Yes, you are right. I've had issues with merge faces for some reason (certain polys were omitted), that's why I went through the process myself, the only thing I considered a plus there is the threshold for coplanarity, which resolved this problem (I assumed it was caused by inner vertices being pushed-pulled randomly, via eye balling, resulting not perfectly coplanar surfaces). If I recall properly, the merge faces component has only one input, and spits out the cleaned brep with some numbers (before/after face count).
The posted script was a part of a larger one with a different purpose, the remainder of the algorithm required a more managable brep input than the one from the initial GS node. Maybe I misunderstood the initial question, and it won't be a solution, but it could be helpful.
odv.hu | actively using: AC25-27 INT | Rhino6-8 | macOS @ apple silicon / win10 x64