We handed a candidate a messy module from a system we had inherited and asked how they would improve it. Almost everyone in that seat proposes a rewrite.
This person came back with one small change, a way to verify in production that it actually worked, and a short list of things they would deliberately not touch yet, because they couldn't tell what still depended on them.
We hired them. We had learned that lesson the slow way: rewrites in our past that were genuinely cleaner and still spent months earning back the trust the old thing already had.
So now we ask every candidate what they would leave alone, and why. On engagements that run for years, restraint is most of the job.
If you have ever brought someone in to fix an old system, you probably noticed whether they started by changing it or by understanding it.