I'm writing this as an (incomplete) answer, as I exceed the MAX char length (for comments). But it is not intended as such - more of an attempt to further narrow down the hunt for clues (towards an answer). My understanding is that only a few people would be able to definitely answer your question, namely contributors to the actual EnergyPlus implementation. Most of them regularly visit UnmetHours, so I humbly defer to their expertise/experience.

"... according to the Engineering Ref and the expected thermal behavior, even with a reduced definition I find no reason for the decrease". I agree that the results (at first glance) are unintuitive - wrong even. However, I disagree that the Engineering Reference is clear on what happens when one maintains the default 0mm values for the following geometric variables:

• outside reveal
• inside reveal
• inside sill depth
• FD outside projection
• FD inside projection

My first comment (I should have explained myself better) was to suggest replicating the same window (i.e. same dimensions, double glazing, same frame, same dividers) using LBNL's WINDOW (which is generally how one obtains FD inputs in the first place). I'm curious as to whether WINDOW actually accepts these inputs (e.g. corrections, warnings), and what the outputs would look like. If successful, results could act as control case against which one could compare EnergyPlus results under winter design conditions. My understanding is that EnergyPlus attempts to harness as much of the WINDOW outputs as possible to correctly model the effects of FD, while being constrained to tweaking the properties (in this case) of a single glazing material. I doubt that 0mm for the aforementioned variables fell within the intended scope of these corrections - I could be wrong.

"You have changed into a higher exposed area and increased overall transmission with the 1.5 edge/center ratio, which degrades window enough to revert the results." Yes, of course. I didn't intend to revert the results - I simply reset the 0mm inputs to realistic values one would expect with FD. My only point was to illustrate that the EnergyPlus FD model works as expected with typical, real-world inputs.

"... compare the same exposed areas with a higher conductance of the frame, where the results are unexpected". I understand the intent. But I'm not sure that these "same exposed areas" are treated equally - there's a presumption (or an expectation) that they are somehow. There's an interesting discussion on the topic (about 7 years ago, here and here) on what happens when one defines a 0mm FD interior projection depth. It's not clear for me whether the merge at the time simply fixed what was being reported (versus what radiative/convective coefficients should be applied to FD items), when dealing with 0mm depth. It is clear from the exchanges that this needs revising or at least clarification, and I'm not sure if/when a more comprehensive fix has been (or is to be) implemented. I wouldn't be ... (more)