We really need tests for the `samples` which get run regularly. Finding errors during release is too late.