There was a recently published academic paper "Bitcoin’s limited adoption problem" which used a statistic that the network latency is around 15 seconds.
Obviously, this is not true: One can provide a quick argument with Poisson distributions and Jensen's inequality that this means we should be seeing over 1000 stale blocks per year with probability near 1.
I'd like to refute this claim empirically, but have no good source for claiming that the number of stale blocks is small. I can only see the blocks that hit my node and go stale, but I'm assuming that not every stale blocks makes it to my node. Any better sources of data?