A left-field approach, inspired by the hard-funnest puzzle I've ever tackled:
run them through a fractal maze1.
A fractal maze contains entire copies of itself as subsets of itself, as one might suspect. I'm not suggesting you actually design one in-game and by DM fiat manage the inevitable conflicts in spatial scale; you can use one to be your one-sheet generator for your characters' experience.
In solving an actual fractal maze one tracks the stack of sub-copies entered/exited in searching for a solution. In this case we're going to use the stack to trigger encounter tables and set a more-appropriate end-condition--the generating maze will likely be non-solvable, anyway2. Arrange the encounters into tables linked to each sub-maze, and either roll or cycle through entries3. Some entries could be modified when you hit them subsequent times, but some should definitely stay the same. One does re-trace one's footsteps while wandering around, after all.
A Prefatory Note
This answer is long. 3,000 words-long, give or take. I know it's long, and wonder if it's inappropriately-long. But I think it's
- a simple idea to implement,
- that answers the question in a very novel way,
- which requires a worked-example to understand.
I invite suggestions as to how to better structure this, or your help through edits. Possibly 3/4 of it should be the content of a well-designed webpage somewhere to which a link is provided? But I don't have that. Perhaps the ~minimal~ working example could be smaller; this is (a simplified version of!) the recent one I ran in a home campaign. Perhaps there's a way to make the tables more... tabular?
I first came across the fractal maze that's linked above in a puzzle-game; I don't know whether the game or the MAA article came first. But I believe Mark J.P. Wolf deserves credit for its creation. Much of the thinking about how elements from the tables connect is influenced strongly by Jaquaying the Dungeon, by Justin Alexander. And the Seekers and Watchers are a nod to Highlander: The Series.
Tailoring Fractal Mazes to the RPG
There's obviously no way that the maze can be solved by the characters or players. So getting to the finish can't be the end condition. Perhaps it should be number of encounters, perhaps returning to zero depth, or hitting a certain depth, or generating a certain subsequence in the stack. It's going to vary widely depending on your setting and purpose for doing this to your players; I hope the examples will give you a sense. For some this is likely a deal-breaker, so it gets first-billing.
There's a good argument to be made that by making it unsolvable I've removed player agency. I think that's less of a problem here than in some of the other proposals: being run through a deck of cards or randomizing tables is no more solvable than is this. But in this approach sometimes the players will get "from ahead is the unmistakable--after visiting the location three times!--bouquet of the sheep pens, to the left you hear the cavernous echoes of the atrium, and to the right are some branching ramps." Players will have cues to return to or avoid areas they've already visited, an experience that doesn't prominently figure in many other solutions.
At what level of detail should this be run? I'd suggest that it be presented to the players as many-minute chunks: "We'll assume you've spent half an hour looking around, checking for hidden features, traps, testing doors, opening chests, scoping out empty rooms, going a-ways down halls, &c. I'll describe what you've found in that time and you'll make a choice of how to proceed."
Keeping track of time, resources. Is lighting a factor? Food? Are they likely to spend days in here? In my example we'll say that traveling an edge generally takes an hour, but is not so restful as to be a short rest. (You're forcing a door here or there, peering at old marks on walls, &c.) The tables have provision to make that doable, I believe. So the party doesn't choose to take a rest at any time: there are opportunities presented, which may be seized.
Ability checks. To insert some more player influence you may want to routinely ask for stealth, investigation, perception, &c. checks. Table entries should then include optional info gleaned from those successes. I make all of these entries beneficial, and I'd set the DC to where you could expect more successes than not. I'd get the checks from the party expert in each skill: the idea is that while spending fifteen minutes poking around the burial chamber the religion-character's going to get "hey--is this interesting" ten times, so it's that character's religion check that matters. For stealth, the thief isn't just being stealthy, but is helping make sure everyone else is, too.
For each encounter/location you've got to think of a way to telegraph its presence via the choice that leads to it. The smell of the sheep pens, fresh breeze coming in from the hangar, loud echoing conversation drifting from the atrium, &c. Probably doesn't need to be done ahead of time, but keep the five senses in mind.
Sometimes you'll enter a location and have different choices in front of you than last time. That's intentional, and needs to be explained in-game. This is another place where long time-segments helps. "Hey, last time we were at the treasury there was a way to get back to the stairs--why isn't that a choice this time?" "Because this time there's a bucket-brigade of Seekers moving supplies a few flights up, and your stealth expert tells you there's no way you're getting in and out of there without getting caught in a spotlight. And the path to the watchpost that 'wasn't here' last time: it was down a long-ish passage that you didn't follow very far last time, and this time you went further down."
Built into an extinct volcano is the long-since abandoned Hall of Dragon Riders. (Think Pern.) But millenia ago dragons forsook the world of humans and haven't been seen since. Flash forward three thousand years: A group known as The Seekers collect all known information about dragons, striving to preserve this history, understand why they left, and to bring them back. Seekers are well-known and -respected through the land, and have made the Hall of the Dragon Riders their ivory tower. Unknown by most are The Watchers, who believe we and the dragons are better off separate; Watchers separately and secretly search the land for dragon-knowledge and have even inflitrated The Seekers.
A Watcher has found a clutch of dragon eggs deep in the levels of the Hall. While Watchers the world over argue the appropriate course, one has hatched and is being secretly cared-for by Watchers. It has communicated that it does not want to live in proximity to humans, and it knows where to find the rest of its kind.
Your party is to escort this baby (curl-up-on-your-shoulder-sized) dragon from the clutch (START) to a caravan outside of the Hall (FINISH). It doesn't matter how long this takes--the caravan you're meeting is a daily supply-run, and that entire operation has been infiltrated by Watchers.
This is just a simplified version of the one I've already linked. I've unintelligently picked some paths and have no idea how well it all connects.
Notes: Colors are only intended to indicate crossings-not-junctions, they carry no other information. Entry/exit points are numbered for reference further on. The maze is traversed recursively: entering any of square A,B,C is to enter a copy of the entire structure. That is, our first move is from START into square A at position 5 (hereafter A5). Now we've entered a sub-copy at position 5 (the bottom edge of the full image) and have choices to enter C7, A3, A0, or exit this sub-maze at 0. Exiting brings us back out of the last maze we entered--A, in this example--at that position. So if we choose to exit at 0 we find ourselves at A0 in the original maze with choices of 0, A3, C7, and 5.
Obviously this maze is not solvable, as I've not connected anything to F. But you'll see it still suits our purposes!
- Dead ends: they don't happen. When you hit one, simply present the same choices as last time, but with that dead end removed. I counted hitting 'S' as a dead end, but one could make that an end condition.
- Each possible entry into a sub-maze generates a foretelling of entries from the appropriate table. An entry not realized is described again the next time that table comes up. Tic-marks next to your printed tables keep you from losing your spot in the cycle. Across the bottom of that same page you can record the stack and current (nesting) depth.
- Leaving a sub-maze generally just leads to other choices, not encounters. I like having a go-to choice-generator for each sub-maze: each time the party leaves A it's through a twenty-story switchbacking stairway, leaving B is getting to the outside slopes and scrambling to the few spots that seem safe, &c.
- My end-condition is "first time hitting depth of 8 or 0, after at least 20 entries and no more than 60 entries." (Remember, they might not get a rest! You may want to be nice and grant some while the maze is dormant, but I'd use somthing to disturb them after only a few hours. Rests really should be earned, in my book.)
(I've made party decisions by die roll.)
Party enters A at 5: encounters a watch post. They've got four choices: enter C7, enter A3, enter A0, or leave A at 0 (hereafter, choice A'0). Those correspond, respectively, to entries for the firepits, the hangar, the sheep pens, or paths that lead up other choices. Now you can improv a description of what they've gleaned from the last hour. "A gently-sloping ramp downward rutted with the passage of countless hand-carts [fire pits], a section of cavernously-wide hallways with a strong fresh breeze blowing toward you [hangar], slightly earthy-smelling halls that piqued the dragonet's attention [sheep pens], and a long set of stairs switchbacking up into darkness [choices]." Party chooses...
A0 (the sheep pens). From there the next choices are A0, A3, C7, A'0: hangar, ground level of atrium, firepits, and paths that lead to other choices. In our improv we've noticed the sheep pens connect to the firepits--perhaps burnt holocausts?5 In any case, the party chooses...
C7 (the firepits). After the climb up, probably-not-resting, the choices are A4, C'3, and C'4: the hangar, a path that only leads to a passage with Seekers6, and a dead end (describe as a gently curving passage with lots of small side-rooms). Party chooses...
C'3 (nondescript passage, but: Seekers!). Quickly choose (having already entered B3) from among B'4, A4, B'7: crypt-temple, hangar, or dead-end. Party chooses...
B'7 (dead-end retconned into the last two choices). So now, scrambling to avoid Seekers--they're ahead of you, not behind!--choose from among B'4 and A4. Party chooses...
A4 (hangar). Choices are A'3, A4, A'7: choices, atrium ground-level, or dead-end (straight path to A6, which is a dead-end). Party chooses...
A4 (atrium ground-level). Choices are A'3, A4, A'7: choices, water storage, dead-end. Party chooses...
A4 (water storage). Choices are A'3, A4, A'7: choices, smithy, dead-end. Party chooses...
A'3 (choices). Choices are A'0, A0, A'5, C7: choices, smithy, dead-end, crypt-temple. Party chooses...
A'0 (choices). Choices are A'0, A3, A'5, C7: choices, smithy, dead-end, crypt-temple. Party chooses...
C7 (crypt-temple). Choices are A4, C'3, C'4: smithy, choices, dead-end. Party chooses...
C'4 (dead-end retconned into last two choices). A4 or C'3: smithy or choices. Party chooses...
Turning off verbose mode: A4 (smithy, finally!), A4 (warm passage), A'3 (choices), C7 (armory), C'4 dead-end cum C'3 (choices), A'2 (choices; the previous choice made was C'3: an exit which brought us out of an A sub-maze--that's why you have to track the stack!), A'1 dead-end cum B1 (dry goods), A2 (unused sector), B3 (Giant Eagles; depending on how this is handled we might get a short rest 22 hours into the ordeal!), B'7 dead-end cum A4 (high atrium).
And you can continue and continue and continue....
Noting that B is the least-connected sub-maze, I assume that'll be encountered least--let's put bad things there. But nothing so bad as to generate a raging battle--I don't want to have that happen and then have the choice between (a) lots of reinforcements show up and (b) explaining why lots of reinforcements don't show up. But I do like that entry 1 connects to B2, so it's possible to hit two baddies in a row. Baddies hit a second time around may be described as "remains of your encounter" or may just be presented as a new encounter with improvved differences.
As for the other two tables, I took thematically-similar groups and split them across tables. I also put the seriously high-value locations at the ends of A and C, so it'd be a while before they're hit. (As A has the most connections it'll be hit most--it might have made sense to swap hoard and library, but I don't think there's a right answer here.) Hitting an entry twice is no problem--it's hoped-for, actually. Nothing like crossing that atrium a half dozen times before finding one's way out!
The hall leads you outside the mountain to a narrow ledge with a railing. It runs 50' along the side before re-entering the mountain. Perception: it's a watch-post.
A huge chamber, open to the outside air (though no paths down from the opening). Insight: hangar-bay for dragonriders.
You've found the sheep-pens. It smells overwhelmingly of sheep-pen. Animal Handling: this dragonet surely needs to eat!
You enter the vent (central opening) of the volcano at 'ground level.' It functions as a huge atrium, and there's a colossal (200') statue of a humanoid dead-center. From this level it's impossible to make out features. You see balconies and 'windows' peppering the surface of the vent. Religion: iconography suggestive of Cerelia, the dragon-goddess of ages past.
Small aqueducts lead you to cisterns and storage tanks. Enough melt-water to float a ship! Medicine: this is perfectly good to drink. Fill up the skins!
The smithy: long-unused, many scraps and tools are still here. Along with any martial weapon on the equipment list, and many that aren't. All dragon-motifed. Perception: you hear engineering-minded Seekers in a nearby passage worrying about why a wall of rubble recently shifted.
Recent passages have been steadily warming, air getting drier. Insight: this volcano's really extinct?
You're in an unused section of the Hall and have been wandering through deserted storerooms, cells, long-deserted kitchens, &c. for a few hours. [DM note: if the last OR next encounter is not strenuous, short rest granted.] Animal Handling: you and the dragonet have taken a liking to each other. It's roosting on your shoulder and preening.
The passages you're walking along have openings in side walls looking out into the atrium; you're perhaps 150' above the statue's base level and can tell it's an elven female in riding armor. History: wasn't there a 'golden general' in some of the ancient dragon-wars...?
The hoard: whatever hoard you're imagining for a fortress that housed dozens of dragons, this is bigger. Mostly coinage, but any hour's search will turn up 2d3 treasures from pp.132-133 of the DMG. (Concatenate all tables and roll %ile dice.) Insight: dumping this much treasure anywhere would really screw up the world!
Table B (bad things)
You hear a group of Seekers talking loudly as they come toward you in this single passage. Investigation: you hear one addressed by the name of a known Watcher.
You enter a room where d6 Seekers are cataloging/stocking dry goods. Stealth: you surprise them.
The gallery you enter has windows cut to the outside, and Giant Eagles have made their aerie here. They freak out at the sight of the dragon and will attack after d3 rounds. Animal Handling: prolonged cries from the eagles will bring any within earshot coming to aid of these.
You come up on a pair of Seekers actively mapping this part of the complex. Stealth: they haven't noticed you. Yet. (Give the players 1 minute on a timer to decide action.)
A rubble pile you're clambering up shifts LOUDLY. Dex save to avoid d6 damage from rubble. Acrobatics: you and a nearby party member auto-save.
Recent passages' air has been foul. (Party has disadvantage on the next four checks, saves, or attack rolls it has to make.) Medicine: only the next two rolls are disadvantaged.
Large and long-unused firepits led you to climb a long worked-stone chimney in the hopes of egress. At the top you find yourself in a small chamber with two small halfling-sized holes in the wall. Looking out you see balconies and the ground level of the vent far below. [It's the inside of the colossal statue's head: fires lit below cause smoke to pour out, and it's a lot easier to clean the soot off her nostrils from the inside than from without!] Insight: we'll not find a better place for a long rest!
You almost stumble into a ritual service being conducted by/for hundreds of Seekers in a crypt-temple. Religion: they are sincerely praying for guidance in finding the dragons, or at least some for understanding of where/why they went.
The armory: see smithy. But you're back in the foul air and are feeling sick very quickly. (Put the players on a stopwatch. At 30sec lowest-CON character drops unconscious. When they've left roll d4 for every ten seconds they were in there--party has disadvantage on that many subsequent saves/checks/attack rolls, or until long rest. Unconscious player remains so until next rest, and takes 1d6 damage.) Survival: halve all of the numeric effects.
The ammo dump: arrows and bolts by the thousands. Most in disrepar, but it'd not be long to find as many in good shape as you might like. For every ten ammo found successfully a critter bites someone for 1d3 damage. Nature: for each of these events you have a 50% chance to notice the critter and prevent the bite.
The main food stores: Seekers pretty-constantly going back-and-forth. Stealth: you can get in, spend some time, and get out without too much worry.
The library: multi-story with lots of table-space among stacks. Seekers actively use it, but there's enough space and entrances that it's not hard to get in and out unnoticed. Investigation: each section is devoted to a single dragon: there are many histories, oral histories, drawings, paintings, &c.
The fractal maze takes fairly little prep--I'd liken it to most of the other methods presented. Some descriptive elements and a simple drawing to guide you through the connections.
It gives you the ability to repeatedly get near some identifiable element, giving the party the sense that they're making-progress-but-not, and that they're in a unified grand locale.
It's easy to end at any time, since the party's probably so confused anyway!
You can do it all on one sheet of paper: I did! Maze in top-left quadrant, three tables in other quadrants, checked off entries and wrote down the stack across the bottom as I went. (I did need a sheet of scrap to track the roads-available in order to generate verbose-mode.)
If you've no idea what I'm talking about, try a fractal maze for yourself. This is the same one linked above, but comes as close as I can find to having an explanation of what's even meant by the diagram. Here's another instance of it, with less (and less-annoying) commentary and a more-clear solution. One thing that none make terribly clear: if you attempt to exit a sub-maze and there's no path there, you've hit a dead-end and need to backtrack.
It'll be solvable in the CS sense, but not solvable in the sense of 'players/characters, taking care, have any chance of deducing their way out.'
My preference is cycling, for two reasons. (1) I think the repeated proximity of an entry greatly enhances the maze, as we'll see with the example's smithy. (2) the reduction of rolls on your part makes itseem like you've got the whole thing there, ready to go, and you're just plotting their path through it. Which you do, after a fashion! Just without the work =)
I don't know if the context helps the example or just over-inflates an already-long answer: I'd particularly welcome feedback/edits on that point. I just thought there'd be no way to explain the concept without an example, and that an example'd need a game-setting.
By this point in the example the whole thing's starting to take on a different feel than it did in home play--because different connections have been forged and so different reasons have been retroactively proposed. With suitable tables I could imagine this one sheet coming out every time you need some twisty ways anywhere on a continent!
This is actually an error--I didn't notice the little path off to exit at node 2 and read this edge as a straight path. But the system's robust!