Testing how well an agent maintains performance when faced with errors, variations, or unexpected conditions.