# Test Manager review: $5 live micro-probe on fleet coins (Analyst 1 brief fw/a1-to-tm-probe.md), 8 Oct 2026

**Verdict:** worth doing. It is the outside money test done with real money, and the answer decides whether a whole family of paper books (the dumb v3 book, A2-V3P204) is real or fake. Ben decides (D-030).

**What Ben must decide first:**
1. **Which wallet.**
   - The only wallet the GMGN key sees is fmmc5's (D-021). Running the probe there mixes its P&L and exposure into the live test, and fleet coins can leave unsellable tokens behind.
   - Options: (a) a new self-signed hot wallet like D-026, funded with the probe budget only (my pick); (b) fmmc5's wallet with separate accounting (simpler, but muddies the live test).
2. **Budget and cap.**
   - Expect to lose most of it: 88% of these coins get pulled on paper.
   - $500 is 10 tickets at $50, and it would count in the org loss cap (D-017) together with fmmc5's drawdown. A bad probe week could stop the live test too.
   - My pick: **first 50 coins at $5 ($250 max at risk)**. That is enough to measure the deciding number (pulls within 10 s of our buy, bar 20%) to about ±11 points. Keep it outside the fmmc5 counts. Add 50 more only if the first 50 are unclear.
3. **GMGN key load.**
   - About 80 launches a day means about 160 swaps plus quotes on the shared key fmmc5 sells through.
   - The probe must use the same limiter and back off before fmmc5 does. A sell of the live test must never wait behind a probe buy.

**Build (after Ben's yes, Test Manager):**
- Its own service on firstmin, a copy of fmmc5's code with the probe rule (no cap filter, entry 30 s after the first swap, 20x / 0.3x / pull / 4 h).
- Its own db and dashboard tab. Its own stop when the budget is spent or 50 coins are done.
- The next day, the harness replay of the same coins goes next to each live trade (Analyst 1's pass / fail bars).

**ETA:**
- About 3 h build plus a 1 h dry run with the buy disabled (why: fmmc5 code reuse; waits on Ben's wallet choice).
- Funding and the switch: Ben only.