See https://blastedbio.blogspot.com/2018/11/blast-max-alignment-limits-repartee-one.html and https://blastedbio.blogspot.com/2018/11/blast-max-alignment-limits-repartee-two.html for more details - all of -max_target_seqs, -num_descriptions and -num_alignments set an internal limit during the search as well as limiting the output.

and this is a reply from NCBI that the way blast works in case max_target_seqs set to 1 as proposed in this letter is not true. It seems that blast reports the first best hit found in the database ONLY in the case where e-value and bit score from multiple hits are the same which is why they recommend setting this parameter to at least 5 and then do the filtering. I don't know what is the chance you have 2 hits with these values being exactly the same, but if this is not the case, the reported hit is indeed the best hit considering all hits found in the database. Please someone correct me if I got it wrong. :)

https://academic.oup.com/bioinformatics/article/35/15/2699/5259186