I don't think a candidate who I put at ease then absolutely hammered would compliment me on putting them at ease would they? Or does your theory stretch to them somehow having short-term memory loss?
If the second group preformed better then presumably the testing methodology played a role with a higher percentage of people being able to actually program than your suggesting.
Granted it takes a lot of interviews to build a reasonable sample, but we aren’t talking about 1% better performance here.
I've tried both. I've tried leaving them to it entirely, I've tried leaving them to it then if they get stuck helping both from a lesser to a greater degree.
It really makes no difference - the bad coders struggle on that and other problems, they just lack the ability to think through the problem.
And again I've seen enough anxious candidates to see the difference. I'm pretty good at gauging things well as to how and when to help.
Overall I'm not hearing many specifics more 'I am great at interviewing'.
The empirical evidence is overwhelmingly that there are many bad programmers out there with long CVs. That's also my experience WITHIN companies, though a lesser proportion.
In terms of a test I use mean, median, mode, and range for an array. Then go into basic performance trade offs.