Here's what he must do:
-- Run the experiment again, several times if possible. Combine or
analyze
the results of the several experimental runs in some statistically
valid way.
-- Run the simulation again, several times. Combine or analyze the results
of the several simulations in some statistically valid way.
-- Then, compare the analysis of several experimental runs with the analysis
of several simulations, in some statistically valid way, to decide whether they
really don't track.
-- If they do, then look for ways to tighten up the simulation so that maybe
they can track better.
-- If they don't, then he's back to before Square-1. He not only has to
deal with the question of whether or not his experiment is valid, but he
also has to consider the possibility that his simulation is trash and
doesn't actually model nature like he expected it to.
