Simulation. Not trading advice.
Inside the experiment
The price here is a random walk: each day's move is a coin toss, so no rule can have a real edge. Each variant is a plausible-looking trading rule (a moving-average cross, a breakout, a mean-reversion trigger) with randomly chosen settings. They are all noise strategies.
We fit on the in-sample window and keep the variant with the best Sharpe ratio. With one variant, that is a coin toss. With 324, the winner looks superb, because you picked the luckiest of 324 coin tosses. Then the out-of-sample window arrives. The winner's past luck buys it nothing: its Sharpe there is a fresh coin toss, sometimes up and sometimes down, averaging zero across seeds. Mostly it drifts inside the grey band, which is where random entries live.
The deflated Sharpe ratio (Bailey & López de Prado) corrects for that. It asks how good the best result would look from luck alone, given how many variants you tried, how much their results spread, and how many of them were really independent (near-duplicate rules count as fewer tries). Then it reports the probability that the winner beats that bar. Drag the slider and watch the bar rise as you search.
If a backtest does not say how many variants its author tried, I would assume it was a lot. "We tried 324 things and this one worked" is a very different sentence from "this worked".