re: your last line:

That is the classic industrial ML advice, isn't it? Read a bunch of papers, look at the stupid obsolete method they all compare against, implement the latter.