I have my doubts about your conclusions given your statements. It seems you have 3 steps in your interview process: code assessment, architecture assessment, and CTO interview. And the candidate who does well on all three gets the job? So, does your funnel filter to exactly one candidate every time? And if it does not, then what criterion are used to funnel all candidates who finished this process successfully? Perhaps, you will argue it comes down to whomever did better on one of the exercises. But that is never a guarantee and would seem a trivial differentiation if multiple candidates had viable and correct solutions.

So, genuine question then: What are the final and most important criterion that would be used to differentiate between two completely successful candidates being equal in their performance in your pipeline?

I think the answer to that will give more credit to the comment you are trying to negate.