True, and that was noted when this game was posted a couple of years ago. Article 17A of the Codex says "Unless expressly stipulated, the rule of dead position does not apply to the solution of chess compositions except for retro-problems." This being a retro-problem, the dead position rule does apply.
Yes, I haven't been paying attention lately. This was changed in the 2015 Codex, it wasn't there before. The problem is that no-one cares to define what a retro-problem is - though some types are recognizable as retro by their stipulations. The weird part is that problems running into this issue can always be considered retro-problems for the simple reason that they ran into this issue. To find out what the author intended you first need to solve it and then reverse engineer that this was probably not a retro-problem! A questionable approach to rulings. Note that standard non-retro-problems need some retro-conventions as well, e.g. to find out about castling and e.p. right and who's on move.
Btw, all these issues could be resolved by introducing the habit to announce retro / non-retro type with the stipulation as soon as there could be the slightest doubt. And the author could use it to place compositions in a different category than would be assumed by default! Then a proof game such as this one could be treated as either a retro- or a non-retro problem.
Btw2, proof games are the least retro-type of all composition types anywhere! The retro-conventions are typically required to provide "choice rules" to fill in unknown states. But proof games are purely retro-analytical and have no unknown states whatsoever! Unlike standard two-movers or endgame studies which may demand access to retro-conventions! So proof games ought to be classified as non-retro with regard to the retro conventions. And then article 17A will save proof games such as the one discussed.
So we arrive at the retro-analytical "release the position" (which is a proof game) type by a detour. Typically the retro-conventions do NOT apply to the retro-analytical type. Nothing is known about states of castling, e.p. or who is on move. You have total freedom to retract whatever you like as long as it is "legal"! This is the place for proof games as well: retro-problems solved without retro-conventions and with pure Sherlock Holmes analytical deductions! Btw, Smullyan in his retro-books never assumed any "choice rules" for the unknowns. No conventions!
what about
This is an illegal proof game by FIDE law as the position is "dead" (no one can checkmate the other side) after 17. .. Kxe7 which terminates the game automatically in a draw. To get the final diagram legally you'll need to make the rook capture as the final move on e8!
This issue has recently been discovered by the retro-community though I've known it for 15 years. It caused problems in standard compositions which were drawn before the point designated by the author. I think they changed the rule now or are in the process of doing so. One of these days I'll look where they have gone with it. Probably the wrong way as they have a habit of doing.
Note that "dead positions" are a basic FIDE law (article 5.2.2) and not some weird "competition rule" or "Codex convention".
ohh okay! I did not even realize that