Item 7 of this dungeon-building project specifies a discussion of dungeon-building philosophy. My first post of actual content was running a bit long just as my spare time was running a bit short, delaying said philosophy until now. MMO design and LARP design have both informed my ideas on dungeons, as you'll see.
The truth is that I have design philosophies about games, not dungeon design specifically. For one thing, I've never been a professional level designer. (This has a lot to do with not knowing a damn thing about using Maya, 3DMax, or any other comparable tools.) As a Content guy, I made some recommendations about what might go where and what the place's overall story was, but in general I stayed out of their way and we were all happier as a result. Having a level design philosophy wouldn't have done me a whole lot of good anyway, since the other main thing we did was stock existing levels in different ways. Just about every MMO, as well as some single-player games, does this. (Quick list of guilty parties: WoW, CoH, Mass Effect, FE.) This is not, in itself, really a sin. It's really a question of how well they discourage you from noticing this fact. The amount of time and energy that goes into texturing a level and making pathing work properly means that it would be grossly inefficient not to re-use levels. The games I listed were also focused on presenting a large world, so they needed to present a huge amount of content. Oblivion handled this with making its repeatedly-used unit something smaller than the whole level, and did a pretty good job of disguising what they were doing. Other games have used the patchwork-level idea with varying degrees of success; I seem to recall that FE's pathing issues were exacerbated by this practice. WoW's cave segments are particularly recognizable. Don't get me wrong, though, this is still comfortably filed under "good practices" as far as I'm concerned.
Dungeons present a completely different set of challenges in a LARP. The one universally-recognized virtue of Indian Springs State Park in Georgia for LARPing purposes is that its cabins, barracks buildings, and the Infirmary are ideal for conversion into trap-filled and monster-haunted dungeons. Other parks have lodges and pavilions that are workable, but IS has an embarrassment of riches on good places to run indoor modules. Given those limitations, games sometimes use outdoor areas and request suspension of disbelief ("okay, guys, the edge of this path is the cave wall"). Everyone knows it's not the most ideal case, but it's how things are done.
Special mention goes to Eclipse's Wonder Twins, who have in the past coordinated taping off an entire unit of the site with police tape. Just having more definite boundaries helped; they also used rope lights to define some parts of the terrain as impassable. The whole unit represented a vast underground cathedral, and was an excellent module that included the entire playerbase as either PC or NPC. Good stuff there, but ruinous on setup time, and not infinitely repeatable.
The longest dungeon crawl I've ever heard of in LARPing was eight hours: the Tomb of William the Black in SI. The longest dungeon crawl I've personally played in an MMO takes around six hours to clear: Blackrock Depths. These are minimal compared to, say, the World's Heaviest Dungeon. A dungeon crawl intended to last pretty much the whole campaign is commonplace in old-school play, and Undermountain, and so forth. (Also, the Tomb of Horrors was certainly intended to include the beginning, middle, and end of the campaign - possible in a single night of play.) Diablo gets a special mention here - prior to the expansion, the first game keeps the entirety of the action in the same dungeon.
There are countless blogs full of advice on how to build dungeons that will remain interesting for as many levels of play as the game runs. I've never played in a game like this. The longest dungeon I've played through in a tabletop setting was an adaptation of I3: Pyramid set in Eberron. We were in this dungeon for around seven sessions, I think, which was something like a quarter of the game's full run. The longest dungeon I've ever run was in a Forgotten Realms game - I would guess that the PCs spent something like eight to ten total sessions, out of the thirty-ish we played that year and the sixty-one we played overall, in this dungeon. More recently, I've had four-session and five-session dungeon crawls in my 4e game. I mention all of this to establish the baseline of my experience.
Because LARPing can't have serious dungeon crawls in the D&D style, and because relatively few computer games do so, extended dungeon sequences are not a major part of my DMing style. A few sessions in a location, probably one encounter on the way and 2-4 at the location, and it's back home for a hearty meal. This might take two or three sessions; the five-session dungeon mentioned above was a very big deal, the finale of a conflict that had taken more than a year of real time. My games are typically about conflicts with bad guys who use powerful magic, not plumbing the dungeon's depths for riches.
On the other hand, there was my first Mage: the Awakening chronicle. Probably 75% of the things that happened in that game took place inside the PCs' home base, which was an old and sprawling mansion with lots of locked doors and lots of secrets. Exactly one fight took place inside the house; otherwise, it was creative problem-solving.
So I don't have a philosophy on how to draw, describe, or stock a dungeon. I have ideas for what makes a good game in general.
1. Lots of moving parts that the players can see, learn about, and manipulate are good. I don't have a metric for this, but I believe that there is Not Enough and Too Much and a pretty decent margin between them. Compartmentalize these to make them manageable. My current example of this are the three new substances I've included in the writeup of Sector 1. These moving parts are mostly ways to respond to challenges, many of which may be puzzles that act as barriers.
2. Uninformed choice is not interesting. (See: why I do not like Tomb of Horrors. I am well aware that it is an abattoir, by design - to which I say, "No, thanks.")
3. D&D is not built for days with just one encounter. This seems to be a lesson I have to relearn in every edition. I need to be careful to write things so that days with one encounter also have at least two more encounters.
4. Two combats in a row is okay, three is pushing it, and four is an outright error. There are a lot of right ways to break that up, though. Friendly interaction, puzzle-solving, or skill challenges are all great. The more these non-combat encounters get PCs to talk to each other, the better.
5. I don't have a lot of use for the randomness touted by the old-school folks. It doesn't save me planning time, it doesn't save me running time, and it means there will be more fights in parts of the dungeon that aren't good for a fun combat. Combat is improved with a bit of space to maneuver, though not complete openness. I don't need to be surprised by my dungeon; the players will throw me for a loop just fine on their own.
6. Including secrets for the players to find is an inherent good, as long as searching for more secrets doesn't slow down gameplay to a crawl. Therefore you want to give cues as to when would be a good time to search.
7. The second or third time players see a place, creature, or phenomenon that matters, their recognition and ability to predict its actions are an emotional reward that they should be allowed to enjoy in full. It's okay to vary things up and have different rules apply, but do not ever write things that punish players for learning your world without warning them that "this time is different."
8. My players do not regard mapping as a thing they need to do, and making them draw a map from a verbal description is not my idea of a good time either. James C., I am so with you on this. Another reason that my dungeons are not very large or are not mapped out in their entirety is that drawing out the dungeon on graph paper, battle mat, or TacTiles can really start to chew up gaming time, and I don't get to run sessions longer than 3-3.5 hours often enough.
9. In terms of improving gameplay, 4e's focus on interesting terrain or scene features players can use may be the best improvement over previous editions. It's the kind of thing DMs could do in previous editions, if they thought to do so. I certainly didn't, and I can't think of any DMs who did. My only complaint with the fantastic terrain is that it's too much wacky-fantasy for me - I would like it more if it were skinned to fit into a grittier setting. An interactive environment is an inherent good, and the terrain or features would have to be pretty lame to make it another other than a pure positive. I am not great at designing these in a hurry. Battle areas with terrain-and-features take me a lot of time and brainstorming, and even then they aren't as good as what I've seen other 4e GMs (or an SotC GM) use.
10. This is a new piece of philosophy that has come up in DtD brainstorming. Three-way conflicts are cool. Procedural stories (dungeons, for example) with a twist are cool. DtD staff calls this Dogsoldiering a module, in reference to the excellent movie. Things are going along just fine, it's a pretty standard situation, and suddenly there is a werewolf munching on your buddy's intestines. This is also known as "When Plotlines Collide," and there was a famous reversal of the Dog Soldiers scenario in Shattered Isles: A group of PCs is fleeing from a group of werewolves. The PCs flee past the ghosts of William the Black &c. The werewolves attempt to pursue the PCs past same. The werewolves learn, to their detriment, that William &c are well-equipped with weapons-grade silver. (Interestingly, this encounter does not trigger my hatred of "NPCs stealing the spotlight," in large part because it was unplanned.) As a result of this philosophical point, future sectors of the dungeon will probably involve sudden violence that is in some way a twist. Not sure what yet.