Hands-on technical assessment

An interview answer is a description of work. A lab is the work. That difference used to be small enough to ignore, because describing a Terraform state migration accurately was decent evidence you had run one. It is not decent evidence any more.

What a hands-on exercise gives you is the sequence. What the candidate checks first. What they reach for when the first attempt fails. Whether they read the error or guess past it. What they leave behind when the timer stops. None of that survives into a transcript, and none of it can be rehearsed from a question bank, because the state of the environment is different every time.

Designing one is harder than it looks. Too large and it measures typing speed. Too small and it measures nothing. One right answer and it measures whether the candidate has seen that answer before. Scoring is a separate problem again: if two interviewers cannot grade the same artifact the same way, you have swapped one argument for another.