I like it. Exploration vs exploitation is used in a lot of optimization algorithms (and in reinforcement learning) meaning that there's a low probability that the agent just decides to make a random choice instead of making a greedy choice.

That would ensure also diversity of staff. Very good idea.

Why do you think that? Isn't it rather sexist/racist to argue this?

The candidate stream is itself not "diverse" (assuming you are using the normal meaning of the word of lots of women+racial minorities). Simply randomly making offers wouldn't alter the outcomes because those people aren't being de-selected by skills testing to begin with - unless you believe the 'diverse' candidates are less skilled programmers.