That being said ...
>Convictions are supposed to be made after considering the facts of the case as presente
The fact that someone robbed 10 banks is a fact. You're literally witholding (relevant) facts from the case.
How is a person's history not a fact? I mean, the robbery happened in the past, which is also a fact. How are the two different.
So we rob a bank at T=10, Imagine that person robbed a bank at T=1, not relevant, what if person robbed a bank at T=2, T=3, .. etc. Even if we rob a bank at T=9.999... it is suddenly not relevant anyore?
I realize the absurdity of the point I'm making but how is previous crime not a valid feature? If we train a machine learning algorithm to predict guilty/not guilty (I'm going into dangerous territory here, I realize this) and the training set contains criminals that are repeat offenders, wouldn't "repeat offender" get a high feature weight in say a random forest?