Is this result luck?
Search a strategy’s settings and Rearview counts every variation you tried, earlier runs and previews in the project included. It reports the Deflated Sharpe Ratio and the Probability of Backtest Overfitting with every input shown, then says it plainly: probably not luck, can’t tell, or probably luck. When a grid would be too big, genetic and CMA-ES searches look for you and count every try too.
- Any run can be set against 500 runs that enter at random at its own rate, and stress-tested: costs doubled, every signal a day late, entries missed, other start days.
- A lockbox holds back the latest years of data until you choose to look, and every result afterwards says it used them.
- A stability map shows whether the best setting sits on a plateau or a lone spike, and walk-forward tests pick settings on the past and trade them on what came next.