[jdcarter]

I recently looked into the research on this topic, and it absolutely agrees with you--failures are not independent. Papers of note:

https://www.usenix.org/legacy/event/fast08/tech/full_papers/...

http://www.cs.toronto.edu/~bianca/papers/fast07.pdf

Take note of section 5 in both papers for their statistical models and how they match with real-world data.

[hinkley]

There was a statistics class in college that kicked my ass. I never quite understood how to determine if two variables were independent or dependent. You get a vastly different answer if you get it wrong.

I run into people all the time that seem to have the same problem, to the point that it makes me wary of any software developer putting forth numbers that seem fantastical.

My gut reaction is that if Backblaze wants to keep reporting their disaster preparedness numbers that they need the assistance of an actuary to calculate them.