incorrect binomial distribution?

This is an excerpt from doc/org/oaut.org:

#+BEGIN_SRC sh :results verbatim :exports both
randaut -d0.2 -Q10 -n1000 a --stats %e > size.csv
R --slave -e "summary(read.csv('size.csv', header=FALSE, col.names='edges'))"
#+END_SRC

#+RESULTS:
:      edges
:  Min.   :14.00
:  1st Qu.:22.00
:  Median :25.00
:  Mean   :24.72
:  3rd Qu.:27.00
:  Max.   :36.00


For $Q=10$ states and density $D=0.2$ the expected degree of each
state is $1+(Q-1)D = 1+9\times 0.2 = 2.8$, so the expected number of
edges should be 10 times that.

The maths say that the mean should be 28, but regardless of the number of automata generated or seed tested, the computed mean tends to 25 instead.

So there is something wrong here. This computation was working fine in cbf1e15b. I wonder if the binomial distribution might have been broken by 734bceff (which is just adding back some old code from before we used C++11, because C++11's random distributions are implementation defined.).