Usefulness of rating / Documentation needs more explanation

[dagelf]

An immediate question that arises is: Is the ERR column, the total number of attributes that met the fail criteria, or the total number of drives that had 1 or more errors?

Additionally, number of samples per drive seems relevant - can drive serial numbers be matched over time? (Or we might end up with a situation where there is no record of failed drives - ie. we only get stats from healthy drives, and when they fail, we don't know about it - or even worse, an incentive for manufacturers to not report accurate stats. Abraham Wald's infamous story from WWII about where the planes needed to be reinforced, springs to mind: The sections with the most bullet holes? Or those with the least? The latter of course, because planes with bullet holes there never made it back for the bullet holes to get counted.

I'd say that the summaries should be sorted by number of samples, rather than alphabetically, but its worth thinking how number of samples per drive could be tracked - I can think of a few ways. Eg. in each set, count the number of drives that went missing, and log them as missing. Or keep a running tally of measurements per year for each drive.