How do you check whether a futures backtest's edge survives out-of-sample data?

Backtests on futures look convincing surprisingly often, and I keep wondering how many of those results would hold up on data the parameters never saw. A single in-sample curve tells you very little on its own, since with enough tunable settings almost any series can be fit to look good. One common approach is to split the history into a development window and a held-out window, then judge the rule only on the second. Walk-forward testing extends this by re-optimizing on rolling windows, though the number of trials you ran before the final result matters just as much as the split itself. If you tried fifty variants and kept the best, the reported Sharpe ratio is inflated in a way the test window alone won't reveal. Some questions I'd want answered before trusting any such result: how many observations and independent trades sit behind it, what the base rate is for similar rules across comparable contracts and periods, and whether costs and roll adjustments were modeled. How do you decide how many trials or parameter combinations you count when adjusting for multiple testing?