Add feedback to the model

There are at least two types:
1. _Quality of the predictions_: we've given lists of predictions to our partners, and they have told us whether they agree. For example, we've flagged some officers as high risk that the department thinks is not.
2. _Quality of the explanations_: we give four or five reasons for a particular flag. They sometimes tell us whether they agree with the explanations.

Our model should take that feedback into account and should weigh the feedback by that individual's past feedback quality (i.e. if we say an officer is low risk but a supervisor thinks the officer is high risk then the officer has an adverse incident, we should weigh that supervisor's feedback more heavily. Item response theory provides useful guidance when thinking about this.)

We can start to model with the feedback they've given us, but down the road the interface should make it easy for them to provide the feedback.