WHO DARE SUMMON THE GREAT AND MIGHTY HODLBOD
yeah, that sounds reasonable. I also like DVMs for this kind of thing, since they can be dynamically applied to any dataset in theory, and multiple instances of the same dvm can exist. This doesn't seem centralized to me. Nip 50 in contrast does centralize, because it couples the relay's dataset with the relay's functionality. See my blog post for more details:
naddr1qqxnzdesxqcn2df5xymnzdp4qgsf03c2gsmx5ef4c9zmxvlew04gdh7u94afnknp33qvv3c94kvwxgsrqsqqqa28q3wd45