@KenGordon's post prompted me, so I just went and had a readthrough of Erik Baker's Fortran code for calculating an ascent.
To the best of my reading (it's been a long time since I was coding Fortran for a living), his code uses GF Lo to calculate the first stop on the ascent, period. It does not check for a direct ascent against the GF Hi first.
I also re-read Baker's article on Understanding M-values and I don't find anything there that really specifies whether GF Lo is to be used always or only after GF Hi dictates at least one mandatory stop.
My thought is that this little nit, in the small window of dive time where GF Hi does not dictate a stop but GF Lo does (or would), is one that is really up to the discretion of the implementer. So, I don't think either way is "wrong" and they only give different results for a pretty small subset of all dives - with the fuzziness of deco theory rendering that difference meaningless in any practical sense.
I have mixed feelings about how I would want it to work. But, as I am now sold (based on other threads) on using a fairly high value for GF Lo (usually, 50), I think the difference is REALLY irrelevant to me.