I think I just answered my own question
I just re-implemented the OSTC4's ceiling algorithm. While the old algorithm does work, and gives results that I did understand, it seems not efficient ... and my new algorithm gives very similar results to hwOS. That was not my intention, so it was a bit of a surprise, when I ran it.
The ceiling represents an ascend relative to GFhigh, and the "normal" deco stops do not do that. In other words, when following the ceiling, you follow GFhigh through all deco stops, so you violate the deeper stops, and you end up at the surface at GFhigh (like you would when following the deco stops).
Not sure I really would want to dive a full deco schedule on GFhigh