I'm sorry to hear that! Feel free to email me (patrick@stripe.com) and I'll connect you with the team if you'd like us to do a deeper dive.

But, yes, part of the intent here is to enable us to achieve better ROC[1] in our models and to block more fraud while also encumbering fewer false positives. From our testing, it's very clear that these bot-detection techniques do substantially improve the accuracy when compared to other, coarser heuristics.

[1] https://en.wikipedia.org/wiki/Receiver_operating_characteris...