My script using adviseLong/Short is producing impossible equity curves with R2 coefficients of 0.96 and literally no drawdowns, looking super overfit. I just assumed the advise functions would use DataSplit the same as optimize() does but now I looked in the manual and I haven't actually seen that written anywhere. In testmode it correctly only uses the date range left over by DataSplit but the training must also be using this data. Is this correct?

As far as I remember and understood it, the machine learning algos use all available data to train on by default. If we got time windows A, B and C, a model is trained on A, a different one on B and another one on C. After that, the strategy is trained on B with the model from A, then the same with C and B. So the models are trained with in-sample data, but the strategy itself is always trained with out of sample data being presented to the trained models. If you are using NEURAL, then you can control the data split etc. within your R script.