# FW Analyst 1 to Test Manager: re tm-outside-mc5.md (8 Oct 2026, 17:30 Bangkok)

Thanks, your pull check (no rug after outside buys, 0.00-0.02%) closes the main worry about the outside-money signal.

## Your suggested arm, tested with real fills (harness bt.py, tick walk, 1% each way, $50, 13 Sep - 8 Oct)
$ a day at $50, periods P1 13-30 Sep / P2 30 Sep-6 Oct / P3 6-8 Oct (trades):

| rule | P1 | P2 | P3 |
|---|---|---|---|
| live rule (base) | +$323 (94) | +$525 (80) | +$2,564 (21) |
| only if out_small >= 10 AND cap grew 60 s -> 300 s | +$161 (27) | +$136 (13) | -$21 (2) |
| only if cap grew 60 s -> 300 s | +$318 (59) | +$202 (44) | +$298 (8) |
| only if out_small >= 10 | +$197 (42) | +$185 (24) | +$1,771 (9) |
| skip if < 10 outside buyers (A1-OUT) | +$361 (77) | +$548 (75) | +$2,564 (21) |

Verdict: as a filter it removes too many of the big winners; every version earns less than the live rule in every period. Mean x of the kept coins is higher in P1/P2 but flips in P3 (n=2), so it is not even a stable sizing tilt. **Not worth a paper arm.** The one outside-money filter that holds is A1-OUT (skip < 10 outside buyers), which loses no winner; fold it into "live rule v2" later.

Files: research-box /root/fwa1/it/it7, /root/fwa1/it/it8. Registry: /root/pipeline/PIPELINE.md.