What metrics matter most for binary signal strategies?

A binary signal can be right often and still be poor if it fires on tiny opportunities, misses the best ones, or trades too often to survive costs. The central question is not just whether the trigger is correct. It is whether the decisions it causes improve the full strategy.

The useful metrics are the ones that separate predictive skill from threshold luck.

Binary strategies are often compared only to cash or to a naive always-trade baseline. A better check is to compare them to the underlying continuous signal. If the smooth score plus simple sizing does better, the binary trigger may just be throwing information away.

A serious review sounds less like 'the model is 61 percent accurate' and more like 'the trigger improves net returns in this score range, at this turnover level, with these false-positive costs.' That is the point where it becomes strategy research instead of classifier vanity.