I've always done my gradient reduction on the linear data being stretching and finishing off.
Last night I did a masked stretch on an integrated image & did the gradient reduction on the non linear data and it came out surprising well. It seemed to look visually better than applying on the linear data. I've begun to wonder now where the best stage in your imaging workflow is best for gradient reduction. The ABE and DBE seemed to produce a very nice result on the stretched data. It was as if the stretching smoothed out all the kinks so your not working with lumpy gradients all over the image allowing DBE or ABE to work somewhat better.
Please let me know your thoughts on this important topic.