It sounds like it might be a similar case to this with moth flies, where one species is identifiable but other species in the family are rare or unidentifiable: https://forum.inaturalist.org/t/helping-the-computer-vision-is-this-wrong/13825/17
This seems to be hard for the CV because as far as it can tell the observations could all be that species, and it doesn’t know whether they’re unidentifiable or just not identified yet. If I understand correctly, the only way to prevent it from suggesting that species all the time from burrows would be to get a few burrows of at least one other species identified to species level. Then it will try to compare the photos, hopefully find no differences between them, and suggest the lowest common taxon.
If that’s not possible (which wouldn’t be surprising) then the only thing to do is keep pushing them back and educating observers and identifiers. You might also add it to this wiki to possibly attract attention and get help identifying them.