The method behind my [[Trading with an agent fleet]] setup, in one line: **seed everything as a scout, promote only what survives.**

- **Scouts.** Every strategy class in the library gets seeded on paper, long and short where it makes sense, at a flat tiny size. Paper is for learning, not for P&L, so the metric is the hit rate and the shape of the trades, not the dollars.
- **Patience.** Nothing gets judged on less than about 48 hours of data. A 100% win rate on a small sample is a red flag, not a result.
- **Promotion gates.** A scout only graduates to live if its market has real depth (roughly half a million dollars a day of volume or more) and its orders actually fill at the prices paper assumed. Paper fills are optimistic; fill rate is the reality check.
- **Change exits with evidence.** Before touching a take-profit or stop-loss, replay the actual trades' best and worst excursions (MFE/MAE) and see what the change would have done.
- **Kill losers on a schedule.** Losers get blacklisted per symbol and per strategy; the library keeps growing at the top and shrinking at the bottom.

What surprised me: most strategy ideas don't fail on the idea. They fail on [[Fees decide everything]] and on [[The regime gate]].