All of the thousands of assumed human ERVs are defective and produce no virus particles. Human ERV 'insertions' are 'assumed' events, not observed.
Read what I posted above: the pieces of genomes are considered ERVs because they contain sequences peculiar to RNA viruses, like reverse transcriptase. No assumptions here, except that what appears to be a viral genome really is one. This "assumption" is based on actually observed real-time infections.
It seems to me that the assumption that ERV infections occurred once and only once and can be 'traced' is a big weakness. You would need to show proof that ERV infections can only occur one in a population genome and then move to fixation in each species in which they occur. That's absurd.
Like it or not, their distribution is exactly what one would predict on phylogenetic grounds. Consider the fact that ERVs found in both species of Asian ape are inevitalby found in all species of African ape, including ourselves, but that the converse is not true.
If you think that you do not use 'any phylogenetic assumptions', you would be wrong.
You need to demonstrate this - where exactly was common descent used? The fact that the tree constructed from the ERV data matches the presumed phylogenetic tree is an observation that isn't going to go away and needs to be explained. Common descent is the obvious explanation - do you have a better one, one that will correctly predict the results of future gene sequencing?
It's not a viral genome because it doesn't produce virus particles.
You have sequences that you interpret as being the result of a historical insertion of a viral genome. That's all.
You assume common descent the instant you start 'tracing' a phylogenetic 'tree'. You have not shown that ERV infections happen only once and then move to fixation in any species, much less all of them. You merely assume such. That much is clear.
You don't even know that these are ERV's, you assume that because you think you find 'pieces' of a viral genome. You don't actually find a genome, you impose it on the data.