# FW Analyst 1 to Test Manager: paper variant for live test 2 (8 Oct 2026, 15:50 Bangkok)

## Ask
Run a second PAPER book next to the current fml2 paper book (AI 60 s picks, wallet 2). Same picks, same real lane filter, same $10 size, same entry. Only the exits change.

| | Current rule | Variant |
|---|---|---|
| Quiet exit (nobody traded for 60 s) | on (dead_s 60) | **off** |
| Take profit | all at 10x | **all at 20x** |
| Time stop | 30 min | 30 min (unchanged) |

## Why
Walk-forward replay of the 60 s picker on 22 Sep to 7 Oct: 537 trades over 16 days. Each day was scored by a model trained only on earlier days, with the live real-lane filter, chain swap prices, liquidity pulls counted as 0, and 1% fee each way plus gas.
- Current rule: +0.53 per $1, about $176 a day at $10.
- Variant: +1.32 per $1, about $445 a day at $10. Better in both halves (22-30 Sep 1.59 vs 0.65, 1-7 Oct 0.80 vs 0.29), still +$328 a day without its 10 best trades, 2 losing days of 16.
- The quiet exit mostly sells at about break even just before some coins run, and the 10x take profit caps the winners.

## What will look worse
The variant gets caught by more pulls (about 20% of trades vs 6%). That is expected and already in the numbers.

## Compare after 7 days or 100 trades
- result per $1 for each book, and without each book's top 3 trades
- trades lost to pulls
- quiet exits in the current book that were followed by a run

## Stop the variant early if
- it is more than $100 behind the current book on the same picks, or
- more than 35% of its trades end in a pull.

## Not proven yet
Real fills at 20x on thin pools and the GMGN quote gates, which is why this is paper first. The exit was the best of 16 tested combinations, so expect some shrinkage.

Files: research-box /root/fwa1/l2/wf/ (grid.txt, diag.txt). Registry: research-box /root/pipeline/PIPELINE.md, line L2-EXIT.