I found, that nothing else works.
I found, that nothing else works.
I'm not sure what you think you'll learn in one day like this. You're throwing someone into a high-pressure peer-programming session with someone they don't know and a codebase they've never seen. In order to get them to complete anything you're probably going to have to stage a bunch of stuff and basically give them a fake assignment anyway ("fill in the body of this 'findUserByOrderNumber()' method for us") which has turned it into the equivalent a whiteboard coding question anyway. If you give them a single pair-programming partner, you only get one hiring opinion which is a pretty bad idea, and if you give them multiple pair-programming partners the work will have to be even more trivial.
I don't think a one-day trial like this will tell you more than a standard interview loop. Some people might do better with it, others will do worse.
I also don't think you should expect to get real work out of someone before you hire them, either. If you want me to spend 8 hours building you a feature, I expect to get paid. And the overhead of paying me for 8 hours of work is probably not worth it to your company, because it's a ridiculous amount of paperwork. (And not even possible for people who need visa sponsoring.)
This has grown on me. It's a lot of time invested per company, but it's still a hell of a lot less time than the interview prep mentioned in the article. It's also more pleasant, since I'm programming something real instead of the fibonacci sequence.