Eups package astrometry_net_data co-opted the individual versions to hold different unique datasets (i.e. not just different versions of the same dataset).
Currently the different datasets are not backed-up nor are they installed in git repositories so the loss of a dataset thru misadventure or other failure requires the recreation of the dataset from whence it came.
This Issue should deal with setting up a failsafe archival mechanism.
Edit (by mjuric): As K-T points out in the first comment, this is true of all other test data sets. This issue is intended to cover the needed work for all of them.