lucy24 - 5:39 pm on Dec 2, 2011 (gmt 0)

If your whole package works in UTF-8, you would never see anything reverting to numerical entities, so it's all down to your eyeballs.

But after reading the original post, I went back and checked a related issue that has worried me periodically. Somewhere along the line g### has learned to recognize "soft" hyphens-- at least to the extent that it will offer up "recognize" as an alternative if you search for "recog­nize", and vice versa. Last time I tried (probably a few years ago) they had not yet learned this.