Most grid bot backtests count one thing: how many times price crossed a grid line. Multiply by profit per grid, done. That number is almost always too optimistic. Here's what it misses, and how I model it in a simulator.

1. Fees are paid on every fill, not every round trip

A grid with a 0.3% step and 0.05% taker fee per side keeps only part of each step. With tighter grids the fee share grows quickly. The free calculator shows profit per grid after fees: grid bot calculator

2. Price leaving the range is the real risk

Inside the range a neutral grid collects small amounts. Outside it, the bot holds the whole inventory on one side, and the open loss can be many times the collected profit. A good backtest has to report that open loss, not just "grid profit".

3. Futures add funding and liquidation

Funding is paid or received on the position you hold, usually every 8 hours (some contracts settle every 4 or even 1 hour). Liquidation depends on leverage, maintenance margin and how much inventory the grid has built up, so it moves as the grid fills. I check it on every price update.

4. Fills aren't instant at the line

On fast moves price can jump past several lines between updates. The simulator walks the price path inside each candle and fills each crossed line in order, with slippage.

I put all of this into Botrader, a grid-trading simulator that runs on live public market data with a virtual balance. It never asks for exchange keys and never places orders.

If you've built your own grid backtests, I'm curious what else you model.