Sweeping a parameter — trying a moving-average length across a range of values, say, and keeping whichever one performed best — isn't a special operation in Reamer. It's an ordinary Python loop calling run_backtest once per combination and comparing the results afterward. That's a deliberate absence, not a missing feature: a dedicated sweep-optimization engine is a different kind of tool built around exploring a fixed parameter space efficiently, and that's not the problem Reamer is built to solve. The problem worth being honest about is what a loop like that actually proves.
The sweep itself is where the overfitting happens
Running fifty parameter combinations against the same historical window and keeping the best one isn't the same as finding a genuinely better strategy — it's closer to running fifty separate hypotheses and reporting only the one that happened to work, on data all fifty saw. The best-performing combination out of fifty candidates will look artificially strong even if every single candidate had a real, roughly equal edge, purely because the best of many noisy draws is expected to look better than any one of them individually. A sweep result reported without accounting for how many combinations were tried isn't a stronger backtest — it's a backtest with the multiple-comparisons problem quietly built into how it was produced.
best = None
for lookback in range(5, 60, 5):
cfg = StrategyConfig(lookback=lookback)
result = reamer_py.run_backtest(data, MyStrategy(cfg), exec_config, ...)
if best is None or result.net_pnl > best.net_pnl:
best = result
What actually earns trust here
The loop above finds a candidate. It doesn't validate one. The parameter that won the sweep still has to clear the same bar any single strategy would — Monte Carlo robustness testing against out-of-sample behavior, not just the in-sample number that made it win the sweep in the first place. Treating the winning parameter as already-proven because it beat forty-nine alternatives skips exactly the step that would catch whether it beat them for a real reason or because it happened to fit this particular slice of history best. The sweep narrows the search. It was never supposed to be the validation.
Full reference: docs · The workflow this maps onto: reamerlabs.com