"My definition of evidence is any data-point that indicates indicates you better have a good explanation"
Ok, but for me that definition leaves too much room for mischief.
"the consultant [nature.com] who ran the software has used it to get a a previous researcher from the same University for faking evidence with copied images in the past"
Yes, and it found this too: "the 2013 Food and Nutrition Sciences paper was retracted, with a citation of “self-plagiarism”. However, the journal noted that the results were still valid and that it considered the issues an “honest error” -- Nature.com
But even with those results it doesn't change the fact that the software isn't enough to go on by itself because it produces way too many false positives to be able to rely on its results alone.
"A good explanation should be fairly trivial for the article's authors to come up with (you aren't supposed to trash all the data you use to write a paper just because it's been published), if they are actually innocent"
If you read the link to La Rebublica in Nature: "according to the Italian newspaper La Repubblica, Infascelli said that there is no substance to these allegations, and that an expert that he consulted about the papers had ruled out the possibility of data manipulation" (Nature.com), it appears the rector's investigating committee has consulted an expert on images and the expert said there was no evidence of any problems. Nature seems to have confused things a bit and its in fact the committee that's consulted the expert.
What I'd like to know, and what Nature doesn't tell us, is the content of the leak to the press: "details of the confidential findings of the investigation committee — composed of scientists in and outside of Naples — were leaked to the Italian press" (Nature.com).