Calling YOLOv5 a "fraud" is a bit harsh. It has many excellent aspects for practitioners: easy to use, fast inference time, scalable model architecture, and it has many helpful utilities built-in for model deployment. In my experience, in real use-cases the models achieve about the same precision / recall / mAP as well as "state of the art" methods that report better stats on benchmarks.

All YOLOs past v3 will be, to some extent, since the original author wont be releasing anymore versions. This one looks to be in pytorch, so it is also not in darknet. I am not really sure what makes it "YOLO" anymore other than being a single-shot detector, but the author claims they took inspiration from the techniques in the original YOLO papers.

Comparing confidence metrics of the networks themselves is like comparing two athletes by asking them each how good they are and declaring athlete B the winner of the race because he thought he was better than athlete A thought about himself.