Actually, I do not disagree that Rd1 is a good try as well, better than the dxc3 try. But that is not my point. It is one of mutually exclusive conditions : If Kofman overlooked the consequence of the retraction history, then Rd1 would not be a try at all! Placing a pawn on g4 could then have 2 reasons: (a) in retro's you always open up positions as much as possible to give them a less cramped feeling and more freedom in forward and backward play - that is also why h6 is not on h7 (b) to make a solver think he could be clever with dxc3 and castle after ...g3. It is not an impressive try but in retro-positions you are considerably restricted in showing forward content.
The second condition: if Kofman was aware of the relevance of retraction history and tuned his problem to get the thematic Rd1 try in - demonstrating he did put effort in his composition - it is even less likely that he overlooked / tolerated the dxc3 dual. That only happens in quick composition weekeends in Nunspeet with time running out under pressure to make an entry.
I therefore remain firmly convinced that Kofman was deluded when making his 1958 original (of lesser importance) as was everyone who quoted it after him (which is of great significance). Nunn not in that flock since he clearly made the right pick.
I have no (logical) problem with Kofmans e.p. version and replaying the e.p. move is a good theme. In my 2019 mind I probably fused the fact that there was an error in it with the fact that I used the e.p. retraction differently in my quadruplet - leading to a false memory. The message of my 2007 article was only that you cannot mix two types of retraction rules - with and without retraction memory - without making any note of it in the stipulation. A very modest statement considering that I really believe that retracting without memory should not be called "retracting" at all. But that is part of the larger picture.
I didn't think that Kofman made or would make a correction after my 2007 article and my database edits (I assume he was no longer alive), but I did think that people might stop quoting Kofmans clearly flawed problem and slowly start removing it from databases, articles and books. It is easy to discriminate them. No-one publishes Kofmans correction by accident because - besides being the only correct one - it is clearly inferior to the original. And publishers going for the original problem evidently have no understanding of the issues.
So here comes the big question. Why is the wrong version of Kofmans problem still in the Die Schwalbe database - and likely in many other places and probably being requoted today and tomorrow ... ? I know all the excuses for it but the bottom line is that you don't want a problem database to be a random mix of correct or incorrect problems. And if you leave them for some special reason then please explain with the diagram what it is. No explanation on either diagram, not even that one intends to correct the other. May be I could edit it as in 2007. Will it help? No, it will not. Why not? Because there are loads of problemists who still believe there is something right about Kofmans 1958 version. Why? Because they are confused and will continue to be until educated.
A point on PbyD. It is not a global logical attribute of a retro problem but a local property of certain items in a retro problem. In orthodox chess it only applies to castling and not to e.p., 3rep or 50m. That quickly changes for fairies. For instance, in reflex chess, no-castling and no-e.p. may have PbyD instances. Problemists like PbyD (once they know about it) because it hands them a quick and dirty instruction for handling retro uncertainties - without understanding context. Mutually exclusive castling is just a collection of mutually exclusive PbyD instances. PbyD and mutually exclusive castling are both instances of global RS logic applied to a local retro-property, in casu castling. All logical levels are governed by the requirement to have a proof game in every time line. Mutual exclusive castling has at least 2 time lines.
Kofman didn't add Pg4 to stop Rd1 but to turn 1.dxc3? into a decent try which he hoped to refute not by ..g3? 2. 0-0-0! but only by ...0-0!. That he corrected the whole problem later with a heavy heart was certainly not because he knew at the time there was an error but likely because someone told him there was!
No, 1.dxc3? as a supposed try defeated by 1...0-0! would work fine too if the g4-P were on g5 or g6. If Kofman were mistaken as you believe, he would've place the P on g5/g6 which would've added the even more thematic "try" 1.Rd1? defeated by 1...0-0! So you're implying that he (an IM of composition) also makes bad artistic choices too. The actual refutation of 1.Rd1? enabled by the g4-P - 1...g3! 2.dxc3 gxf2+! is a unique sequence and as an experienced composer, you should know that it doesn't happen by accident. But your theory requires that he "accidentally" stopped the 1.Rd1? dual. Put in another way, suppose he had actually placed the P on g5/g6 and later the correction with the g3-P appeared, would you then have tried to claim credit for "contributing" to the fix of the 1.Rd1? dual as well?
Decades ago as a new composer I made a retro showing mutually exclusive castling and only later did I read about the "prove by doing" concept. The former is a retro idea and the latter is an arbitrary convention and I can tell them apart without your assistance, thank you. I don't think I'm a genius for telling them apart, and have yet to hear compelling evidence that the vast majority of retro experts confuse the two ideas.
Here's a link to the two versions of Kofman's castling retractor. The correction was reprinted in diagrammes in 1994 (i.e. it was originally published even earlier), years before your 2007 article: Schwalbe PDB
And here's a link to his e.p. retractor: Schwalbe PDB. It's obvious that after retracting e.p., the only M3 is to capture e.p. and there's no alternative "try" that's supposed to be defeated by 1...0-0. So there's no evidence that the composer was trying to demonstrate "prove by doing". Retracting and replaying e.p. is an interesting theme in itself.