A lot of people here also make the assumption that the human user would make no errors.

What error rate this same person would find if reviewing spreadsheets made by other people seems like an inherently critical benchmark before we can even discuss whether this is a problem or an achievement.