From a quick look, this doesn't have the model evaluate code quality, but it runs a heuristic analysis script over the code to determine the groan signal. Did I miss something? Why not leave it to the model to decide the quality of the code?