Hmm, I think you could maybe lean harder into the one screen thing?
Having to keep an eye on the dragon is good, but I think it would be great if there were more mechanics which required you to see what was going on in different parts of the screen, as that's your main mechanical twist.
I'm sort of torn on the torch, because it effectively undoes your twist! A small light radius is effectively the same thing as a small screen, right?
I think if some of the other mechanics were based around characters and unexpected interactions, you could get a chuckle out of seeing a wizard who's been working in his workshop the whole time becoming a roaming enemy when he gets annoyed at you flooding his workshop during solving another puzzle, that kind of thing.
Have you heard of a game design technique called the false ceiling? It's when a game looks like it's going to end at a certain point, but then surprises you by continuing past that point.
It's usually a nice little addition to a game, but in your case I think it may be mandatory to get the most out of the concept.
Having the castle transformed, or the context changed, by player actions which create new challenges would be excellent.
Effectively, I think you want to basically design dozens of levels, but then contrive a way that they can all be the castle, modified by some event (usually caused by the player).
Does that make any sense?