Don't you think that bacteria exist though, and don't you think that metagenomic assemblers like MEGAHIT produce valid contigs for bacteria even if the sample contains other organisms? Then why would you need to separate out viruses from the sample?

USMortality's second-longest contig was 16,024 bases long, and had it had a 99.2% match with "Leptotrichia hongkongensis JMUB5056 DNA, complete genome" with 16,023 overlapping bases (so is that contig not valid either?):

brew install blast seqkit;curl raw.githubusercontent.c… sort -lr final.contigs.fa|seqkit range -r 2:2|blastn -db nt -remote -outfmt '6 qseqid sseqid pident length qlen slen qstart qend sstart send stitle'

You wrote: "The Fan wu sars-cov2 genome from there own dataset can only be recreated with short read aligners , taking segments from reads and seeing if they match the reference genome chopping off the first 30 base pairs that dont fit and the last 70 and selecting the 50 in the middle that do". However in this post, didn't MEGAHIT also recreate the genome apart from the last 30 bases of the poly(A) tail and the two extra bases at the beginning? It didn't need any reference genome.

You also wrote: "I can also find the sars-cov2 genome in datasets going back to 2016 using the same software". Can you post your code, and which datasets are you talking about?