We presented test of what we believe are ``building blocks'' of microsimulation models. Further ``building blocks'', not included here, are probably freeway ramps with merge lanes, and freeway weaving sections. We plan to include these tests into future versions.
As pointed out earlier, we believe that ``clean'' real world measurements of the ``building block'' situations are hard to obtain. Thus, one may consider them primarily useful for comparing simulations with each other and with theory; nevertheless, we think that one can judge from the results at least if the simulation is ``in the right ballpark''. It would certainly be desirable in the future to also have test suites for more complex situations. - For the same reason, we did not make any attempt to get ``better'' results than the ones presented here: we know that the results change in more complex scenarios, and it is therefore unclear if a change ``to the better'' in the test cases may not be a change ``to the worse'' with respect to reality.
Also, we would shortly like to point out again that ``verification'' of simulation models, i.e. the question if an actual code corresponds to a (possibly incomplete) specification in a paper, is in practice a difficult question. An alternative approach would be to try to find a suite that decides if we are macroscopically convincing without the need to go through testing the rules on an individual scale. Arguing about the microscopic rules could then be left to a small group of specialists; the end user could just look at the test suite results and judge in a matter of minutes if the simulation has faults that would seriously affect the analysis of their problem.
Last, all these problems imply to us that one should expect that simulation models will undergo continuous improvements, and it seems more realistic to us to expect ``test suites'' to be run at regular intervals instead of expecting that parts of simulation models can be validated and calibrated ``once and for all'' at certain stages and then never be touched again. In consequence, we would like to shift the argument from a discussion whether a model is ``correct or not'' to the discussion about which tests should be run to enable the user to make that decision, and how these tests can be made comparable between different simulation models.