# Brief for the test manager: live test 2 (AI picker, real lane) — outside money check (FW Analyst 2, 8 Oct 2026)

## Status: keep live test 2 in PAPER. Do not switch it live yet.
Ben's rule (8 Oct): every test must pass the outside money test before it goes further. Live test 2 has not passed it yet.

## What live test 2 is
- Bot fml2 on firstmin 178.128.20.136 (/opt/fm/fm_l2.py, l2_config.json, l2.db), wallet 0xb8eba78637744887ffc06034037d49e8c14e5505, paper mode (no /opt/fm/LIVE_L2 file).
- Buys the AI picker's 60 s picks (aipaper book), real lane only, $10 a trade, max 4 open. Sells: 10x take profit, or no trades for 60 s, or 30 min.
- Paper so far: honest board real lane +$613 per $200 trade (72 coins since 5 Oct); its own paper book 13 closed, mean 1.85x.

## Outside money test, first result (picked coins that are in the Oct 5-7 chain day files: 97 picks, 33 real lane)
- GOOD: no rug-on-buy. After an outside buy in the real-lane coins, the pool was pulled within 5 s 0.0% of the time (n 3,424). This is NOT the launchpad trap that killed the 10-second flip (there: 67-88%).
- GOOD: real outside money exists in these coins (10% of buy ETH real lane, 26% all lanes; the flip coins had 0.04%).
- BAD: outside wallets did badly in the SAME coins at the SAME time. Outside wallets whose first buy was 40-120 s after launch (like the 60 s pick), real lane: n 317, nothing back 35%, mean 0.47x, median 0.41x, profitable 15%. Those who sold within 30 min: n 206, mean 0.72x, profitable 22%. 31% of these coins were pulled within 30 min of the buy.
- Same group, all lanes: n 926, mean 0.72x; those who sold within 30 min: mean 1.20x, median 1.02x, profitable 52%.
- Gap: paper says about 4x per trade on real lane; real outsiders in those coins averaged under 1x. Outsiders do not follow our exit rule, so this does not kill the test, but the gap is too big to go live on paper alone.
- Caveat: only 33 real-lane coins overlap the chain data (day files end 7 Oct).

## What to do (in this order)
1. Keep paper running. Add per trade, from the chain: the real outside buys and sells in that pool within +/-30 s of our paper buy and our paper sell (wallets NOT in the operator fleet = wallets trading 20+ pools a day). Log their price next to our paper price.
2. Pass bar for the outside money test (pre-registered), on 30+ paper trades:
   a. pull within 5 s after an outside buy: under 5% (now 0.0%);
   b. real outside sells exist within 30 s of our paper exit on at least 80% of trades;
   c. our paper exit price is within 15% of those real outside sells (median);
   d. paper mean still above 1.15x when every exit is priced at the real outside sell price.
3. Only if all four pass: $10 live, 20 trades, stop on 10 losses in a row (Ben's rule) or minus $100. The live fills are the final outside money test: log fill vs paper quote on every buy and sell.
4. If (b) or (c) fail: report to Ben and the pipeline (FW Analyst 1 and 2), do not go live.

## Pointers
- Test code and output: research-box /root/fwa2/edge/l2test.py, l2test2.py, l2test.out, l2test2.out; picked coins /root/fwa2/edge/aip_picks.json.
- Why this test exists: the 10-second flip looked like +44% a trade on paper and lost 90% of every outside buy on chain. http://159.223.88.168/hx-c9ois/fw/a2-flip-KILL.md