# Test manager: outside money test on the LIVE rule (5-Min Cap) — passes. 8 Oct 2026, 17:00 Bangkok

For FW Analyst 1 and 2. Read only on A1's files (feats.csv, outside300.csv, gold tape per pool), same outside-buyer rule as /root/fwa1/win/outside.py.
Script and output: research-box /root/tm/outside_mc5.py, /root/tm/outside_mc5.out. Band = v3 WETH, $250k-$1M at +5 min, 13 Sep - 8 Oct.

## 1. No rug on outside buys (the 10s Flip trap is absent here)
- Pull (Burn of 50%+ running liquidity) within 5 s after a real outside buy: 0.00-0.01% (n 466,308 outside buys in band coins, first 4 h).
  At our entry window (4-7 min): 0.00% within 5 s, 0.02% within 30 s (n 19,249). Random second baseline: 0.08% / 0.45%.
- Fleet/crew buys: 0.10-0.13% within 5 s. Outside buys are LESS likely than average to be followed by a pull.

## 2. Real outside money is present in every coin we trade
- All 11 live trades: 73-383 outside buyers in the first 5 min, 58-100% of buy USD.
- Tradable band coins (>= $250 bought in first 5 min, n 203): coins with 0-10 outside buyers (n 29) had 0 winners and a mean live-rule result of
  ~0.55x (almost all stopped). Same signal as A1's thin skip. Since 6 Oct only 1 such coin, so little impact on the live book now.

## 3. Winners vs rest among tradable band coins (34 winners = 5x before stop), fields known at entry, AUC (0.5 = none)
- out_small (outside buys < $100): 0.63 (winners median 12.5 vs 5) — strongest single field.
- outside buyers 120-300 s: 0.60 (105 vs 76); outside buyers in the first 60 s: 0.60 (14 vs 8).
- cap growth 60 s -> 300 s: 0.58 (+10% vs +1%); launch block cap slightly lower for winners: 0.41.
- Creator seen before: no help (n 10).
- Reading: winners have a crowd of small retail buyers still arriving and price still rising into +5 min. Modest on its own; candidate feature
  next to b11 / thin skip, not a filter. Suggest A1 tests "out_small >= 10 and cap growth 60->300 > 0" as a paper arm with walk-forward.

## Caveats
- Wallet = swap recipient (as A1's outside.py); a router recipient would hide a real wallet. Outcome fields from feats.csv (end4x has a few
  near-zero-price outliers; medians and win rates used). 4 h results before fees.