If you watch the Caltech talk (embedded in the blog post) or read the original published paper, they did quite a bit of work to validate that their method was actually giving them good results and not just what they wanted to see. The author of the post doesn't mention any of this, presumably because it doesn't fit her narrative.
