Backtesting

A backtest replays history against your rules. Bar by bar, the engine feeds the strategy exactly the data it would have had at that moment, records the signals, simulates the fills, charges fees and slippage, and produces the equity curve and tearsheet that result.

What separates an honest backtest from a flattering one

No look-ahead: decisions use only data available at decision time, which means evaluating on the bar close, filling on the next bar's open, and confirming swings only after their confirmation lag. Real costs: fees, slippage, spread, and (for perps) funding, charged on every trade. Realistic execution: stops that gap, and size that moves thin books (AlphaProve walks reconstructed L2 depth where coverage exists). And enough data to mean something: multiple regimes and hundreds of trades, not one lucky quarter.

What a backtest can't do

It cannot promise the future. It can only disqualify strategies that failed the past. A good result is a hypothesis that survived one test, and the follow-up tests, forward-walk analysis and Monte Carlo simulation, exist because the past can be memorized. That failure mode has a name: overfitting.

On AlphaProve

The simulated exchange exposes its cost model as separate inputs — taker and maker fee rates, spread in basis points, market slippage, an extra penalty applied only to stop fills, optional order latency, and perp funding — so the same strategy can be rerun from an optimistic to a pessimistic fill regime with the net result moving each time. Where that leaves you on the four levels of backtest honesty is a choice you make per run, not a fixed property of the tool.