I do filter structural junk per backbone, but only from evidence that backbone itself supplies: synonym/accepted status, "not-a-taxon", environmental, hidden, unplaced flags. I also insert neutral buckets such as "classes (no phylum)" where ranks are skipped, so thousands of taxa don't spill into one flat level.

What I deliberately don't do is move a taxon to the parent that I think makes more biological sense. Ptree (Ptable gets the same kind of requests to classify metalloids) is downstream of academic taxonomic authorities operating under something like Wikipedia's "no original research" policy. It can expose disagreement among authorities, but it shouldn't quietly become another undocumented taxonomy.

Naldaviricetes is a good example. It is unquestionably a class of viruses, but "Viruses" is not necessarily its formal kingdom. ICTV currently leaves it unassigned to a realm, kingdom, or phylum so Ptree's ICTV backbone places it directly beneath a display root called Viruses. NCBI places it beneath "Viruses incertae sedis," while GBIF supplies it without that viral higher lineage, producing the confusing placement you found.

As a whole it's brilliant and humbling. Both the vastness and complexity of nature and biology on Earth, and also the absurd amount of work and human lifetime that has been spent on gathering the information. It's also a testament of the power of Wikipedia.