A while ago I thought I came upon a situation, that I think MTGA resolved wrong, but wasn't sure if I remembered correctly. Yesterday that scenario happened again for sure.
Situation: I have a Nightmare Shepherd on the field and a 1/1 Gemrazer token it created when the original Gemrazer died. Now I mutated another Gemrazer onto that token (on top). As per rule 721.2d, that permanent is now not a token anymore. A turn later my opponent killed that Gemrazer, and I would have expected the Shepherd to trigger again to make another token. Yet MTGA didn't give me that trigger.
Am I missing something here, or is this a bug in MTGA?
For reference:
721.2d If a merged permanent contains a token, the resulting permanent is a token only if the topmost componentis a token.
So, there are two pieces here that are relevant. The first is the one you pointed out: it is not a token even though a token is a component in the pile. So, Shepherd should have triggered and potentially allowed you to exile the cards in the graveyard.
However, the more important piece is the "if you do" clause on Shepherd. This requires that every component of the pile be exiled in order for you to get a new token. While the trigger should have happened, it is impossible for you to exile every component since the token component ceased to exist.
It certainly sounds like a bug if the trigger didn't occur, but you can't get the new token anyway so the end result ends up being correct. Since you can't just perform part of the process, you also can't exile the card components just because you want to. It is all or nothing and since you can't choose "all" you end up exiling "nothing".
You have understood correctly that the ability should have triggered: under C.R. 723.2d, a "nontoken creature" includes a merged creature even if one of its components is a token, as long as its topmost component is not a token. If such a creature is not treated as a "nontoken creature" for the purposes of Nightmare Shepherd, then that is a bug in MTG Arena.
By the way, where Nightmare Shepherd says "it" (as in "you may exile it"), what is meant is the cards and tokens that make up the creature referred to (but note that any token components of that creature will have ceased to exist in the meantime [C.R. 704.5e]) (C.R. 400.7d, 400.2, 723.3c).
EDIT (Sep. 16; Oct. 3): Some rules were renumbered in the meantime.
Situation: I have a Nightmare Shepherd on the field and a 1/1 Gemrazer token it created when the original Gemrazer died. Now I mutated another Gemrazer onto that token (on top). As per rule 721.2d, that permanent is now not a token anymore. A turn later my opponent killed that Gemrazer, and I would have expected the Shepherd to trigger again to make another token. Yet MTGA didn't give me that trigger.
Am I missing something here, or is this a bug in MTGA?
For reference:
Former Rules Advisor
"Everything's better with pirates." - Lodge
(The Gamers: Dorkness Rising)
"Any sufficiently analyzed magic is indistinguishable from science."
(Girl Genius - Fairy Tale Theater Break - Cinderella, end of volume 8)
However, the more important piece is the "if you do" clause on Shepherd. This requires that every component of the pile be exiled in order for you to get a new token. While the trigger should have happened, it is impossible for you to exile every component since the token component ceased to exist.
It certainly sounds like a bug if the trigger didn't occur, but you can't get the new token anyway so the end result ends up being correct. Since you can't just perform part of the process, you also can't exile the card components just because you want to. It is all or nothing and since you can't choose "all" you end up exiling "nothing".
By the way, where Nightmare Shepherd says "it" (as in "you may exile it"), what is meant is the cards and tokens that make up the creature referred to (but note that any token components of that creature will have ceased to exist in the meantime [C.R. 704.5e]) (C.R. 400.7d, 400.2, 723.3c).
EDIT (Sep. 16; Oct. 3): Some rules were renumbered in the meantime.