This post describes how I view the highest-level structure of Dungeon World. Most of what I’ll say also applies to Apocalypse World, so I’ll mix discussion of the two with little warning.
The first level — everything is the conversation
Dungeon World is a game of rules that modify a conversation. That conversation refers to a simulated world, which is modified by the conversation and in turn shapes the space of what is reasonable for the conversation to say at any point. The world does not have any more reality than that — it doesn’t “do” anything itself, unless the (rule-governed) conversation causes an update.
For example, if the Red Knights haven’t been mentioned for six weeks of game time, they exist in a space of possibilities – they could be here, they could be there, they could be anywhere within six weeks travel of wherever they were last. The conversation can bring them into the current scene (or show their very obvious effects e.g. having burnt down a town and left their flags all over it) as long as that is consistent with the time, the distance, and whether of all of the chaos shrines in all the mountains of the world there is some plausible reason for them to show up in this one.
What the Red Knights don’t do is move around in the background, in some factual sense, even in the GM’s head. The GM may have ideas about this, but they’re merely ideas until the live conversation makes them fact. The GM’s prep is raw material, prompts, aids, but not reality.
I.e. the possibility and plausibility space is “real” beyond the conversation (and each player plus GM will be independently monitoring it to some degree) but the precise facts there are not.
But is that right?
I went on Google+ and asked —
1. Is the above consistent with how you play?
2. Is the above consistent with the current RAW?
3. Do you think the above what is Latorra and Koebel intended?
The consensus was “yes, mostly, but you’re ignoring the rules about things ‘offscreen'”.
The second level — “Think offscreen too”
So DW, RAW, isn’t as simple as I’ve made out above. There is a principle thus:
Think offscreen too
Just because you’re a fan of the characters doesn’t mean everything happens right in front of them. Sometimes your best move is in the next room, or another part of the dungeon, or even back in town. Make your move elsewhere and show its effects when they come into the spotlight. (p162)
Similarly, Apocalypse World 2e has this principle:
Think off screen too. When it’s time for you to make a move, imagine what your many various NPCs must have been doing meanwhile. Have any of them done something off screen that now becomes evident? Are any of them doing things off screen that, while invisible to the players’ characters, deserve your quiet notice? This is part of making Apocalypse World seem real—and if you pay attention to your threats, it’s part of making the characters’ lives not boring too. (p86)
(thanks to Morgan Davie for pointing me at the AW version of this)
Neither game has a move “do something offscreen” — instead, the principles above tell you to sometimes make your move offscreen, and thus only in your head (or notes), not in the conversation. If that actions is to be meaningful, and is to integrate properly with DW’s action economy (fail a roll –> get XP, but suffer for it), those offscreen decisions need to be real, permanent, factual.
The third level — Fronts and prep
DW doesn’t say that the GM can’t prep. And there’s nothing in the rules that prevents them pretty much hewing to that prep, come what may. If their notes say the Pitiful Stair is haunted, then they can make the Pitiful Stair be haunted when it is encountered. They have that latitude under the rules. 
(NB I’ve previously thought that the principle “Draw maps, leave blanks” as an instruction to indeed prep some, but re-reading it now it seems like an instruction to draw shared maps at the table.)
DW provides “Fronts” as a tool for prep. But it’s quite clear in the text that they are meant to be advisory
Fronts are built outside of active play … [but] … You may tweak or adjust your fronts during play (who knows when inspiration will strike?) [although] the meat of them comes from preparation between sessions.
Fronts are designed to help you organize your thoughts on what opposes the players … Consider them an organizational tool, as inspiration for present and future mayhem. (p183, emphasis mine)
In my Google+ thread, Jeremy Strandberg and John at Deep Six Delver both talk about doing substantial concrete prep. The latter talks about his Fronts being in between the evanescence of the possiblity space and the concrete reality of the established facts — he tries to stick to them so as to help him maintain a coherent, believeable world. For my current DW game I have quite a few pages of maps and notes, although that’s partly because we started it with Zweihander.
But is this distinctive? Is it different to any other game?
Over on Google+, Jeremy Strandberg said “I don’t think there’s much to your description that’s terribly specific to DW or AW style play.”
To an extent, sure. Dungeon World play, done RAW, is not exceptional. Lots of games are played this way, especially if we only look at the player’s view. A recent blog post by Belock Shrike tries to describe the (implicit) core mechanic of D&D, and ends up with something very like DW’s (explicit) core loop.
But here, as it does with so many other places, DW makes explicit what other games leave explicit. It explicitly codifies one style of GMing, and rules out others. As +Morgan Davie points out (above, and in http://taleturn.com/these-9-games-have-hidden-goblins/), there’s more than one way to handle reality-of-offscreen issues, and different games give different advice (more rarely, they give different rules). Most games commit to much less on this issue — they leave it to the GM to work out what to do.
On the specific issue we’re describing, even Apocalypse World, as written, is slightly different to DW — it says that some prep should be held as definite and certain. Morgan Davie summarises it better than I could, but here is the key quotes from AW:
During play, you leapt forward with named and motivated NPCs, you barfed forth landscapes and details of society. Now, between sessions, it’s time to go back through your notes and create those people, places, and conditions as threats.
Creating them as threats means making decisions about their backstory and motivations. Real decisions, binding ones, that call for creativity, attention and care. You do it outside of play, between sessions, so that you have the time and space to think. (p106, emphasis mine)
As Morgan puts it:
Note […] a crucial run of four words: “Real decisions, binding ones”. The GM isn’t supposed to leave these things vague or undefined so they can pull it out of the air in later play — they make a call about the hidden fiction now, and they stick to it.
In other words, the Bakers tell MCs to to do some kinds of prep and to then keep that as binding and real.
Most importantly to me, what I say above is not always how I’m trying to behave as a GM. When I run DW as above, I’m thinking “only the established fiction is real”; when running Immergleich, I deliberately thought “the world in my notes is real, even when unseen”. It was quite different GMing experience.
Committing to concrete prep comes with benefits (e.g. I can design the world to be “balanced” in some way) and costs (e.g. it can stop me narrating exactly what the game needs in this moment). Even in this mode, I never keep to my prep perfectly. I can’t, and I don’t try. But when am in “my prep is real” mode, I experience myself as starting from that ideal and then degrading as gracefully as I can. That feels very different to not trying at all.
I take the different feel to be a cue that different things are happening, and that if you studied my GMing closely you’d be able to distinguish the two modes. In any case, I’m a player too. How the GMing experiences feels to me matters.
It’s worth noting that I naturally think of strong-GM rpgs as being a simulation that runs in the GM’s head. The *W “the game is a conversation” model is a second language for me. I suspect it’s more useful, at least for the kind of games I like, but I have to make effort to use it.
 Maybe there are rare cases where making the Pitiful Stair be haunted would be pushing the bounds of the GM’s agenda and principles. It might just be that it’s a dick move to make it be haunted right now whent the PCs discover it, that it doesn’t serve the game in any way.
Maybe, more straightforwardly, there might be cases where it being haunted conflicts with established fiction, e.g. that all ghosts have been driven from the earthly world. But that seems, to me, like a straightforward case about consistency, not really one of “does the GM think offscreen” at all.