Because we generate the data on the fly, the accuracy bounds might be violated now and then. While it's valuable to keep an eye on this, failing unit tests create too much friction. We should freeze the seed and test for accuracy in some other way.