Level(3) faux paux

How many places does AS3356 connect with other AS's, and how many places does
AS1239 connec with other stuff? I'd expects an AS with 500 interchange points
would have 25% more whoopsies than one with 400 interchanges even if they were
otherwise equivalent. Another factor would be number of miles of fiber their
net is run over - If backhoes per mile is a constant, a 3,000 mile link is more
likely to be hit than one half as long, and so on. Or maybe Level3's network is
more openly visible from outside, so it's easier to tell that they are the
source of a problem than a net that's not easily debugged from outside the AS
(leaving you wondering if it's them or somebody on the other side of them). Or
maybe past experience has shown that the two have the same *actual* failure
rate, but asking for a Level3 help is more likely to actually get you a clueful
*and* helpful engineer.

Plus, I don't think *any* provider gets mentioned enough on NANOG to be able
to draw any realistic statistical inferences. The short-form highly inaccurate
handwave is that you'd need *two* providers and at least 20-30 datapoints
on *each* to draw conclusions - which would probably take you 2-3 years to
collect, and at that point, changing management policies at both providers will
render your results inaccurate (you'd be comparing 3-year old data to current).