The testing is a bit lax at the moment. There are many ways that this could be improved without too much effort. Once test coverage gets a little bit higher we might even want to put a little "coverage" badge on the front page!
For example, there is virtually no unit testing around the DSL, with only a handful of functional tests to ensure it works properly. The learn tests have also been disabled for some time, since they never were updated after the algorithm was really finalized.