Finds bytes decoded with the wrong encoding, length limits counted in characters where the column counts bytes, emoji cut in half and names that match only after normalization. Returns the input that breaks each one and the fix with a test.