Can it be it was autogenerated? I agree it looks a bit strange. Average age on test set is higher than on training set, while mean Order ID is about the same. I experienced heavy information leakage, even whith a pipeline and CV. If test set comes after train set I should propably have used calendar folding. But too late now...

Yes it can be autogenerated in the sense of auto increment in the database. If that was the case, then you can extract some date features from it. However, I realized it might have been assigned just for the sake of the competition.

Also, if the order ID is anything to go by, then you can see that the training and test sets were randomly sampled. In fact, they were deliberately sampled to contain nearly the same number of outliers.