Idea: mandate N independent uses before stabilizing a feature

There was a discussion about this in the loop_break_value tracking issue.

Edit: and another discussion related to this in internals.