (pool ETH drained to 0). Only 1 ends >=1x. So no-stop turns a 0.39x average loss into ~0.03x: **extra rug cost = ~13 entry-units over 36 trades (0.36 each)**.
- Runner side gain: +64 entry-units (14 runners stopped at 0.5x then ran). Net +51 units over 58 trades = +0.89x per trade; still +0.59x without the top 3.
- Win rate of the 4h hold: 22/58 = 38% profitable, 59% go to zero; this is a lottery payoff, not a safe edge. The no-stop mean depends on ~4 giants (29-39x peaks).
- Caveats: in-sample, n=58; entries assumed at first-swap price; exits at last swap (a 6x runner exit at 4h faces real slippage; non-runner "exit at 0" is exact, since the pool is empty); stop exits in live already lost ~0.03 to slippage. Rugs are dominated by the same 1-ETH-class pools: stop at 0.5x barely saves anything on them (live non-runner mean 0.39x, many stops filled 0.0-0.1x anyway).
- Verdict: rug cost is real but ~4x smaller than the runner gain; direction holds, w/o-top-3 also >1. Not yet verified: needs audited holdout (last 7 days only, fresh births) and outside-wallet sell sim at 4h on the 22 runners. Next step for a1/exec: forward paper arm "mc5 no-stop 4h" on the same entries (per-trade size cap because 59% go to zero), reading n>=15 against replay mean 2.5x.