Along the lines of #16 we should follow webPPL's example closely and make the inference test interface more generic, automate test creation, and write test models independently of the tests themselves so that they can be reused. We could even reuse the test data/ground truths from webPPL and other projects where possible.