@darius i dont read many papers any more, but by happy coincidence, the last one i read *was* about bot detection: OpenAI's "did this text from from ChatGPT" classifier. And they genuinely and helpfully *did* break their results into both false-positive and false-negative rates, rather than hiding behind the more typical reporting of "it is a classifer" or "it has 99% accuracy [on a dataset with 98:2 class imbalance]".