It is true that the errors are **unlikely** to have the same sign. And they are unlikely to all be a whole epsilon. Hence, to assess **probable** error, some analysis is necessary, together with some assumptions about probability distribution, and likely deviation from the means. But Mr. Damon qualified the "1e-2 error" as being "possible". None of that fancy statistical analysis is going to contradict that assertion; it can only show likelihood, not the extremes of possibility. (In fact, such analysis can obscure those extremes.)