Selectively replaying recorded execution boundaries while running new code live to test changes against past failures.