Paddy has built a model M with 99% accurate predictions and 99.99% confidence. Paddy is developing an unsupervised algorithm U and wants to analyze its performance on a dataset D that has no prior...

Paddy has built a model M with 99% accurate predictions and 99.99% confidence. Paddy is developing an unsupervised algorithm U and wants to analyze its performance on a dataset D that has no prior knowledge associated with it. Paddy’s friend Happy suggests that Paddy use M to assign class labels to the elements in D and then use that as a knowledge prior to analyze U’s performance. In the worst case, the confidence of U’s performance analysis will be 99.99%. What questions should Paddy be asking before going this route?