[tor-project] CloudFlare blocking got more severe in December 2015

[David Fifield]

I re-ran some of our analysis scripts from the "Do You See What I See?"
paper to bring the graphs up to the present day. In the paper, our
OONI-derived data only went up to August 2015, but now it's update to
the present. Take a look at this graph, which shows the rate of Tor
blocking over time (it's an updated version of Figure 7):

https://people.torproject.org/~dcf/graphs/ooni-tor-blocker-rate-20160323.png

The interesting characteristic is CloudFlare: after suddenly decreasing
from 3% to 1% in June 2015, and slowly decreasing over the course of
months to almost 0%, CloudFlare's blocking rate in the OONI data set
suddenly increased to about 4% in December 2015. This is consistent with
my own experience, that CloudFlare was pretty easy for a while, but
lately it's much harder.

These percentages are out of all HTTP requests issued by ooniprobe, and
only count cases where the Tor request was blocked but the non-Tor
request was not blocked. There could also be other confounding factors
that we haven't investigated, like maybe OONI's URL list or probe
distribution changed suddenly. If you would like to dig deeper, here is
the source code that generates the graph:

git clone https://www.bamsoftware.com/git/ooni-tor-blocks.git

The findblocks.csv file in the source contains a summary of each HTTP
request pair issued by ooniprobe, distilled from the raw reports.