# A1.18 mc5 exit test: hold longer or trail? (FW Analyst 1, 6 Oct 2026, 15:26 Bangkok)

**Verdict: hold 4 h, not 2 h. Every trailing stop loses.**
- The H1 pick was "time stop 4 h". Tested once on H2: **1.67x vs 1.35x for 2 h**, and **$946 vs $491 a day** at $100.
- Without the top 3 coins: 1.24x vs 1.00x.
- 3 to 4.5 h is a flat plateau in both halves. Past 5 h it falls apart, and 12 h loses (0.86x).
- Trailing stops (armed at 2x, sell at 0.5 / 0.6 / 0.7 of the peak) and "half at 3x, trail the rest" are all worse than plain 2 h. Swap-level prices shake out the winners before their big leg.
- **New rule: hard stop 0.5x, pull = 0, time stop 4 h, no take profit, no trail.** a1-mc5-spec.md is updated.

## Data
- Swaps plus Mint/Burn for all 338 band pools, now out to 12 h after launch: 835,246 more swaps, 749 Mints, 722 Burns. Store: research-box /root/fwa1/v3/v3ext.db.
- Trades with a full 12 h of data: 197 (H1 42, H2 155).
- The 4 trades from 6 Oct are out: under 12 h old. That removes the 35.6x coin, so the "2 h" numbers here are lower than in A1.14.
- Same entry and fills as A1.14: real depth from swaps, GMGN 1% + pool fee + gas, 3 s delay, sell capped at 90% of pool WETH, pull = Burn of 50%+ of minted liquidity units.

## H1 grid (pick made here; mean, without top 3, $ a day at $100)
| exit | H1 mean | w/o top 3 | $ a day |
|---|---|---|---|
| time 2 h | 1.87x | 1.16x | $365 |
| **time 4 h** | **2.56x** | **1.23x** | **$654** |
| time 8 h | 0.89x | 0.34x | -$46 |
| time 12 h (= hold, data ends at 12 h) | 0.36x | 0.28x | -$269 |
| trail 0.5 from 2x, 4 h / 8 h / 12 h | 0.79x / 0.78x / 0.77x | about 0.60x | about -$90 |
| trail 0.6 from 2x | 0.73x to 0.74x | about 0.61x | about -$110 |
| trail 0.7 from 2x | 0.73x to 0.74x | about 0.61x | about -$110 |
| half at 3x, trail 0.5 / 0.6 / 0.7 | 0.86x to 0.94x | 0.67x to 0.73x | -$27 to -$59 |

## H2 test (once)
| exit | n | mean | median | w/o top 3 | $ a day | green days | top 5 share of profit |
|---|---|---|---|---|---|---|---|
| **time 4 h (pick)** | 155 | **1.67x** | 0.48x | **1.24x** | **$946** | 9/11 | 93% |
| time 2 h (old rule) | 155 | 1.35x | 0.49x | 1.00x | $491 | 8/11 | 129% |

- Top 5 share over 100% means the other trades lose in total. At 4 h the rest of the book is close to break even instead of losing.
- Best trail on H2 (context only): trail 0.5 from 2x, 4 h = 1.49x, $695 a day, 5/11 green. That is worse than 4 h on every measure.

## Is 4 h a lucky cut? (time stop sweep, both halves; context, not used to pick)
| time stop | H1 mean / w/o top 3 / $ a day | H2 mean / w/o top 3 / $ a day |
|---|---|---|
| 2 h | 1.87x / 1.16x / $365 | 1.35x / 1.00x / $491 |
| 2.5 h | 2.00x / 1.24x / $422 | 1.52x / 1.07x / $739 |
| 3 h | 2.05x / 1.17x / $440 | 1.61x / 1.15x / $860 |
| 3.5 h | 2.15x / 1.07x / $484 | 1.82x / 1.28x / $1,151 |
| **4 h** | **2.56x / 1.23x / $654** | **1.67x / 1.24x / $946** |
| 4.5 h | 2.63x / 0.94x / $684 | 1.69x / 1.20x / $977 |
| 5 h | 2.23x / 0.81x / $515 | 1.33x / 0.95x / $466 |
| 6 h | 1.56x / 0.52x / $236 | 1.33x / 0.91x / $467 |

- Every stop from 2.5 to 4.5 h beats 2 h in both halves. 4 h sits in the middle of the plateau.
- Past 5 h, coins that ran fade and the 0.5x stop or a pull catches them.

## Fragility of the 4 h rule ($100)
| set | bootstrap under 1.0x | 5th pct | median | worst 3 trading days | worst day |
|---|---|---|---|---|---|
| 4 h, H2 | 0.1% | 1.24x | 1.67x | +$112 | -$721 |
| 4 h, all | 0.0% | 1.40x | 1.85x | -$457 | -$721 |
| 2 h, H2 | 1.1% | 1.09x | 1.35x | -$727 | -$767 |
| 2 h, all | 0.6% | 1.15x | 1.45x | -$727 | -$767 |

- 32 trades did better at 4 h than at 2 h, and 13 did worse.
- Biggest gains: 0x6e8a6a1af8ec53f3f09c7303e5403558dee42d0a 20.3x → 33.6x; 0xc8b416104e0b4bd5e5e14489f90ac807b67300e2 5.5x → 18.8x; 0xe1f227ad6f50b3c8905e75ed621610becf4510d9 8.1x → 19.7x.
- Biggest losses: 0xaf69d9d9d0dbeaaa2e03eddbd7971d5b536627ba 6.3x → 0.1x; 0xad4cfc3433988aa9495d94b4163f84e2422bb1d2 3.1x → 0.0x (pulled); 0xeb1646dbfbeacf76483df1092c208388c029c534 21.1x → 15.3x.

## Why trails fail here
- The trail is checked on every swap of 0.01+ ETH, and one big sell can drop the print 30 to 50% for a few seconds. A bot watching swaps sells into that dip and misses the run.
- A trail on 1-minute closes was not tested. It is the one variant that might do better. Not proposed until tested.

Code: a1code/v3ext.py, v3exit.py, v3exit2.py (research-box /root/fwa1).