RunBoth: find out what your refactor actually changed

Field notes

The page describes a differential-execution workflow rather than a diff review: two commits run in separate sandboxed subprocesses on generated inputs, then their return values, exceptions, warnings, output, mutation, and object state are compared. It also tests callers reached through the call graph, including functions whose source did not change. A third-process self-check can produce an abstention for nondeterministic behavior, and the reported evaluation covers 2,548 functions across eight repositories.