Random vs procedural generation for game content

Random generation and procedural generation (which is a little more than 'random generation with the crusts cut off') are powerful tools. As GMs and designers we use them to spark inspiration or to populate a game or setting with interesting content. But we use them, or ought to use them, in different ways.

Here's the difference, as I see it.

Random generation

Roll a dice, get an outcome. Or variants thereof.

Random generation is

👉 Extremely quick
👉 An opportunity to use big random tables, which everyone loves
👉 Useful for inspiration by hitting possibilities you wouldn't have thought of, even if you don't use the actual prescribed results 

But it may also result in content with major dissonances and peculiarities.

I think a little incongruence is good, but generating too much randomly can create a sort of silly homogeneity where everything is dissonant, nothing quite makes sense, and no world elements are linked or correlated except by accident. You get ooh-so-zany slop.

Over-reliance on randomness thus requires work to repair, which can itself be entertaining, but takes time and effort.

For example, suppose you roll a random encounter across multiple tables, and get: [roll] hundreds of ... [roll] skeletons ... dressed as ... [roll] jesters .... and they're [roll] hauling a cartload of ... [roll] scented soap ... through [roll] a desert in bloom.

It's something you wouldn't have thought of, sure. But it's too dissonant to be usable in most settings. You hear it and maybe an inkling of a consistent narrative jumps to mind, but the designer needs to plaster over the gaps, tweaking or adding things, to actually get there.

Zany. Too zany? Image by CDD20, via Pixabay.

Procedural generation

Technically a subtype of random generation, procedural generation uses a more complex function to go from inputs to results. It uses randomness, but has deliberate human design built in to create plausible outcomes.

This takes longer, both in the up-front design work and the use of it for creation.

Procedural generation...

👉 Both uses and creates interlinked/ordered/self-consistent information
👉 Combined human decisions and design with randomness to spark novelty but to constrain it in certain ways
👉 Results (ideally) in more plausible or verisimilitudinous outcomes

Examples include:

👉 Branching tables and subtables
👉 Little bits of pseudocode for dungeon room generation
👉 Features added to a consistent core, like a themed monster which gets appropriate thematic features generated via branching dice tables or if/thens

One Application

I'm currently working on a zine. Its premise is that wizards are powerful, crazed, unique, and have strange features which align with their individual theme. For me, that means procedural generation.

I'm exploring a process of taking a wiki article from, say, Wikipedia and pulling out a bunch of descriptive phrases. From these, you get semi-structured thematically-coherent content. Then you follow some guided rules to apply them to the underlying core, which is 'a creepy and erratic wizard'.

After enough customisation, you have generated yourself a wizard. The random element of procedural generation means your character has features you wouldn't have thought of yourself. The human-guided element means the wizard should be more plausible, or at least thematically consistent, than pure randomness generally achieves.

(More news on the zine as I progress with this idea!)

On edge cases and magic

There's an innate design tension which RPG magic systems must contend with.

Magic must be more powerful than enacting the equivalent effect by mundane means, otherwise it's not fun and not fantastic. But if it's powerful and its uses are limitless (or at least non-rate-limited), then you become committed to an exceptionally high-magic setting.

If it takes ten seconds to give yourself a magical boon which lasts for ten minutes, the default player behaviour and the reasonable in-world behaviour will be to cast the spell every ten minutes... Unless they have a specific reason not to. That's almost always 'resource management'. In almost ever fantasy TTRPG, magic (whether spells, items, rituals, innate powers, etc) will be powerful and limited, i.e., a valuable resource; something to be conserved for when it will matter. Therefore, players will want to know the exact effect of some magic they might use, so that they can make an informed choice.

And this creates a design tension. How predictable to players is the magic their characters are harnessing?

The predictability predicament

Assuming you want magic to be able to do a huge range of things (it's magic), to get breadth of coverage, you either end up with these huge, minutely-detailed lists of magic effects and parameters (almost certainly player-facing), or you have a fuzzy system where the GM instantiates vague wordings at runtime (possibly via what amounts to negotiation with the player). These game design choices have different costs in terms of usability and satisfaction.

Most big, popular fantasy TTRPGs like D&D go with the huge, minutely-detailed lists of magic effects option. This arguably results in greater player agency, in the sense of being able to make good and meaningful decisions about resource allocation, but it constrains magic to the big official list (and if it's not exceptionally well-designed, you end up with weird codified corner cases and interactions). You also need to know the system really well to be effective as a player or a GM. Finally, there's a risk that presenting in exhaustive detail what is possible strongly implies that anything else isn't possible. You get cases like D&D 5e, where many destructive spells officially don't damage objects because a small subset of them specify that they do.

The fuzzy system option is attractive because it makes magic more 'mysterious' and wild, and thus fantastical. But it can be hard for a GM, especially one who isn't already operating in a "rulings not rules" tradition. But it can also be hard for players to come to terms with, unless either (a) it's the case in-world that a mage never knows exactly what they're going to get when they do some magic (which is flavourful, but risks a poor play experience), or (b) the GM and player together determine what happens when the magic is used. The latter might sound good (an opportunity for collaborative buy-in) or bad ('Mother May I' gameplay). Some GMs are surprised to find that many players hate 'collaborative storytelling exercises' spilling over into their roleplaying games. Either way, it doesn't actually resolve the tension: it's still in the player's best interests to stop and discuss the possible magical effect before they can make an informed choice whether to use their magic.

Variants of the fuzzy system where e.g. the GM has to follow certain rules to come up with a power-level-appropriate instantiation of a spell at run-time are close to being the worst of both worlds: mechanically fiddly but without giving a player enough information to make decisions with, and you just end up building that minutely-detailed canonical list slowly over time.

Resolving the tension

I consider this an open problem. As I've suggested before, I think the best practises are to

• Pick a path and stick with it, rather than mixing up how rigid and predictable the magic system is in different use-cases within the same game; and
• Have an extremely strong, well-thought out metaphysics which either underpins the mechanical design for the minutely-detailed lists option or informs what 'should be the case' for the fuzzy system option.

Food for thought

Here are some questions your magic system must be robust enough to answer, consistently:

If you turn flour to silver dust and mix it into acid, changing it to silver nitrate and creating heat, what happens to the silver nitrate if the spell wears off? What happens to the heat? If the compound is left over, can you reduce it to get silver? Does the resulting silver still count as being magical?

If you turn a pile of stones into loaves of bread and spend a week eating them, what happens to your tissue, the partially-digested bread, and the waste, if the magic ends? Do you die instantly? Does the energy you got from eating disappear?

If you turn a stone into a cat, and that cat eats and sweats and breathes makes waste, then the magic is dispelled, what happens to the bits of world (food and air and water) that got incorporated into the cat's body? Are they in the stone? Do they turn to stone? Can they be turned back? What happens to the cat's sweat and waste and loose hairs and skin particles out in the world? Do they turn to stone? What if the cat was a kitten, and it grew much bigger and older before the spell ended?

Or if the original subject was a mouse rather than a stone, what happens to the cells that changed in its body? Was there a 'mapping' from one mouse cell to multiple cat cells, so that injuries and shed cells and replaced cells carry back over? Do microbial diseases carry back over? Does cancer?

Are all your answers consistent so far?

What happens if you turn someone into a statue and snap their arm off? What about if you turn someone into steel and then temper or anneal the metal? Or take a mould of their body, turn them into ice, melt them, and then refreeze the water in the mould?

What about if you turn a piece of metal into a piece of wood and burn it inside a big tank of air? Does the carbon dioxide turn into a metal oxide? In general, if you heat something up while it's magically transformed and then turn it back, is it still hot? What if you turned something really small into something really big and then it just got slightly warm by soaking up a million times more sunlight than the tiny thing would have?

If you expand a mouse to be the size of a horse and it eats a bucket of oats, what happens to the oats if it turns back? Does it matter whether they've been digested?

If you expand a mouse to be the size of a horse inside a mouse-sized terrarium that a mouse couldn't possibly break but a horse easily could, does the spell work? Does the answer change if the terrarium is nearly horse-sized? What about if you try it on a mouse that's inside a crack in solid rock? Is this meaningfully different from trapping a mouse under a piece of heavy cloth and then expanding it to the size of a horse?

If you cast a spell on a subject, then another one that seems to have the inverse effect (flesh to rock then rock to flesh), do you get back the exact original? If so, why? If not, what happens if only the first spell ends or is dispelled? What if you cast the first spell, then make some orthogonal change to the subject, then cast the second?

What in-universe mechanism decides whether an object counts as meeting the targeting requirements of a spell? Does a spell that works on chairs work on a one-legged stool if the caster doesn't think a one-legged stool is a chair?

Do any of the answers to these questions change depending on whether it's a temporary effect, a dispelled permanent effect, a magically reversed effect, an intrinsic ability of a magical creature, the effect of an item, etc?

How I think about story games

What follows isn't an attempt at a codification or taxonomy or definition. In fact, it deviates from some of the language people use to describe TTRPGs and story(-telling) games.

It's a way of looking at the design space that I've personally found useful, and a sketch of some of the ways people end up arguing at cross-purposes about story.

 

Story  ∈  TTRPG

First, the process of playing a tabletop role-playing game does inevitably create stories: whether you think that's the point of playing or not; whether you think the GM should try to steer it or plan it or not; whether you think there's story from start to finish or only after the fact.

Back in 1987, Gary Gygax advised that

When the principal characters in a story (the campaign) are free-willed and have a multitude of choices regarding how to proceed, it is counterproductive – and, in fact, impossible – to preordain just how the events in the campaign will unfold.

In 2009, the Alexandrian made the firm case most of us are familiar with: don't prep plots.

A plot is the sequence of events in a story.

And the problem with trying to prep a plot for an RPG is that you’re attempting to pre-determine events that have not yet happened. Your gaming session is not a story — it is a happening. It is something about which stories can be told, but in the genesis of the moment it is not a tale being told. It is a fact that is transpiring.

In 2021, the Angry GM noted that

the story—the narrative—is an emergent thing. Your game’s going to have a narrative when it’s done. Like it or not. And people are going to interpret it based on how they think stories work. Players won’t experience the game as it was. They’ll experience it based on their warped perception of it. [...]

Once your game’s done, it’s going to have had a plot. It’s going to have followed a single sequence of events. There’s going to have been a climax. People—even you—will remember that s$&%. You can either use what power you have to steer the game toward the best narrative possible. One that does the things that thirty millenniums of storytelling have taught us make for good stories. Or you can hope and pray that the emergent narrative is accidentally a good one. It’s your call.

And just recently in 2024, Tom van Winkle wrote a superb article about the nuances of all this. If you only have time to read one blog post today, go read that one instead of this.

The assumption seems to be that to create a story, you need a plot in advance. It's important for the entire entry here to understand that that is not true. [...]

What all the "anti-story" gamers have in common today is that they allowed other people to define the term "story" for them as "pre-designed plot," even though that's not what story means.

He analyses "anti-story" gamers, who

have, apparently unwittingly, accepted the false idea that RPGs would need plots designed in advance to have stories in a story-telling game. Their objections are therefore confusing for many people who are aware that they experience stories arising spontaneously in RPGs.

The underlying claim is: playing a tabletop role-playing game is intrinsically a process of telling a story. And yes, in a technical sense, this is absolutely true. In a TTRPG session, real-world people are making deliberate decisions that lead to the details of a fictional universe and events within it being filled in. Therefore, definitionally speaking, the players and GM have collaborated to tell a story.

Tom van Winkle in particular discusses the literature, and what scholars of story and narrative have agreed on, and how that applies.

The problem, though, is that in everyday, natural English for the layperson, the phrase "tell a story" comes with all sorts of baggage that doesn't necessarily apply to TTRPGs.

The phrase is laden with subtle implications in common English:

1. Telling a story means the storytellers are doing so with the primary goal of producing a 'good story', or at least are telling a story intentionally
2. Telling a story means the storytellers have equal parts, or at least similar amounts of creative control
3. Telling a story means the process is done selectively, without random elements
4. Telling a story means that core elements of the narrative are planned (at least a little in advance), not suddenly discovered
5. Telling a story means that people are making choices in order to further the narrative in a satisfying way, which might not be the 'optimal' or even plausible choices for a particular character
6. Telling a story is unregulated by formal mechanics, or at least largely so

None of these are true of TTRPGs in general. They're not even all true simultaneously for most story-telling games!

So telling a novice who's interested in entering the hobby that a TTRPG is necessarily 'about telling a story' is frankly just misleading that person, unless you stop and break down why 'telling a story' doesn't mean most of what they think it means. (And it may lead to further misunderstandings, like when they hear the term 'player agency' and think it means creative agency over the fictive world, and so on...)

...We still have this phrase, 'story game'.

Two axes

When I think about 'story games' and TTRPGs, I envisage two parallel tracks running left to right.

On the first axis, we have 'rules' to the left and 'freeform' to the right. This axis mostly has to do with the GM, if the game has one.

• A game can be towards the left, 'rules', with lots of rules and processes for decision-making. A game here encourages system mastery for both the GM and the players. It doesn't necessarily imply more control or omniscience for the GM: the game could be heavily reliant on random tables, for instance, or there could be lots of codified rules for how the players get to mess around with the game world directly rather than through their characters.
• A game can be towards the right, 'freeform', taking an open-ended and freeform approach to actual gameplay. In this case the GM is making up a lot as they go along, relying more on the characteristics of the world that have so far been determined than on rules and frameworks provided by the game.
• A game can be somewhere in the middle, with moderate rule support.

On the second axis, we have 'decisions' to the left and 'insertions' to the right. This axis has more to do with the players than the GM.

• A game can be towards the left, 'decisions', where players control their character's mind – the decisions they make – and nothing else. The GM determines what is in the world, controls events and NPCs, and adjudicates PC decision outcomes. (But even the most leftward TTRPGs usually have a carve-out here for character generation, where players get to pick things about their character which the character realistically couldn't have chosen within the world.)
• A game can be towards the right, 'insertions', where players have power over aspects of the game that isn't through their characters. In some situations characters can create, choose, affect, invent, or veto, but in games like this the players have an ability to create, choose, affect, invent, or veto.
• A game can be somewhere in the middle, mixing approaches.
  

The reason I picture the tracks as parallel is that many games, if you map them onto the axes, make a line that's close to vertical. That is,

• Most, though not all, 'traddish TTRPGs' (games with few to no overt player-available story-telling elements) are also mechanically rigorous. They are towards the left on both axes. In general, the GM has lots of power and mechanical support, and the players will know that most things they choose to do will be covered by rules. Whether those rules are good or bad, players may be uneasy with situations where the GM has to go with their gut over having an 'official' answer. It will be unusual (troubling, even) for a GM to invite a player to fill in gaps in the world or determine action outcomes.
• Most, though not all, 'collaborative story-telling games' (in which players are permitted to insert content directly into the world rather than being restricted to attempts to alter the world with their characters) have relatively few rules and processes. They are towards the right on both axes. In general, there isn't anywhere near as much of an authority disparity between the GM and the players, or there might not be a GM at all. The game rules tend to be 'softer' or 'fuzzier', the players are given more overt license to override or ignore them, and the mechanical framework is usually designed more towards 'crafting a satisfying narrative' than towards 'playing a skill-requiring game' or 'simulating a fictional world'.

I'm personally most interested in the leftward side of both scales. I've designed at least one little game that wanders around the middle of the axes, and have a few others sitting on the project pile, but as a designer and player I mostly lean left, towards 'traddish TTRPGs'.

But I'm also interested, from a design point of view, in the rare oddballs. The really open-ended, rules-light TTRPGs with high GM authority, perhaps like this idea. And the highly constrained, rules-heavy, almost formulaic, versions of story-telling games, with no referee.

 

Ways of playing

We can map the way people actually play games onto these two tracks, too. As opposed to how games were designed to be played.

The big example here is (as it often is) D&D 5e. The system is to the centre-left on the rules/freeform track: it's crunchy and extensive compared to a lot of indie games, but not as much as previous editions of the same game, or GURPS, or whatever). And it's quite far left on the decisions/insertions track: the official rules only contain (outside character generation) a few optional and tacked-on suggestions about direct player control over the world.

Interestingly, this isn't always borne out in play. It's possibly due to the rise of people entering the hobby via watching televised 'actual play' by actors, or possibly a trend that's always been there but is just more visible now. A lot of people these days play D&D as if it was further right on the decisions/insertions scale. Players in a lot of games end up with a certain amount of creative control over the world.

To put it in fuzzier terms, they inject elements of a collaborative story-telling exercise into a relatively traddish TTRPG when they play. And that's fine. I suspect they might be better served by playing an actual story-telling game with design elements supporting that, but ultimately, let people have their own fun.

To sum up

This is just how I think about things. It's not a formal system or a position on how things ought to be done.

I think it's fair to say that all TTRPG involves the creation of story – and again, go read van Winkle's article for a really well-considered take – and that some people have the creation of that story as a more primary concern than others. I think that the phrase 'telling a story' has misleading implications when it comes to TTRPGs, which is why I personally avoid saying that TTRPGs deeply involve it, even if (getting into the weeds) it's true. And I find it interesting that when it comes to story, we can try to pin games onto two axes which quite often correlate with one another.

The 'Four Humors' Theory of TTRPG Players

The Angry GM recently published an article about the distinct ways in which players play RPGs, with the specific example of D&D.

He mentioned in particular the rise, or increase, of an attitude towards the game as an intrinsically theatrical activity, adding that to the old GNS classification:

Some players see D&D as a Game. They think it’s about overcoming obstacles, winning adventures, accomplishing goals, and advancing characters.

Some players see D&D as a Narrative. They think it’s about telling a story of fantasy adventure and they want to share the most interesting story possible.

Some players see D&D as a Simulation. They think it’s about entering a world as a character and dealing with it as it is. Whatever happens happens.

And now, some players see D&D as a Performance. They think it’s about putting on a show. They have a role to play and they’ve got to play it to the hilt. In the most interesting and engaging way possible.

It's a good article, worth checking out. While reading, it occurred to me that we could treat GNS Theory as just a way to describe some of the emergent properties of a simple pair of dichotomies, along the lines of another well-respected and evidence-based organisational schema, that of the humoral theory of human health.

Take all of this very seriously please.

A simple pair of dichotomies

Somebody is playing a tabletop role-playing game. They are engaging with an activity, making decisions, thinking about what they are doing. The act of play is, ultimately, a realisation of the act of thought. What is the player spending most of their brainpower thinking about?

There are two dichotomies here.

1. Is the player thinking mostly about their real world environment (the people they are playing with, the game rules, the miniatures, the side conversations)? Or are they thinking mostly about the fictive setting? Or something in between?
2. Is the player thinking mostly in an inward-looking away, about themself and what is theirs or squarely within their purview? Or are they thinking mostly in an outward-looking way, about things around them and outside of their control? Or something in between?

We can put this on a pair of axes, as in this professionally-drawn image.

And now a picture emerges, of what GNS+P is describing, but defined in terms of these innate characteristics of human thought: wet, dry, hot, and cold fictive, reality, inward-looking, and outward-looking.

 

These of course are what modern medicine ludology teaches us are the fundamental building blocks of blood, phlegm, yellow bile, and black bile different ways to engage with a role-playing game.

GNS+P Theory doesn't map precisely to the four resulting quadrants, of course. There's some interplay and outliers. We have debatable blobs, not square pigeonholes.

Let's break this exciting new bit of science down.

Yellow bile 'TTRPG as Game'

Temperament: choleric (ambitious, driven, aggressive)

What is a Gamist player thinking about?

Their thoughts tend to be INWARD-LOOKING: upon their own (and usually necessarily, the group's) ability to win; to this end they channel attention into their own player skill, their own knowledge of the rules, and making decisions about their own character's capabilities.

Their thoughts typically tend towards REALITY: the nature of the game and how to win it, the rules and mechanics, win and lose conditions, and how to get other players to go along with their plans. But they do still focus some of their attention on parts of the fictive environment: the obstacles to winning that their character encounters, and the mechanical ways they can get advantages with their character (in particular, advancement). The yellow blob in the diagram therefore stretches away towards the left.

To put it another way, Gamists have an excess of yellow bile.

Black bile 'TTRPG as Narrative'

Temperament: melancholic (sad, delusional, deep in fantasy)

What is a Narrativist player thinking about?

Their thoughts tend to be OUTWARD-LOOKING: towards the group having a satisfying play experience and how the game's play will become a story after the fact. At times, they may care more about their own enjoyment of the game-as-story, thus leaning towards the inwards-looking position.

Their thoughts tend towards REALITY: making decisions for the good of the players' play experience, the emergent story, and what will be 'interesting'. If they care about the game rules, it is in case those become an impediment to higher-order goals. (The Narrativist's thoughts may also drive towards a story with verisimilitude and/or immersion, creeping towards engagement with the fictive environment.)

To put it terms we can all understand, Narrativists may have an excess of black bile.

Phlegm 'TTRPG as Simulation'

Temperament: phlegmatic (calm, reserved, analytical)

What is a Simulationist player thinking about?

Their thoughts tend to be OUTWARD-LOOKING: at the world around their character, and at things they don't already know. They want to discover things about the fictive world and see how they mesh together, and inwards things like their knowledge of the rules and their character are mostly ancillary to that. But the act of engagement still requires making plausible decisions about the character, and the Simulationist does typically spend some attention on their character's place within the world (thus, the blue blob on the chart reaches upward).

The Simulationist's thoughts are almost strictly FICTIVE: They care most about the world they are playing in. The game's rules are meant to simulate a world, the GM is the arbiter of those rules, and the outcome of those rules is to be accepted and examined, never challenged or overruled for the sake of higher-order player/GM goals.

So of course, Simulationists have an excess of phlegm.

Blood 'TTRPG as Performance'

Temperament: sanguine (passionate, mirthful, self-indulgent)

The Angry GM suggested 'Performance' as a new and increasingly popular playstyle category, characterised by (amongst other things) creating

characters with clearly defined motivations, personalities, beliefs, goals, and fears right from the get-go. Their character’s character doesn’t emerge from gameplay and it doesn’t change as the game plays out.

What is a Performer player thinking about?

 

Their thoughts are ultimately INWARD-LOOKING: Most of all, they care about their own character, deciding its intrinsic properties, and then getting across those preselected properties.

Their thoughts are largely on the FICTIVE: How to affect the fictive world by determining a tiny part of it; looking for opportunities in the world to accomplish their secondary goal of expression. External things like 'the game rules' and 'the other players' don't impact this very much for trad RPGs.

Or as we normal, modern people would definitely say in normal, modern terms, "Performers have an excess of blood".

I put something weird on that chart, though.

 Entertainer island

I think there's a little outcrop of the Performative playstyle occupying the diametrically opposite position.

A player can start with a solid foundation of hot, wet inward, fictive humors. That is, they have a firmly imagined static character with predetermined characteristics and sometimes even an intention to pursue a 'character arc', whatever on earth that means when you're (a) playing a game that (b) involves, at its most fundamental level, making decisions about situations that you have no foreknowledge of.

But with those humors as a firm foundation, most of the player's attention is freed up to focus on the actual performative act, which is ultimately outward-looking in reality (given that it is a process of interaction with an audience). With this inversion of the expression of their bodily gameplay humors, the Performer is now thinking primarily about the theatrical entertainment of the other players / the GM / their Twitch audience / passersby in the real world. When taken to excess, to the game's detriment.

Balancing the humors

So it seems likely that people can have a predisposition towards certain humors, and we might or might not regard this as an imbalance of humors.

What are the implications? Let's sketch just a few, presupposing (naturally) that two thousand years of high-quality humoral medicine can be extended to the modern environment of the game table.

• A player who is desirous of a more neutral or 'balanced' playstyle may examine the degrees to which they are inward- or outward-looking, in the fictive world or reality, and deliberately change their behaviours accordingly.
• If the players and the GM at a table all have a similar humoral temperament, then we would expect their games to run well if they indulge in that temperament, although players of a different style would scarcely recognise the game they were playing!
• In rare circumstances where the group dynamics are just right, players of completely opposite humoral temperaments may 'cancel out' at the table due to the ways they use and express those temperaments, resulting in (against the odds) a harmonious, near-maximally-enjoyable game.
• Obviously, players may be able to select the right foods and drinks to modulate any excess of humors. By careful review of TTRPG play I have discerned some of the modern-day nostrums: pizza (hot + dry); coffee (hot + wet); soda (cold + wet); chips (cold + dry).

Science marches ever onwards.

Improving Swarms for D&D 5e

As part of a creature project, I'm taking a look at improving the 'swarm of creatures' mechanics in Fifth Edition Dungeons & Dragons.

The problem is one of verisimilitude balanced against the quirks of the 5e combat system, and a desire to make swarms uniquely vulnerable to the things they should be vulnerable to while dangerous if you're fighting them without those things. Only fun solutions are allowed!

What are the essential characteristics of a swarm?

A swarm of creatures is a notable opponent because each of its diminutive components is, almost by definition, (a) mostly harmless and (b) trivially defeated. Only en mass are they dangerous and require effort to destroy.

Characteristic 1: Weapons shouldn't matter, but strategies might

A lot of TTRPGs massively undersell the killing power of a bladed weapon compared to a blunt strike in almost all situations. One exception is fighting a swarm.

Typical swarms are made from creatures so small that the more effective approach to fighting them would be to hit them with something broad, in order to crush as many as possible at a time. If you have six seconds to spend and you're fighting a homogenous area of a thousand flying bugs in the space in front and to the sides of you, then we can imagine shooting an arrow might kill ~0.5% of the swarm if you're lucky, swinging a razor-sharp blade could kill ~1%, and swinging an ordinary dining chair could kill ~5%. If they're ground-based creatures, stamping on them or rolling around on them are likely to be far better strategies than swinging a weapon.

So if a player character wants to fight a swarm, then we only have to consider whether they're in melee or shooting at range; to simplify we might generously allow melee weapons to be as effective as unarmed or improvised attacks, so that players don't feel silly just having their characters stamp up and down every turn. Ranged attacks should be much worse, though.

As the swarm's component creatures become larger and more dangerous, the martial skill and physical power of the person fighting them might begin to matter. But for heroic fantasy like D&D 5e, I think we would generalise 'fighting skill and power' to 'character level': CR 18 swarms should be made up of individual creatures of a kind that any 18th level character can destroy trivially and in quick succession in a fight; the same creatures might individually pose a small challenge to 1st level characters, but the 5e design parameter of bounded accuracy limits the degree to which this can be true.

Characteristic 2: Big, flowing, and diffuse

A swarm is made of tiny little freely-moving elements, and therefore behaves something like a fluid: it can get into small spaces, surround creatures, and so on. Healing, strengthening, or weakening an individual creature shouldn't have any impact on the swarm.

How large and dense is a swarm of animals? Dense enough to see it clearly, as a bare minimum. In practise, I think it would be more viscerally horrible if it's really dense, with creatures crawling all over each other.

• As a bare minimum density, I think we should imagine at least one flying bug in every 4×4×4 cm cube. Then we would have more than 15,500 bugs in a cubic metre, which is about the space you could swing a chair through.
• Alternatively, we can imagine one larger flying creature like a small bird or bat in each 25×25×25 cm cube. There would be 64 such creatures in a cubic metre.
• For a dense mass of beetles crawling over each other on the ground, we need at least one beetle every 2×2 cm. Then we would have at least 2500 bugs in a square metre.
• Likewise, for rats or tarantulas or other larger creatures, I would expect at least one animal every 10×10 cm, for at least a hundred critters in a square metre.

I envisage swarms as having a big dangerous centre part and a much less dense 'periphery' of creatures that continually enter and leave the centre, but are mostly harmless. We can think of the swarm as being all the creature in it, but it's mostly the ones in the core attacking and grouped up in ways that makes it possible to kill them at any reasonable rate.

When I sit back and imagine a swarm, the whole thing must be at least five square metres on the ground, or at least ten cubic metres if flying – preferably much larger. Anything smaller seems pitifully small. Of that, the dangerous core is maybe a fifth to a half of its area (or volume).

This back-of-the-napkin maths implies you're fighting ~150,000 wasps or ~640 bats or ~12,500 centipedes or ~500 rats.

It's a good place for a reality check. Here on Earth...

• A bat colony has several hundred bats
• A honeybee swarm has 5,000 to 40,000 bees
• A murmuration of starlings has 5,000 to 50,000 birds
• Army ants form swarms or columns of between 200,000 and 20 million ants
• A desert locust swarm is pretty diffuse, with 'only' 50 to 70 locusts per square metre but covering hundreds of square kilometres in total

Real-world swarms are probably much less dense than the ravenous fantasy swarms we imagine. But whether they are or aren't, the numbers don't seem off by multiple orders of magnitude.

Characteristic 3: Time required for attrition

Assuming a swarm is only dangerous because it is a very large group of marginally dangerous creatures, it should take a long time to destroy in melee combat.

If you have a six-second turn, maybe you can completely destroy half a cubic metre of flying creatures by swinging a chair through them, or annihilate a quarter of a square metre of ground-based creatures by jumping up and down on them. I actually think that's generous even for characters in a heroic fantasy genre.

Then with the parameters we've established it should take a minimum of 20 character turns to kill a swarm in melee (as a level-appropriate moderate challenge, 5 rounds for a party of four characters). For the kind of bigger, more dense swarms that we'd prefer, it 'should' take much longer to destroy a swarm without something that takes out a huge chunk out of it at once (see Characteristic 8, below).

Characteristic 4: Easy to hit

It shouldn't really be possible to try to harm a swarm and fail.

It doesn't matter if you're swinging an axe at bats or stamping on centipedes, we're talking about a packed area of low-intelligence creatures with little situational awareness, a flocking instinct, possibly driven into a frenzy, and surrounding their enemies on all sides. They're not getting out of the way.

In D&D terms, it shouldn't require an attack roll to take a swing at a swarm, and they should at least have disadvantage on most saving throws.

Characteristic 5: Dangerous in numbers

A swarm should be able to attack almost indiscriminately. Ideally, anything inside it should get hit; we might limit this to a 'multiattack' to make it faster.

As a swarm gets smaller, the amount of harm it can do should fall.

• If a swarm takes up less volume as it suffers attrition, keeping the same density, then the number of targets it can attack simultaneously goes down.
• If a swarm takes up the same volume as it is attrited, its density falling instead, then the amount of harm it does to each target goes down.
• If a swarm's volume and density are both reduced, then you could justify either option.

Characteristic 6: A reason for cohesion

The main reason (on Earth) that animals band together in swarm is to defend against predators. There's both 'safety in numbers' in the sense of an individual becoming statistically less likely to be predated, but also, the movements of the group can confuse or in some cases intimidate predators. There are rare exceptions, like army ants, which seem dangerous because they group up to hunt. I think it's safe to say that fantasy swarms are more like the latter.

A swarm sticks together, but clearly doesn't have to – its individual components aren't physically connected. It's a matter of taste as to whether parts that get separated can rejoin the swarm to continue acting in unison; I would rule that they can.

A swarm also can't react quickly or precisely or perform complex tasks, and its individual creatures attack pretty much blindly. In D&D terms, I wouldn't let it make attacks of opportunity: if you're in the swarm, you're being constantly bitten (with lots of miniscule harm adding up to HP worth of damage), and if you're not, you're not. If anything, a creature should suffer less harm for leaving the swarm's area sooner, but that's difficult to reconcile with D&D mechanics for only a small payoff.

Something is motivating the swarm to act as if it were a single creature. Maybe...

• A true hive-mind has emerged from the swarm, or
  
• A foreign overmind is controlling it, or
• There's some spell or fantastical natural force in play, or
• The individual creatures have a psychic bond

It seems likely that forcibly disrupting the swarm's cohesion would be particularly harmful to it. Depending on the nature of the outside motivator, the swarm might be particularly vulnerable to psychic damage, or it might shrug it off completely. I personally favour the former.

Things that affect a creature's physical body (conditions like paralysis, poison, and petrification) probably only affect one or a small number of bodies within the swarm, for negligible effect. We could model this as the minimum unit of harm in D&D, 1 hit point loss.

Characteristic 7: Extreme susceptibility to area damage

A little creatures which is individually below notice can be killed instantly by almost any attack. Some attacks are indiscriminate, hitting everything in an area. Even a really dense swarm probably has some room between creatures for such effects to penetrate.

It follows that effects like fire, force waves, toxic or corrosive clouds, rough weather, etc should do massive damage to a swarm. A swarm caught entirely within such an effect might struggle to survive at all! In D&D terms, it is not sufficient to model these kinds of things as different damage types. "6d6 poison damage to any creature in the cloud of poison" should kill a swarm; "6d6 poison damage to a creature hit by the arrow" should barely inconvenience it.

So does 5e capture these essential characteristics?

Not sufficiently, although the writers made an effort in that direction. Consider the six swarms in the D&D 5e System Reference Document (SRD).

Characteristic 1: Weapons shouldn't matter, but strategies might
Characteristic 7: Extreme susceptibility to area damage

✅ SRD swarms do resist the kind of damage usually associated with weapons.

❌ ...But on average a longbow will still do more damage than a club, which will still do more damage than fists and feet, i.e., the opposite of what we'd expect.

❌ More martially skilled characters deal substantially more damage than less martially skilled characters when smacking at a SRD swarm in their space. We wouldn't expect this, especially at low levels (where flailing around with something broad and heavy to crush dozens or hundreds will always be a good strategy).

✅ SRD swarms don't resist any of the classic area spell damage types, making damage spells a good choice against them.

❌ ...But nor do they resist single-target spell damage. And if an area spell does deal weapon-like damage, an SRD swarm takes half damage from it (which it really shouldn't).

Characteristic 2: Big, flowing, and diffuse

✅ The SRD absolutely nails the 'Swarm' trait. I don't think it needs to change at all.

❌ The SRD swarms are Medium in size. Even if we imagine this as the most dense centre part of the swarm, this still feels too small for any but the absolute weakest of swarms (admittedly all the SRD swarms have a low CR).

Characteristic 3: Time required for attrition
Characteristic 4: Easy to hit 

❌ The SRD swarms have similar HP to comparable SRD monsters, but their resistances effectively double it. However, their HP is unremarkable per the DMG monster creation rules, even factoring in resistances. They should be harder to destroy using single-target attacks.

❌ Likewise, their AC is typical for their Challenge Ratings, so a fair number of attacks (at least a third) will just miss them, even though that makes no sense.

Characteristic 5: Dangerous in numbers

✅ The SRD swarms lose half their damage output when reduced below half health, calling to mind 4e's Bloodied condition. I think only changing it one time at half HP is reasonable. It means a swarm's current number of component creatures is directly reflected in its damage, but doesn't create more mental load for the GM with a more complex way of doing that.

❌ To have a SRD swarm attack lots of creatures at once, the GM has to fake it by using multiple swarms together. Really a swarm should be larger and dangerous to anything inside it. Think about the verb form of the word 'swarm'.

Characteristic 6: A reason for cohesion

✅ SRD swarms are immune to most conditions, as we would expect. There are some edge cases like blinded, deafened, and poisoned which the designers let affect them (the first one possibly due to how light and obscuration are defined).

❌ Swarms probably shouldn't distinguish 'attacks of opportunity' from just continually attacking. Also, SRD swarms are affected normally by psychic damage despite being made up of hundreds or thousands of minds.

 

Evolution of solutions

When I initially wrote a more plausible 5e swarm based on that analysis, playtesting showed that I'd gone too far in the direction of realism. The fun was being lost.

Swarms could be blown up by quite low level spells, and a combat run strictly with melee attacks was a grind: arduous but not very dangerous.

In 5e, combats that go beyond three rounds are uncommon, ones that go beyond four rounds are rare, and so on. When the four PCs took six rounds to destroy just one of three lower-CR swarms I knew I'd miscalculated. Out of a sense of morbid curiosity, I gamed out eleven(!) more rounds before the fight was over.

Fortunately, I found some tweaks that solved this:

• Harmful area effects should do a lot of extra damage to a swarm (especially if all of it is in the area) without completely trivialising an encounter.
  
• A melee attacker can't waste every turn inflicting really trivial damage on a swarm. (It's fine to make single-target ranged attacks useless, because 5e has no strictly ranged glass cannons, and the character types that come near that all have access to area damage.)
• Swarms can have middling damage that drops when they hit half HP, as long as they deal some small damage on a miss, for an effect akin to a damaging aura.

With these changes, fights using just one or two spells last a good number of rounds. Multiple spellcasters willing to expend their best spell slot in their first turn can destroy swarms easily if they want to pay that cost. Fights with mostly mundane combat go longer than normal, but not outrageously so. More importantly, damage-on-miss makes swarms feel dangerous; between that and 'you can't fail to hit a swarm', the two sides are continually wearing each other down.

The final stat block isn't as completely verisimilitudinous as my initial vision, but overall it's a fun challenge to fight with a reward for thoughtful play, which is important. And the other changes to the SRD 'swarm' type monster implied by this article's analysis all worked well, so they'll be included in the final write-up.

Of course, to make this all hang together, the modified swarm needs a large number of traits (as well as some slightly unusual stat block entries like damage-on-miss and broad spectrum condition immunities). That's a no-no; it imposes substantial cognitive load on the GM, especially a GM trying to use it on the fly.

The next step, then, will be a challenge: Edit the text into a shorter form which presents the most important information in a way that's easy to find while keeping the writing crisp and clear. But isn't that always the case?

Narration and creative control of the world: A cautionary follow-up

The other day I talked about the advice that 'the extremely unlikely success or failure of an action means something about the world must be different'. I gave some reasons why this can make things harder for the GM and less satisfying for the players. I emphasized how important it is for the GM to be careful with the 'world-state' (the body of information about the fictional environment).

Van Winkle's article which prompted my writing also touched on a related issue with either the GM or a player establishing new facts about the world-state:

What if a player says, after landing a blow, "My character slashes the orc's forehead, causing blood to pour down his face"? Sounds cool, right? So you, the GM, nod your head and continue the excitement. But then the next round the player says, "Hey, shouldn't the orc attack me with some penalty because there's blood in his eyes?"

Van Winkle endorses a solution to this from ktrey: The GM should say "Let's see what happens when the orc attacks you. If he misses, we can say that the blood in his eyes played a part in his failure".

But this advice boils down to: ignore what has actually been established to be true in the game world. We could rephrase it as "the GM/player can establish something that definitely changed in the world but treat it as not actually having changed unless future dice rolls support it having changed". There's a strange implication that the GM can and ought to say "actually, this fact about the world can't have consequences", because the rules (mere tools the GM uses to interact with the world-state) are somehow more important than the underlying validity of the world-state!

I think it's fair to say this is contrary to the GM's fundamental role. Nothing that the GM (or player, in this case, where a player has been handed creative control) establishes about the world should be mere paper-thin "set dressing" which looks nice but which the characters can poke a finger through. Once something is established as true, it becomes usable by any PCs or NPCs or natural forces. Any knock-on effects must actually happen.

The combat example is a minor one, but even it has repercussions. What if the game has a spell that works on bleeding creatures? What penalties does the orc have? Can it benefit from attacks of opportunity / combat advantage / mechanic X? What about if it wasn't an orc, but a creature has corrosive blood, or a gaze attack? How long does it take to clean blood out of your eyes, for bleeding to stop, for the wound to heal?

Just take a more systematic approach

This pitfall isn't hard to avoid.

To begin with, the GM should keep careful track of their game's world-state. They should understand how the game's systems work and what the outcomes of those systems mean for the world-state.

They should only narrate beyond the extent of those outcomes with care, especially in situations like combat when PC life is on the line and players will feel cheated, not just disappointed, if a new piece of the world's reality gets established and then its consequences immediately ignored.

They should be extremely cautious in allowing player involvement in settling what's true in the world (beyond what the rules of the system allow).

For example, in a game that has hit points, the GM must understand what hit points are, what they represent in the world, and how systems interact with them. In some games, if a creature is knocked down to 5 hit points, you can narrate them being stabbed through the thigh. In some (most, frankly), you can only narrate the near-miss, the ensuing exhaustion, the draining of luck, etc. Being stabbed through the thigh, once established, is something that actually happened in the world. If the unfortunate victim is a baseline human and the world is anything like ours, then there are consequences: the victim almost certainly falls down, they may pass out from pain, they may bleed out slowly or in a matter of seconds, and their ability to attack, defend, climb a rope, and so on is drastically reduced. Is that a ton of extra GM work and possible interactions with parts of the rule set? Yes. Is it absolutely necessary? Yes. This is why games with hit points usually don't make them meat points.

But understanding the game's systems means it's easy to narrate. You know that a sword through the thigh is a grievous wound, so you know if the rules say a creature isn't mechanically penalised by injury, then you can't narrate any non-lethal attack as a stabbing. You know that kind of wound can't be slept off, so you know if the rules say a creature can get all their hit points back overnight, then when a creature is knocked to 0 hit points you can't say they were stabbed. In fact, in such a system, you can never narrate any wounds unless the creature also died – unless (a) you are just playing the game for laughs or (b) there's something in your world's metaphysics that makes wounds of all kinds short of death close overnight, with all the ensuing consequences to society.

All of this applies to the example with the orc getting blood in his eyes, of course. There the possible solutions are:

1. Don't let players narrate their attacks and don't ever say anything about the world that isn't true and therefore supported by game mechanics, or
2. Change to a system which actually mechanically supports whatever a player establishes as true about the world via the creative control they have been given, or
3. Switch to free-form role-playing or storytelling.

If you study your game system and can't help but conclude that hit points are neither meat points nor luck points but also both those things, then it's only a matter of time until you have to contend with this issue. There is no solution other than fixing the game system or changing to one that's coherent. Even if you try to sidestep by deliberately narrating combat really vaguely, eventually a player will notice "I suffered wounds so terrible that I failed all but the final saving throw against instant death, and now it is ten hours later and those wounds have literally disappeared without any magic or miracle involved".

Finding a conclusion in all this

Ultimately this is part of a broader point: narrating an action's outcome isn't just a fun little bit of flavour, like it would be if you were telling a story. It's a major part of the role-playing game loop. If a player isn't treating the GM's description of the world as something their character can really interact with, then they might as well not be playing at all! The human interactivity, the "role-playing" part of a role-playing game, is the only advantage it has over other forms of media!

I think it's difficult to create a good role-playing game with collaborative story-telling elements because of issues like these. Role-playing is all about seeing what's in the game-world around your character and treating it as a real situation. When an RPG has story-telling elements, a player has to be both reactive/interactive (role-playing) and creative/declarative (story-telling): the same human brain making semi-godlike decisions about what is true in a world is also trying to make reasonable decisions on behalf of one person with certain goals and traits within that world. Disentangling those roles is difficult. It can cause unpleasant cognitive dissonance.

Think about the consequences before giving players additional creative control in a role-playing game.

"Why Not To Play This Way" – An unlikely success or failure means you learn something new

I'd like to push back on something.

For a very long time – and most recently in an article by Tom Van Winkle last month – lots of GMs and RPG writers have recommended that when a player character's attempt at a task results in a very unlikely success or failure, it should be treated as 'discovering something new about' (i.e., changing the GM's mental model of) the game world.

When the task resolution mechanic reports a surprising success, the task was actually much easier than it seemed (for some reason that the GM comes up with). A surprising failure likewise means something in the world made the task much more difficult.

The lichvanwinkle article aptly sums the idea up as 'The "failure" [or success] with the dice gives an opportunity to define the world in ways that are not just about the character.'

(I like pretty much all of Van Winkle's writing; this is just one of those cases where you nod along appreciatively in silence for years and then only bother to pipe up when you disagree.)

To extend two of his examples:

Surprisingly, the accomplished locksmith fails to pick the simple lock on the store. So...

•  The lock mechanism must be broken, rusted solid, or jammed shut
•  The lock does open, but it turns out the door is stuck or barred
•  As the locksmith is just starting, she hears guards approaching, so can't complete the covert work

Surprisingly, the puny wizard manages to bend the metal bars in the prison window. So...

•  The metal bars are rusted completely through, and give way.
•  The bars were only an illusion all along!
•  The bars are solid, but the wizard finds the stone is old and degraded – he can just chip it away and pull the bars out.

(Note that the probability and mechanic for the unexpected success or failure doesn't matter. It could be 'natural 1 = automatic failure' or 'exploding dice leads to preposterous successes' or 'disregarding the rules, this particular GM always tweaks the Target Number so any PC can always succeed or fail'.)

Now, I usually like the idea of 'overloading' dice with extra meaning. Reducing the number of dice rolls is almost always a great design goal.

So why not here?

This is one of those ideas that's really mostly innocuous, but is so widespread that I want to push back on it for fear that it will become the status quo. "An unlikely dice result changes what's in the world" is definitely not a panacea solution, and you may find that it can create some problems of its own.

And what it boils down is that overloading the action-resolution dice in this particular way thwarts a really fundamental, really effective TTRPG loop:

(i) The GM determines, by whatever means, what is true in the world: its world-state.
(ii) Whenever a player character's passive senses would tell them something remotely meaningful about the world-state, the GM gives the player that information.
(iii) A player can have their PC investigate (interact with) their environment to get further information about the world-state.
(iv) The difficulty of a task is determined entirely by the world-state (including as it does all PCs and NPCs, secrets and fixtures, physics and metaphysical forces, gods and demons, self-inserts and metanarrative powers, etc).
(v) Players can estimate the difficulty of a task by how much they have learned via (ii) and (iii); they can attempt to reduce risk by doing more of (iii) and the GM can make the game require less back-and-forth by doing more of (ii).
(vi) Players have a single 'game action': declare their intention to attempt some task which they describe.
(vii) When a player uses their game action, the GM: first checks whether anything should happen before or simultaneously by consulting the world-state and/or the other players, then checks they and the player are envisaging the same thing with regard to the player action, then considers difficulty via (iv), then (only if the task could either succeed or fail) determines which game mechanics to use to resolve it, then resolves it, then narrates the attempt and its outcome via (ii).

Notice how many of those steps are disrupted if the action-resolution dice can make a retroactive change to the world!

An idea in conflict with TTRPG best practises

More specifically, the idea that 'unlikely dice results change the world' conflicts with six important bits of TTRPG advice which are ubiquitous in modern discourse:

1. Don't make the characters do anything the player didn't try (outside of minor, obvious, safe, unstated things; always assume a character is competent). An 'unlikely dice result' should not change what a character was doing.

• The outcome of the action "I try to pick the lock" can't be "you were so overconfident that you tried to pick the simple lock without torsioning the keyway, so you failed". The player didn't attempt that and the character shouldn't be incompetent.
• Likewise the outcome of the action "I try to bend the prison bars" can't be "you chip away a bunch of what turns out to be ancient, crumbling stone and pull the bars out instead". We can easily imagine cases where the player would decide not to take the alternate option. (And if you stop and give them the "new information", they're now deciding based on information that was first created and then discovered by the outcome of a "use crowbar to bend metal bars" resolution mechanism which didn't do anything. Weird.)
• So this substantially constrains your options for retrofitting the world to an unlikely dice result.
  

2. Don't roll if the outcome doesn't matter. If the outcome of the roll is so surprising that it's incompatible with your vision of the world as GM, to the extent that you have to change your mental model of the world-state in order to keep it consistent, it means you should never have had the player roll in the first place! You should have just said "yes" or "no"! (If the game procedures told you to, it may also mean that those procedures are bad, but that's a different matter). If an expert lockpicker encounters a simple (functioning) lock, and has time and tools and use of her hands, then anything short of a lightning bolt from above means that she opens the lock. There should not be a flat 5% chance that she just "can't". This is true in the real world and in any game that has verisimilitude.

(Rambling aside: Breaking a pick or torsion tool in the lock is rare, at least for someone who knows how to use them. Even if you ascribe that fact to the wonders of modern metallurgy, older locks tend to have larger keyways, so older lockpicks can probably afford to be thicker to make up for it. And most importantly, note that snapping a tool in a lock will almost certainly not jam the lock, preventing it being opened; a competent locksmith will remove the piece and continue to work.)

3. Decide and envisage the world-state before you determine how to adjudicate an action. Whether you need to use any given rule, or involve dice mechanics at all, depends on what is currently true in the world.

• Just because there's a rule for "falling object damage" doesn't mean you use it if the PCs manage to make the moon fall onto their planet.
• Just because there's a rule for "whether an attack hits" doesn't mean you use it when a PC attacks a paralysed target in an otherwise empty room.

Knowing what's true in the world makes the GM's second-most important job (adjudicating action outcomes) very simple. If you were envisaging the lock as functioning, then don't roll: the locksmith cannot fail. If you were envisaging the lock as rusted solid, then don't roll: the locksmith cannot succeed. If you were envisaging the lock as being complex, or being simple but someone has lovingly filled it with honey, then work out an appropriate Target Number and go to the dice. The 'unlikely dice result changes the world' idea makes all this impossible.

4. Decide and envisage the world-state plus the outcome of an action before you describe an event or outcome. If you already know what's true in the world, and you're only letting dice have a say when success or failure are both possible, then you can go from outcome to description immediately.

If an unlikely dice result can change the world, this is more difficult: You have to leave space for things that even you, the GM, don't know about the world. When the dice give a surprising success or failure, you have to – on the fly – determine a change which is compatible with what the player characters already know and also what the consequences of that change will be and more specifically whether this 'breaks' anything else in the world-state or in the current scenario. And then you have to narrate the action outcome while incorporating how the player characters discover the newly-created information.

5. Players are given key information up front, can get as much descriptive detail as they ask for, and can find out more relevant information by interacting with their environment. But, whoops, now an unlikely dice result can change the world retroactively, so what info can be given out?

The GM's single most important job is providing concise, accurate, memorable, relevant descriptive narrative. You just can't do this if you have to worry about 'saving room for' corner cases. If the prison bars are strong, rust-free, corporeal, and set firmly in strong rock "unless in the future I need them not to be, if the weakling wizard rolls high enough to bend them", you have a big problem which grows bigger with every interaction the player characters have with the bars.

At some point, as more and more 'explanations' of the surprising dice result are ruled out by what the players know about the world, you're left with just two options, both unsatisfying.

1. A preposterous deus ex machina ("a sudden earthquake rocks the room just as Feebliander begins to push the bar! The wall caves in!"). Even if you try to disguise its extempore nature, you're likely to get the problem of "We immediately waste a bunch of time and resources investigating what has caused this very unexpected thing which we take to be a big clue about other parts of the game world."
2. A fake or fiat perception failure ("I know I said Feebliander thought the bars looked new and sturdy when he spent ten minutes studying them carefully, but as he applies the crowbar, he finds them to give way in a spray of rust.") Many game systems have perception mechanics, including ones that are meant to be passively or automatically applied, in which case as soon as you make surprising dice results retrospectively change the world, you've immediately failed in your job of correctly and impartially applying the rules, by either not invoking them or not taking into account information about the world.

6. Don't let the players think you don't know what's true in the world. This is more important than it sounds, especially for invested, smart, skilled players.

But suppose they notice that a lot of the time when they roll really high or really low, there's a bit of a pause and then the GM tells them their action fails or succeeds because of something they hadn't already known about the world. They're going to lose faith that the GM is thinking about their own world. They'll (accurately) regard that as a troubling sign. They'll start thinking things like "the GM can't have known the lock was completely rusted, because then what would be in the point in making me roll?"

More pragmatically, they're going to start checking for all the little factors – is the lock old, is it rusted, is it gummed or jammed or filled with honey, is the door stuck, is anyone else around – BEFORE attempting to pick the lock, and then what are you going to do when the expert locksmith rolls a 1?

Okay yes but

Most of the rationale for 'unlikely dice result changes the world' was that the GM wants to be surprised sometimes. That's valid, right?

Yes, of course.

• You might be worried about fairness ("otherwise I'm just deciding by fiat that the lock is in good working order").
• You might be aware of your own cognitive blind spots ("otherwise I'll just end up using the normal defaults – normal locks, sturdy bars, guards who can't be bribed, trees with enough branches to climb").
• You might be using a system that's simpler than you'd prefer ("Am I really meant to just make up all the Target Numbers?").
• You might want the PCs to encounter more variety than you're capable of providing ("I'm always thinking about the next encounter, so I never remember to have a more mundane task throw an unexpected boon or bane").

So how can you attain this without overloading the task resolution dice?

I think the answer is: take a more rigorous approach, using a dedicated dice roll. You could do it any number of different ways; the key is to identify early on that there's something with various possible states/aspects/nuances which could change task difficulties, and having identified it, quickly make a dice roll to see what's true in your world. Once you have determined the world-state, all the rest of the GM's job is exactly as easy as it ever was.

The solution

It's one of those things you can learn to do without a formal process, just as an instinctive part of the regular rolling-dice-for-no-apparent reason that you ought to be doing anyway to keep the players alert.

But if I had to encode it as a specific procedure, I'd say:

• Pay attention to cases where you start describing something with an obvious PC interaction, especially where you're likely going to have to later improvise that task's difficulty.
• In such cases, first envisage how the world probably is – the default state of the lock given who bought it and used it and when; the condition of the bars given whether this is a gaol in use or a crumbling ruin; the loyalty of the guards given who hired them to do what; etc.
• Then just roll a dice to see whether it's actually better or worse overall for the PCs, with a 1 being "version of the worst case that's still completely plausible" and a maximum roll being "version of the best case that's still completely plausible", the gradations between being easily decided on the fly. Worse cases make for harder Target Numbers for most approaches to most tasks, and so on. A "middle" result (either of the two numbers around the dice's average, which a GM should know) means "don't change what you first envisioned at all".

Pros:

• This is fast, simple, and easy to remember.
• If you forget to roll, or only remember after the characters have started fiddling around with things, it doesn't matter. You don't need to remember every time, only sometimes.
  
• You can pick up whichever dice is closest and use it.
• Your expectations about the world as the GM often hold true, and things only deviate from your expectations as far as you'll let them.
  
• When you 'discover' something new about the world, you get that information before you go to action resolution, available to give to any player who has their character investigate how hard various approaches to the task are going to be.
• You never get weird action results that are borderline- or outright-incompatible with the world you envisioned.

I think this better serves the overall purpose of "freedom in narrating the events in a fantasy game without hesitation or squabbling over rules and rulings. Let the dice serve your narration without requiring your narration to mess with the rules further."

Cons:

• That pesky extra dice roll...