How were the results for removing stopwords versus not removing them?

*sigh* Well, it's actually doing a little worse when I remove stopwords and do the other thing I'm doing, and I don't know why. But I rigged up some fancy automated testing and a couple of the other anomalies went away (like the accuracy on the test data being way better than on the training data), so I think that that, at least, must have been Human Error™. I'll send you the big fancy report!