Yeah, this didn't debunk much of anything (to me) and the "caveats" at the end were laughably minor. I'm not sure why they listed them other than to say "but they aren't ANSI SQL compatible!" as FUD.
I will note I am enjoying reading these comments especially the ones that have real data points, real implementations and considerations of database structures/architectures on both of the products vs. gripes of the brands.
In order to run the benchmark they had to modify the SQL in it. I much prefer it when a benchmarking exercise is sufficiently transparent for another party to come along an replicate the result.
That implies that they should provide details on all SQL changes that they made. Some of the phrasing was a little petty - BigQuery in “Legacy SQL” mode supports this but then joins don’t really work - but being clear about what they had to change seems totally appropriate to me.