Errors cancel out when they're errors of magnitude, which is what I believe he was referring to, and is often the most important property of Fermi estimates. I

Errors cancel out when they're errors of magnitude, which is what I believe he was referring to, and is often the most important property of Fermi estimates. If one estimate is off by 2x and another is off by ½x, those cancel out because the exponents are additive. Edit: Parent deleted comment…oh well.