Research
The research behind the tool: 192 rule features on 25 CFDs, discovery frozen in April 2025, every rule run across the following year it had never seen, then checked again on the most recent quarter. Results are in R, risk units, with the trade counts and the out-of-sample figures stated every time. Nothing here is advice.
- One week of rule mining, end to end
What a weekly sweep of 18,620 single-condition rules on 41 CFDs produces, from the first evaluation to the sixty-odd that get a slot on a demo bot, with the survivors by symbol and by feature family.
2026-11-04
- What the rules did live, and the noise floor under every backtest
11,775 demo trades of mined rules over six months came to minus 0.109 R per trade, the same in both halves. The per-trade spread of outcomes, 1.81 R, is why a 20-trade result means nothing and a 0.2 R edge needs about 330 trades.
2026-11-04
- How a result is judged
The rules behind the honesty panel, the measured fill offset and spread factor, the evidence floor of 100 trades, the trades-needed line, the last quarter and the locked holdout, with the measurements behind each.
2026-11-03