Spent some time scumming precrowns on three characters to see what fun new artifacts are available in 3.0.x and discovered something quite strange: over hundreds of tests they were getting identical sets of new artifacts but clearly different sets of the previously existing ones. Unless I've hit some fluke of probability, the 3.0.x artifacts seem to be automatically added en masse to the end of the random artifact pool rather than being drawn from it, which is not really consistent.