In statistical terminology, I would prefer an "unbiased estimator with low skewness", such that it's as likely to overestimate as underestimate, even if occasio

In statistical terminology, I would prefer an "unbiased estimator with low skewness", such that it's as likely to overestimate as underestimate, even if occasionally the estimates are quite far off. But as discussed, manufacturers of course have a strong incentive to make these predictions very biased to overestimate the range.