Six months in, I had a strategy that looked good on paper and a live account that disagreed. The gap was eleven basis points per trade, consistently, in the same direction.
Where it went
Paper fills assumed the mid. Live got the far touch plus impact. On a strategy with a thin per-trade edge, that difference is the whole edge.
The fix
I rebuilt the paper engine to replay recorded order books against the decisions the strategy made, rather than pricing at the mid. Simulated results dropped immediately, which was the point: they finally matched reality closely enough to make decisions from.
The lesson I keep relearning is that the strategy is rarely the problem. The infrastructure around it is.