Computational stylistics is a broad term. Some of the techniques are sketchy, if not down right bogus. Others have been shown to be highly reliable. Also different techniques are applicable to different types of analysis. If the researchers know what they’re doing, it can be conclusive to a high degree of confidence; if not a good researcher will state the appropriate caveats. I wouldn’t trust newspaper articles to accurately assess the reliability of any particular study.
I remind you that fingerprints have turned out not to be unique
Is this the case? I thought the status was that the uniqueness of fingerprints had never been quantified in peer reviewed literature (not helped by the fact that the FBI, with the world’s largest collection of fingerprints, refuses to fund or participate in such research, presumably for fear that it might be found to be so), but that no one seriously doubts that fingerprints, when the technique is properly applied, is not a valid and useful method for high-confidence identification.
From what I’ve read, the cases where fingerprints have been tossed out of court have to do with inadequately trained collectors and analysts and the fact that there is no accepted standard for the number of points of comparison needed for a reliable identification—which goes back to the point that no one has ever systematically quantified the discipline. An identification with twelve points of comparison is almost certainly unique for all practical purposes, one with only four has a pretty good chance of matching several people.