They are pretty accurate, though of course when you are dealing with the person at the very top it's going to be hard to say how accurate it is. I think broadly though most experts agree that Hans beating Magnus was an unlikely but possible ting to happen. For comparison a 1400 rated player, which is someone who plays and studies a decent amount of chess but isn't devoted, would have a 0.0000014% chance of beating Magnus.

"So you're telling me there is a chance!"

I think as the skill differential becomes greater you have a better chance of identifying where the "master" screwed up allowing the neophyte to win. But it sounds like Hans and Carlsen are too close in skill (at least in this game) to identify a flaw in Carlsen's play that was able to be exploited.

And perhaps Hans went in expecting to lose and played loose and free and surprised himself with a win.

Also, it is likely that Hans is higher rated than he is, which would make his chances better. Chess rating is earned so you can be underrated.

I doubt they are very predictive in those rating bands. Based on a quick Google, since 2011, Magnus has lost a total of 20 games as white, mostly against much higher rated opponents than Hans. It had been almost 2 years since he lost as white, against Levon.