My backtest assumed I could always get filled

Ran a mean reversion thing on a handful of alt pairs, looked fine in sim, then paper traded it live and the fills kept coming in worse than the model said, sometimes a lot worse. Spent a weekend assuming my timestamps were off. They weren't. That research piece on spreads and book depth put it plainly enough — the tight quote you see at the top isn't the price you pay once your clip is bigger than the first level or two, and on thinner venues a decent chunk of minutes just can't absorb a normal size at all. My backtest was filling at mid every single time, which is basically fantasy. Rebuilt the cost model, half the edge went away.