Does that match up with the 99.999999999% vs 99.99% durability figures? I don't know what the proper math is here.

edit: they do explicitly say "Designed to sustain the concurrent loss of data in two facilities"

Inexact at best, but ..

  P(3 locations failing) = 1-0.99999999999

  P(1 location failing) = P(3 locations failing)^(1/3) = 0.000215443475

  P(2 locations failing) = P(1 location failing)^2 = 0.999569159

I think you're right ... I don't see how these figures are consistent. Maybe they're factoring in the time it takes to restore a copy, maybe 'normal redundancy' S3 is doing something clever, maybe it's simply another instance of Amazon being economical with the truth.

There is no proper math. 99.999999999% is a very silly number. The probability of all of Amazon's datacenters being destroyed in nuclear war within the next year well exceeds 0.000000001%.