Maybe also build in a reliability check detector?

[fsingletonthorn]

In a similar vein, also could try to ID explicit manipulation checks?

TBH, the reporting of both of these things may not allows us to get to sufficient accuracy outside of the narrow psych domain