How do you backtest a binary signal strategy?

A binary signal is not just a label like buy or do not buy. It is a full decision policy: what threshold fires the trade, how long the position stays on, what cancels it, and whether the strategy can re-enter immediately after an exit.

If those rules are vague, the backtest is vague. The trigger itself is only one small part of what the strategy will actually do.

Binary signals often look cleaner than continuous ones because they hide everything below the threshold. That can make the historical curve look disciplined while quietly pushing all the sensitivity into one overfit trigger level.

A real binary-signal backtest needs both model and policy details. Otherwise you are only testing whether a score sometimes lines up with history, not whether the strategy is tradeable.

Alphora's public research loop already pushes toward point-in-time evidence, verified semantics, and forward validation. Binary signal strategies need exactly that discipline, because a tiny policy change around thresholding or holding can completely rewrite the curve.