Comment on page
Building A Better Dungeon
There are three keys to building an effective dungeon crawl: Believability, variety, and non-linearity.
The first key is believability. Dungeons are sometimes laid out essentially (or even literally) at random, with the first room containing a pack of goblins, the second room containing an ogre, and the third a displacer moose. If the players attack the goblins, the ogre won’t hear and come to investigate. If they kill the goblins, retreat, rest up, and return, the first room will still be full of dead goblins and the ogre will still be waiting around in room two. The dungeon will have no notes for what the ogre and goblins and displacer moose are even doing there. They’re just the dungeon, a bunch of monsters camping on a bunch of treasure for players to kill. This problem is getting more and more rare as published modules and adventure paths are getting more and more popular, and they almost never make the mistake of having a dungeon whose inhabitants don’t even have any reason to be there, but in the interests of being thorough: Your dungeon should have a reason for existing. The monsters should live there for a reason. They should have treasure for a reason. Preferably they should pose some kind of actual threat to the surroundings rather than just being murderhobo targets.
The reason why the monsters are in the area should inform the way they use the dungeon space. If they’re raiders, they’ll need a place to keep their loot and may not be as organized as if they were a forward attack base for an invading army. If they’re scouts for a dark lord, they’re probably much more interested in keeping exits open to flee and report back than if they’re a stronghold that’s meant to serve as a secure location from which to patrol the area and to resist siege.
The reason why the dungeon was originally constructed should also inform its layout (which we’ll discuss in more depth when talking about non-linearity). A mine that’s been overrun by kobolds will have a very different layout from a dungeon intentionally constructed as a fortress. It’s particularly important to bear in mind this aspect of believability when placing traps. If the dungeon was not originally meant to repel invaders, then unless the current inhabitants are themselves engineers and have access to enough manpower for serious remodeling, they probably can’t build any traps directly into the walls or floors, like a dart trap triggered by a pressure plate. However, they can still set up crossbows at one end of the hallway and rig them to go off if someone breaks a thin wire strung across the hallway at knee height.
Also, bear in mind that monsters have to actually live here. The path from the sleeping quarters to the toilet can’t have active traps going all the time or it’s going to kill more of the defenders than anything else. Traps between living areas and dining areas are eventually going to kill their owners unless they’re only armed when the dungeon is under attack. Traps which can be armed and disarmed can be just about anywhere, but bear in mind when running the dungeon that someone has to actually arm them. Whose job is it to do that, and how long does it take them? It might keep a goblin or two out of an encounter for a while.
On a related note, the monsters live in the dungeon. Even if they’re a raiding party or soldiers and have no civilian population (and most dungeons are, because most dungeons do not want to deal with the baby orcs problem, although if you do plan to use that particular moral conundrum in your game, everything about monsters living in a dungeon used as an army base goes double for a dungeon used as a village), soldiers and raiders still need to eat, sleep, and play a rousing game of “can the human eat it?” to keep up morale. They need places to eat, places to sleep, and if they’re going to be here longer than a week, they need places to relax, in addition to guardhouses and chokepoints and so on. A significant chunk of the dungeon’s population will be eating, sleeping, or relaxing until the alarm is sounded, and only at that point will they grab their weapons and get to battle stations. Particularly for monsters who were asleep and thus out of their armor, it may take 2-3 minutes from the sounding of the alarm to get ready for a fight and arrive at a guard station. Players may find monsters who, in an ideal (for the monsters) situation, would be fought together are instead split up in corridors and barracks pulling their boots on and heading to rendezvous points to form up with the other monsters in their cohort, and thus be able to pick off piecemeal an encounter that might’ve been much more difficult when fought all at once.
Sword fights make noise and monsters from the room just down the hall will probably come to investigate or reinforce their buddies. Disciplined monsters might have specific rendezvous points and might leave their friends to die in the room down the hall in order to head to that rendezvous point. Less disciplined monsters might rush straight for the fight. Cowardly monsters might stay in the room they’re in (unless bullied into doing otherwise by bigger and braver monsters), but they might also flee deeper into the dungeon and seek safety in numbers. Even if they do stay put, they’ll still bunker down and be prepared for a fight when the players enter, and probably have a pretty good idea of which entrance the players will come through and ready a bunch of actions to pincushion the first living thing that walks through.
If a dungeon is big enough (and many dungeons are), you might need a random encounter table to keep track of monsters who meander around, rather than keeping track of all the monsters yourself (something which is really only possible with a one page dungeon). Instead of figuring out exactly where each and every monster is, populate major rooms with specific encounters and then have reinforcements, patrols, monsters just wandering around restless and maybe a little drunk with their buddies, and so on be pulled from a random encounter table. Don’t make this table infinite, though. If players encounter five orcs led by an orog when they roll a 10 on the encounter chart, maybe have three of those random encounters (something which I usually write down as “five orcs led by an orog (3)”) and once they’ve encountered and slain or driven off all three of those encounters, that slot on the random encounter table is just empty. As the dungeon’s population grows more sparse, the odds that there will be anyone around to reinforce a guard room that sounds the alarm grow slimmer, the hallways are less populated and the party is less likely to encounter creatures at random while wandering them.
Another important reason to have a finite amount of random encounters is because, once alerted, dungeon lords will want to gather up those random encounters into a reserve force (monsters who are in static encounters hanging out in the barracks or currently asleep or what-have-you may also end up in a reserve force), which they will then drop on top of the party at first opportunity. Dungeon lords don’t have the omniscient perspective of a dungeon master, though. First a monster needs to survive long enough to sound an alarm. If the alarm doesn’t alert the entire dungeon (for example, if the dungeon doesn’t even have a proper alarm, just sentries who run to alert the dungeon lord when the entrance is attacked), then the dungeon lord will have to send out runners to alert all those random encounters and off-duty monsters to show up and form a reserve force.
Even if the alarm goes through the entire dungeon and all off-duty monsters immediately report to the reserve force, that’s still going to take much longer than the ~30 seconds (5 rounds) the party will take to finish up the dungeon’s first encounter and start exploring. This means that the party won’t be where they were when the alarm is sounded, which means the dungeon lord needs to find them, which means he’ll be chasing after alarms and scout reports trying to figure out where the party is going and beat them there, or at least catch them while they’re still fighting. If the reserve force ever does catch up with them, however, it’s going to be a lot of opposition dropping on the party all at once – probably enough for a party wipe if they aren’t smart enough to flee (at least far enough to find some terrain advantage like a choke point).
Of course, the dungeon lord needs to be ready for any kind of attack, including the possibility that the PCs are just a distraction from a lone infiltrator or another PC attack through another entrance. This means that they probably don’t want all the random encounters to join the reserve force. They need to leave some behind to check for infiltrators, and to sound the alarm if a second party attacks a different entrance and the reserve force needs to split in two. These tactics are rare for PCs, but mostly for metagame reasons that it’s more complicated to split the party and there’s almost never enough players in the group for an effective two-pronged attack. In-universe, the dungeon lord has no reason not to suspect that a nine-man adventurer party may have split into one party of four and one party of five and used one party as a distraction while the other makes a beeline for the treasury after the reserve force has mobilized in the wrong direction. What all that means is that the dungeon lord still wants his guard posts manned, even on the other side of the dungeon and still wants random encounters patrolling as much of the dungeon as possible. This is good, because it helps give the party a fighting chance against a fully alerted dungeon without harming believability at all.
Monsters have goals. Usually this is some kind of military goal, they’re here to raid or invade and are using the dungeon as a base. The dungeon should be set up such that they can actually accomplish that goal. An army marches on its stomach, and it occupies a dungeon on its stomach, too. What’s the monsters’ food supply? Do they have several weeks or months worth of food reserves in their dungeon, and when those run out they’ll retreat back to friendlier territory? Do they feed themselves based on constant raiding, or by demanding tribute from nearby villages? Is there farmland surrounding the dungeon used to keep the dungeon supplied by monsters who flee to the dungeon’s safety when the adventurers roll into town? Do the monsters have Dwarf Fortress style underground mushroom farms within the dungeon itself?
What is the monsters’ goal, and how are they going about accomplishing it? If they’re here to invade, then they will need to leave the dungeon to do battle with defending forces. Do they plan to return to the dungeon and use it as their stronghold to occupy the area, or is it too remote for that purpose? If the former, how many guards do they leave behind when they leave the dungeon to give battle? If they’re using the dungeon to patrol and maintain control over territory, how much of their forces are deployed to those patrols? Do they come back each night (or each day, for nocturnal monsters), or do they patrol for days or weeks at a time? If they’re raiders, how much of their forces do they commit to a raid, and how long does it take them to get back from a raid target to the safety of the dungeon?
Make sure your dungeon’s inhabitants are actually accomplishing their goals (or at least trying to). This will almost always make them more vulnerable to attack, but that’s a trade-off that the dungeon’s inhabitants have to work with. On the subject of tradeoffs, a dungeon shouldn’t be prepared exclusively for an attack by a party of 3-6 powerful adventurers. They should be equally prepared for a lone assassin, a large army of weaker creatures, or a small number of powerful but distinctly non-human assailants, like dragons or beholders. Spreading guards out helps to detect a lone assassin, but also makes them more vulnerable to adventurers. If you’re focused only on defeating adventurer parties, that seems like a stupid move, but the dungeon doesn’t exist just for the players, and sometimes the dungeon lord might need to allocate their resources in a way that makes them more vulnerable to players in order to defend against other threats.
Can the dungeon defend itself against a large army? What about aerial opponents (if they’re a classic underground dungeon, the answer is “yes, because we force them to land in order to get into our dungeon at all,” but a military camp that works like a dungeon but is open to the air needs some kind of medieval AA)? What about one especially sneaky opponent? What about enemies with lots of breath weapons or other AoE attacks that punish tightly grouped opponents? What about enemies with high damage, single-target melee attacks that punish spread out opponents?
Lower level dungeons won’t have an answer for all of these threats. A goblin war camp’s response to an attack by an adult dragon is probably “grovel,” a bunch of kobolds relying on traps probably can’t reset them in the middle of a fight and probably can’t do anything against a large force of weak creatures (aka the Robilar strategy) except run, and a necromancer lord’s skeleton hordes might be dumb enough to fight tightly packed (all the skeletons just make a beeline for the nearest threat) even when it is advantageous to be spread out, so there is no point in having a strategy to funnel AoE enemies into a wide open room where a loose formation can be employed. The skeletons can’t actually adopt a loose formation, so why bother? However, monsters should still be prepared to deal with as many threats as their resources allow, even if that means they’re less specialized for defeating the player characters. The exception is if the player characters have proven to be far and away the greatest threat to the dungeon’s occupants, in which case they might reorient their defenses to deal with them specifically.
For the sake of variety (which we’ll get into later) there are often multiple monster factions in a single dungeon, and they may be only loosely allied with one another, or even openly hostile to one another. If this is the case, after players retreat from a partially cleared dungeon, consider how this affects the factional politics. If a faction is weakened, another faction may push to exterminate them completely and claim whatever’s left of their treasure, as well as any valuable rooms of the dungeon (like the barracks with the good beds or a more defensible guardroom). This might result in a scramble for territory that depletes all the factions’ forces. Don’t spend too much time calculating exactly how many orcs die when mopping up remnant goblins and skirmishing with drow doing the same thing, but do reduce their forces a bit and have the orcs and drow go ahead and slice up the weakened goblins’ territory if none of the three really get along.
Even if all monsters in the dungeon get along, they still won’t just leave valuable territory vacant. While it’s possible that a wing of the dungeon wasn’t used for anything else except for the goblins to live in, and when some adventurers came through and killed all the goblins there was nothing left to do with that section, it’s also sometimes true that a section of the dungeon contains important parts of the living space like the kitchen or locations of strategic importance like a guard room or watchtower, and these locations are going to be repopulated by monsters taken from other sections of the dungeon.
All of this makes a dungeon more believable, the first and most important key to designing and running a good dungeon. By following these principles, the dungeon becomes a living thing with a purpose that reacts believably to player actions. This verisimilitude not only helps immersion, it also means the players can more effectively plan ahead. If the world makes no sense, then there’s no difference between a good plan and a bad plan. The more sense there is in your dungeons, the more it makes a difference how good the players’ plan is.
The second key to good dungeon design is variety. Dungeons are big and players don’t want to fight the same encounter over and over again while clearing them. Even if you have a one-page dungeon that’s just the lair of some orcs, you should still include enough variety to keep things interesting. Use an Eye of Gruumsh and an Orc Warchief as boss encounters in certain rooms, give some orcs longbows instead of greataxes and have them engage at range while their standard orc buddies hold the frontline, give some orcs dire wolf mounts to make them into powerful cavalry, and now you have five different encounters (including the standard “just a bunch of orcs”) without even dipping into room design, traps, making unique monsters, or having multiple factions in the same dungeon.
Room design is an important part of encounter variety, but make sure your monsters are able to use the room’s layout to their advantage. For example, if you have a sneaky orc ranger in one room, you might make it a large natural cavern, but give it a lot of stalagmites for him to sneak off behind to lose the party if they find him. An orc sorcerer with lots of fire magic might fire on players from a ledge above a cramped corridor where they can’t spread out to avoid his AoEs, then retreat into the side-path leading to that ledge if the players turn out to have some serious ranged firepower of their own. Monsters live in a dungeon, so even if they’ve only been here a week or two they probably know the place well enough to have figured out what areas are best suited to their abilities, and will be assigned there (or just gravitate there naturally if they’re too disorganized to have any assigned patrols or guard stations).
You can squeeze more unique encounters out of the same set of monsters by taking advantage of room design. Using just regular orcs and orcs using longbows, you can have a brawl in a mess hall where the ranged orcs use their melee buddies to keep their distance, a room split in half by a chasm where orc archers on one side will attack the players with impunity if they come into the opposite end, but will be forced into melee if players come through the door on the archers’ side of the chasm, and a bridge surrounded by arrow slits that archers use to soften up the party while a mob of orcs rush across to meet them.
Traps are another way to add variety to a dungeon. You can just let a trap stand on its own, but they’re often more interesting if they’re part of a larger encounter. For example, a pressure plate triggers a fusillade of darts, softening up the party, and then immediately afterwards ambushing goblins leap out to try and finish them off before they can heal. If a trap is going to stand alone, make it a set piece, like a room whose walls close in to crush the occupants or which floods to drown the occupants. Bear in mind believability when designing these traps. Whatever monsters live in this dungeon have to actually build these things. If they can’t make a room watertight, they can’t make a drowning room trap. They’ll need some serious mechanical power to make a crushing walls trap. Also bear in mind that some “traps” may not be intentional at all. A falling bricks trap might just be because the dungeon is old and starting to fall apart. You might have them strike squares at random during a large melee in a cramped area. You can be pretty sure they’ll hit something (especially if one or two of them go off every round), but they might hit the monsters just as much as the party.
Varying the rewards can also keep a dungeon interesting even if the encounters (including their room design and traps) are starting to feel same-y. Players might have fought a half-dozen-ish orcs led by an Eye of Gruumsh twice already, but if the first time they got boots of speed and the second time they got a mace of terror, then by now they’ll have picked up that the Eyes of Gruumsh are the keepers of the dungeon’s magic items, and wondering what magic item this one will be guarding can keep them invested in an encounter that might otherwise be repetitive. This is doubly true because the Eye of Gruumsh will probably use the magic item he’s guarding, which means fighting him might have a touch of flavor to it that sets the fight apart from the other Eyes.
We covered this in the Art of Encounters, but it is important to repeat here: Players decide what kind of encounter any given room will be. If you sense they’re getting sick of fighting through things, you can suggest that they might take a stealthy or diplomatic approach instead. Preferably, a dungeon should have enough variety that even if you just slaughter every creature inside of it, it still won’t get old, but if that fails, encouraging players to switch up their own tactics can help keep content interesting for longer. You can also design some encounters such that they lend themselves more clearly to one approach or another. More specifically, designing some encounters to be more open to negotiation than others can help draw more variety out of a dungeon. The party might stumble across prisoners who, once freed, might be convinced to help you fight against their captors instead of immediately fleeing into the wilderness, or they might find evidence of a rift between the dungeon lord and one of his lieutenants that they might be able to parlay into a civil war.
Most of these tricks will help squeeze just a few extra rooms’ worth of variety out of a dungeon. If you want to keep a very large dungeon from growing stale, you need to have noticeably different areas to the dungeon, with different monsters who fight in a different style with different terrain. For example, hobgoblins tend to directly engage opponents with well-balanced teams that are prepared for many different threats, and will usually take up positions that favor what their group is good at (which may vary from one encounter to the next, as different hobgoblin war parties have different units). They’ll fortify whatever part of the dungeon they’re occupying to better suit their defenses, but they won’t make very heavy use of traps or ambushes, instead using things like chokepoints and reach weapons to force one enemy at a time to fight two ranks of hobgoblins, or using arrow slits in fortified towers to fire on enemies who will have difficulty shooting back. Regular goblins, on the other hand, will have more ramshackle living quarters, and rely more on traps and ambushes.
Orcs will be as straightforward as the hobgoblins, but less disciplined, usually charging straight towards the nearest enemy. They’re plenty capable of some tactical planning in advance, but once the fight starts they’re much harder to control, and their plans will take this into account. Orcs fight for glory, and you can’t expect any hot-blooded orc warrior to stick to a plan that involves hanging back while there’s glory to be had in battle. Ogres are very big, very strong, and not so bright, so they’ll tend to be very bold against creatures smaller than them but more cautious against creatures their size or especially ones larger. Trolls are cunning ambush predators who lurk in narrow crevices and wait for an unsuspecting creature to wander by, then attack the luckless victim in melee, and in tight quarters where escape is difficult.
Examples keep going for as long as the Monster Manual does. Use the culture and abilities of the monsters you’re using in the dungeon to inform how they fight and how their section of the dungeon is built (especially if they built the dungeon from scratch, which means each and every room was placed because they wanted it there). Not only does this improve believability, it more importantly means that going from a section of the dungeon controlled by orcs to a section controlled by goblins will be a very distinct shift in how the opposition fights.
As discussed in the Believability section, different types of monsters can also make different factions who may be anywhere from closely allied to one another to openly at war, and this can open up lots of possibilities for player interaction. Two monster factions which are very closely allied are essentially the same faction, and may even be intermixed in the dungeon. If orcs and ogres get along swimmingly, they’ll probably share barracks, mess halls, and so on with one another.
Two monster factions who formed a coalition for a specific purpose, however, will keep to themselves, and may even turn on weakened members of the coalition if they don’t think they’re strong enough to be a meaningful contributor anymore. Some member factions of the coalition might be more loyal to it than others, and some coalition factions might be convinced to back out of or even betray the coalition. All of this is significantly exacerbated if the coalition is between normally enemied factions brought together by a common enemy or a strong-willed dungeon lord. In the latter case especially, the death of the dungeon lord will probably see the factions of his dungeon coalition immediately going to war with one another.
Just because monsters share a dungeon doesn’t mean they aren’t fighting one another. Dungeons may be in a cold war in which any sign of weakness (say, because players depleted forces) will result in an attack by neighboring factions, who will occupy the territory. Capturing new territory doesn’t give a monster faction new troops out of nowhere, though. They’ll have to stock the new territory by either draining some of their random encounters to make some new static encounters for the captured territory, or else by moving some of their static encounters out of their section of the dungeon and into the captured territory. Either way, monsters who aren’t reckless should weigh the benefits of newly captured territory against the risks of spreading themselves more thin. To adventurers, a dungeon is a pinata full of treasure and blood, but to monsters, it’s a Risk board, and only foolish dungeon lords will make the mistake of going for Asia.
Speaking of random encounters, different areas of the dungeon should have their own random encounter tables, which is another way that one faction controlling multiple areas gets stretched thin – they have to split their random encounters between two tables. If monsters are in a fairly friendly coalition, random encounters from other factions might show up, as a couple of drow drop by to chat with their orc buddies while off-duty or what have you. In dungeons where monsters are actively at war with one another, random encounters might include an incursion of forces from a hostile faction, who might be persuaded to team up with the players for so long as they have a common enemy.
With the possible exception of enemy variety, faction variety is by far the deepest well to increase the variety of your dungeons and keep larger dungeons interesting for their duration.
The third key is non-linearity. In recent years, the dungeon crawl has often been the victim of linear design. The subterranean nature of a dungeon makes it pretty easy to keep players on track, and particularly in the 3e and 4e days it was often considered close enough to a real dungeon crawl to just have some twists and bends in a linear route without actually branching the corridors at all (you get this sometimes in the Adventurer’s League, too, which I hope is mainly because they need to make sure the adventure takes a specific amount of hours to complete and having a very open dungeon crawl would interfere with that something fierce).
This harms believability, because a dungeon is almost always a location that would not have such a linear construction. Whether it was intentionally built as a subterranean fortress or it was repurposed as such from a mine or natural cavern, there’s almost never a good reason for a dungeon to have few or no branching paths. It also removes one of the best sources of fun from a dungeon crawl: Exploration. Deciding where to go next and slowly filling in the blanks on a map. It even harms challenge, because exploring a challenge and deducing where targets are based on contextual clues from a well-designed dungeon is a skill. Even if a party lacks the resources to clear an entire dungeon, they might be able to find the parts they’re looking for with some deduction and/or interrogation. Clever scouting hardly makes a difference if you have to go through nearly every room in the dungeon to reach the goal anyway.
The most basic elements of dungeon layout are loops and branches. While a branch leads to two separate areas that don’t lead to one another except by backing up to the branch, a loop is a series of areas that are all connected to one another. A branch provides a simple choice, although if one branch is easily cleared then it’s hardly a branch at all. The party will go down one branch, reach the end, then turn back and go down the other. However, branches can be used to interesting effect if they contain secondary goals (and sometimes it just makes more sense to have a branch than a loop – don’t sacrifice believability on the altar of non-linearity, especially since believable dungeons will be plenty non-linear anyway). For example, the monsters’ treasury might be located behind a series of traps and guardhouses that branch off of a loop in the dungeon, or perhaps there is a lair of a wyvern that branches off a loop in a goblins’ lair.
Loops are the cornerstone of non-linear dungeon design, however, because they provide the choice of going around encounters. This is particularly true if the dungeon has been alerted to intruders. An encounter might do its damndest not to be gotten around, but if there’s a loop, the party can get around it if they’re quick or sneaky enough. Loops let you decide not just what order a party wants to deal with encounters in, but gives them more options for dealing with those encounters and lets them decide if they want to deal with certain encounters at all. Players skipping an encounter might seem like the encounter is wasted effort, but the ability to skip encounters lets players know they have actual freedom within the world, and that is worth the effort of making a few encounters that get skipped.
Hidden entrances, both into the dungeon itself and onto specific dungeon levels, are another way to make loops not just for a few areas but for entire levels of the dungeon. A second set of stairs (or a ladder or chute or teleporter pad or whatever) between levels 1 and 2 turn the entire dungeon level into a loop. Likewise, if there’s a hidden entrance straight into level 3 of the dungeon, then levels 1-3 become one giant loop. Readers might reasonably wonder if potentially cutting out two entire levels of a dungeon isn’t perhaps pushing the limit of sacrificing work for player freedom, but remember, dungeon inhabitants shouldn’t be waiting around to be mulched into XP by the player characters. If the players find a hidden entrance to level 3 of the dungeon and they decide to use it, that means there are two completely untouched levels of the dungeon lying around to make a reserve force out of, and if that reserve force follows the PCs down, they will be between the PCs and the exit. If the party finds a secret entrance between level 3 and level 4 they can sneak around that reserve force, get to the hidden exit on level 3, and slip out without fighting that reserve force, but the key word there is “if.” If the players don’t find that hidden connection between level 3 and 4, then in order to get back up to level 3 they will have to use the main stairwell that every spare monster from levels 1 and 2 (and any they missed on level 3) is now camping on. This is the kind of considerations that make clearing a dungeon feel more like an assault on an enemy fortress and less like a Diablo clone with hideously sub-par action resolution speed.
Sub-levels are when one floor of the dungeon has two or more connectors that each lead to a lower level, and these lower levels do not connect to one another. For example, on level 4 there might be two different stairways down, one leading to level 5A and the other leading to level 5B, but there are no corridors connecting levels 5A and 5B together. One of these might be a dead-end, so for example the stair down to level 6 might be only on 5A, or they might both have connectors leading down to level 6, in which case levels 4-6 together form yet another loop. Non-looping sub-levels can be a good way to cordon off an optional sub-section of the dungeon, especially one that contains some important side goal. Maybe the goal of the dungeon is to kill the monsters’ leadership in the fortress on the final level, but there’s a sub-level branch containing their treasury, including several magic items that none of the monsters can use themselves, but which might be helpful to the party, who have a wider variety of character builds and explore a wider variety of fantastic environments (a driftglobe might not be seen as very helpful to monsters with darkvision and a ring of water walking isn’t very helpful to a monster who lives in the desert, but a multi-racial party probably doesn’t all have darkvision and wandering adventurers are probably going to end up at a large body of water they’d like to walk over sooner or later).
Sufficiently large sub-levels (looping or otherwise) can be dungeons unto themselves. In fact, one way to make a large dungeon in a hurry is to take multiple small ones and Frankenstein them together into a bigger one. You have to be careful about making the seams too obvious, but adding a few connections between a series of natural caverns with goblins, a catacomb overrun by undead, and adding a surface lair that’s a small temple of Tiamat controlled by kobolds and dragonborn isn’t unreasonable. The catacombs were formerly a burial ground for the temple, which was once dedicated to some other god before being seized by the Cult of the Dragon, and some goblins from a nearby natural cavern broke in mostly by accident while tunneling out some new accommodations. Connect the entrance to the catacomb to a room in the temple, then connect a room somewhere in the cavern to another somewhere in the catacomb. The temple and the cavern both have separate entrances. Even if the dungeons were fairly linear before (for example, if you took a few dungeons from the Adventurer’s League adventures for Tyranny of Dragons) adding these connections between them will make them passably interesting.
This is a relatively basic example, and nested dungeons can get much more complex. For example, imagine two (or more!) dungeons that both descend somewhat parallel to one another into the earth. Level 4 of dungeon A connects to level 3 of dungeon B, and further down level 6 of dungeon A connects to level 4 of dungeon B. These kinds of connections are especially useful for megadungeons and won’t come up so much if you’re hoping to use a dungeon as just one adventure.