Look-ahead bias: the number-one backtest trap

A backtest simulates past decisions. Look-ahead happens when a decision relies, even unintentionally, on data that comes after the supposed moment of the trade: the close of a candle not yet finished, a day's high known in advance, an indicator recomputed over the whole series.

It's insidious because it doesn't feel like cheating: you sincerely think you see a great signal, when you only “see” it because you already know what comes next.

Entering at a candle's close when you'd only have known that close at its end; using the high/low of an already-finished session; fitting an indicator (average, level) computed on future data; or cherry-picking, after the fact, the periods that “work”.

Survivorship bias is a cousin: testing only on assets that still exist today ignores those that disappeared, artificially inflating performance.

An ounce of look-ahead is enough to turn a neutral strategy into an apparent cash machine. The catch: that edge doesn't exist in the future, where information arrives in order, one candle at a time.

The outcome is always the same: flawless test results, then brutal disappointment live. Worse, the trader often concludes “it's psychology” when the backtest was flawed from the start.

The fix is bar-by-bar replay: the chart is frozen at the present, you advance one candle at a time, and decide only on what's actually visible. No access to the future is possible.

On GetBacktest, upcoming history is hidden server-side: even by inspecting the page, you can't see the next candle. That's a structural guarantee against look-ahead, not merely good intentions.