On Randomness

Much of TTRPG design is based around randomness. We roll for random stats, encounters, reactions, morale, searching, listening, attacking, doing damage, treasure, etc.

This article will attempt to untangle what the point is, when it is beneficial, and when it is not.

TL;DR

Randomness

• Allows for blameless drama generation and fictional buy-in.

• Helps us makes complex systems game-able.

• Describes a lot of information in a small amount of text.

• Gets us out of mental ruts, both during prep and play.

• Costs time, coherence, and quality.

• Can be edited down mercilessly. Remove weak entries, specify outcomes, and see if random tables can become shuffled lists or be replaced by their best result.

Blameless Drama Generation

I think this one is the simplest. Frequently, there are major emotional stakes on the line when the dice are rolled. Maybe an enemy is attacking a cherished PC on low health and the players really need the enemy to fail. Perhaps a daring PC is attempting to leap a bottomless pit.

When everyone involved knows the odds and the stakes, all that’s left is to roll the dice. The moments between when the dice are picked up and when they’re finished rolling are dramatic; anticipation is high.

Equally important are the social and emotional dynamics. If the GM declares that the skeleton attacks Daphne and she dies, folks could (rightfully) be upset about the GM’s decision. If the GM declares that the skeleton has a 55% chance to attack Daphne and everyone agrees this is reasonable, then when it hits and Daphne dies, then it feels like the dice’s fault.

This is the other side of the coin of something Vincent Baker (the designer of Apocalypse World) wrote circa 2003:

So you're sitting at the table and one player says, "[let's imagine that] an orc jumps out of the underbrush!"

What has to happen before the group agrees that, indeed, an orc jumps out of the underbrush?

1. Sometimes, not much at all. The right participant said it, at an appropriate moment, and everybody else just incorporates it smoothly into their imaginary picture of the situation. "An orc! Yikes! Battlestations!" This is how it usually is for participants with high ownership of whatever they're talking about: GMs describing the weather or the noncombat actions of NPCs, players saying what their characters are wearing or thinking.

2. Sometimes, a little bit more. "Really? An orc?" "Yeppers." "Huh, an orc. Well, okay." Sometimes the suggesting participant has to defend the suggestion: "Really, an orc this far into Elfland?" "Yeah, cuz this thing about her tribe..." "Okay, I guess that makes sense."

3. Sometimes, mechanics. "An orc? Only if you make your having-an-orc-show-up roll. Throw down!" "Rawk! 57!" "Dude, orc it is!" The thing to notice here is that the mechanics serve the exact same purpose as the explanation about this thing about her tribe in point 2, which is to establish your credibility wrt the orc in question.

4. And sometimes, lots of mechanics and negotiation. Debate the likelihood of a lone orc in the underbrush way out here, make a having-an-orc-show-up roll, a having-an-orc-hide-in-the-underbrush roll, a having-the-orc-jump-out roll, argue about the modifiers for each of the rolls, get into a philosophical thing about the rules' modeling of orc-jump-out likelihood... all to establish one little thing. Wave a stick in a game store and every game you knock of the shelves will have a combat system that works like this.

Modeling Complex Systems

A society of goblins living in a cave is complex. They each have their own wants, needs, and emotions. They travel, hunt, get into mischief, create “art”, release spores (or however they reproduce), etc. Whether or not some goblins (and how many) are going to come across the PCs is a result of their complex behavior. Whether or not the goblins (or the PCs) will be caught off guard is likewise complex. How happy the goblins are about seeing the PCs is, once again, complex.

Instead of trying to simulate the physics and complex emotional states of goblin society, we simplify all of this into random encounter rolls (1-in-6 chance every 10 minutes) and random reaction rolls.

In abstract, vastly complex systems produce random-looking results. Instead of modeling the systems, we can model their outputs to get something easy to play at the table.

Densely Encoding Information

“1d20 orcs” represents 20 possible different orc sizes; this is way more compact than saying:

1. 1 Orc

2. 2 Orcs

3. 3 Orcs

4. …

This is especially useful if a situation is likely to come up multiple times in roughly the same context. Variety is needed to preserve verisimilitude. If the players keep finding exactly 100 copper or fighting exactly 4 orcs, suspension of disbelief starts to become difficult.

Escaping Ruts

There’s a wonderful lesswrong article on divination that explains that things like tarot cards, horoscopes, or the I Ching are good for making sure that different sorts of perspectives and approaches (consult a friend, self reflection, etc) are applied to challenging problems.

This, in effect, makes sure we don’t get stuck in a mental rut where we’re only applying a subset of the tools available to us as we approach a particular problem (I know I’m guilty of not asking for help and performing lonely deep research to solve almost anything).

In a D&D context, it’s easy to get stuck in the same tropes when creating situations or adventures, and so “divination” like tables from the Tome of Adventure Design help a ton. It’s easy to get stuck in a rut where NPCs play to the same stereotypes, monsters are aggressive, fight to the death too often, etc. Being inspired helps us to increase conceptual density.

Consulting random oracles like charts and tables helps to decouple moods from decisions; ultimately getting us out of ruts.

Randomness Gone Wrong

Randomness, especially in adventure design, that doesn’t do one of the above makes the game worse. The costs of randomness are

• Time. It takes longer to roll a dice than to directly have a result.

• Coherence. A specific result can be intentionally crafted to a situation; random results can seem arbitrary.

• Quality. A specific result can be given more love and evocation.

We pay these costs for the above benefits. In situations where the benefits don’t apply, we’re left with just costs. For example, consider the random encounter chart from The Black Wyrm of Brandonsford

We roll a d12 to get an encounter; is this better than picking an encounter from that list (or going top to bottom in order?). Rolling to see what is encountered isn’t player facing, so there’s no drama. So long as we’re not picking the same encounter twice it isn’t helping to simulate a complex system. It could help us escape a rut if the reality is that we’d pick less-fun encounters than the dice would. I tend to think for a list like this, it’s better to pick the most fun/interesting encounters and play them immediately.

Further, say that we rolled #5: 1d10 Stirges. We could roll a 1 here and then the text is nonsensical (there would be no other stirge to squabble with and there would be no half-joining later dynamicism).

Does rolling for Stirges create a better at-the-table experience than specifying that there are 6 Stirges in the encounter description? If we roll #5 again, are we actually going to repeat the Stirge encounter, or is it better for at-the-table fun to add variety?

Or, consider random loot:

This is a lot of bad all at once. It’s unclear whether multiple players can search at the same time (and if that speeds the process up), and it’s unclear what happens when you roll the same result multiple times. Say that a player rolls a 4 (1d4 gems), and receives 3 gems (which also have their own random value). Then they roll a 4 again. I assume that this is not an infinite money generator, and so on a 4 nothing happens. In order to find everything, they eventually need to roll all of the d6 numbers at least once, the #2 7 times, and #5 5 times.

That’s a lot of boring rolling.

Why not specify 13g instead of 2d20? Why not specify 20g citrine, 50g amethyst, 100g ruby instead of 1d4 gems (10-100g each)? The players don’t know the loot is random; so the drama is created between when they ask “what do we find?” and the GM tells them; extra dice rolls aren’t adding here.

While we’re at it, why is the loot random? It’s not modeling a complex system. This is pure output randomness and it’s not player facing.

Compare this to something like:

For every 10 minutes a character spends searching, have them roll 2-in-6. If they succeed, they get the next item in the list. Let them know when they’ve found it all.

1. 14 iron ingots and a Mining Pick engraved with “Brol”.

2. 11g and a dwarven corpse a mining pick engraved “Grelli”

3. 20g citrine, 50g amethyst, 100g ruby, a mining pick engraved “Grimni”, and a dwarven corpse (Grimni) wearing the Soul-Catching Ring

4. A dwarven corpse and a mining pick engraved “Durni”

5. A bejeweled Silver Beard Comb (300g), a mining pick engraved “Breni”, and a dwarven corpse.

6. A dwarven corpse and a mining pick engraved “Kedri”

7. A dwarven corpse and a mining pick engraved “Hruni”

Note that we condensed the text about spending 3 turns to dig into the third entry, and each mining pick is now found with it’s associated dwarf except Brol’s pick, which is found first (since it’s the most important information), and Brol’s corpse is missing (since he’s the dragon). This runs much more smoothly.

Auditing

Go through each random element. How many times will that source of randomness be drawn from?

If it’s only once, replace it with it’s most interesting result. This allows you to both replace random table entries with pre-chosen results, and random tables with pre-chosen entries.

If it’s more than once, but significantly less than the number of entries, cut the weaker entries. For example:

Hole in the Oak includes 20 entries on its random encounter table. With a 1-in-6 chance of having an encounter per turn, we’re 95% confident that we’re good for ~90 turns turns before we start repeating. I think we can cut down to 18 entries; so we can remove the weakest event on each side (#1, since it’s a red herring version of #10, and #16, since the sacred caverns beyond the waterfall do not exist as written).

Additionally, many of the random creature encounters (#11-20) have random quantities. We have enough events to not have to randomize here. 1d3 sodden ghouls can probably just be 2 ghouls. 1d3+1 fire giant beetles can be 4 fire giant beetles. 1d4 heretic gnomes can be 2 heretic gnomes, etc.

Note the difference between

1d4 heretic gnomes (from Areas 51–60), carrying empty sacks and satchels, on their way to the surface to forage for mushrooms and roots.

and

2 heretic gnomes, Fimir (Butcher) and Batabsh (picking teeth with a fishbone), carrying empty sacks and satchels, on their way to the surface to forage for mushrooms and roots.

The specificity allowed by knowing how many gnomes makes the encounter easier to run and more evocative. Players will very frequently ask intelligent creatures what their names are. There’s a gnome name+trait table provided on p34, but now we’re page-flipping.

Now that we have higher quality sources, we need to decide if it needs to be random at all. For instance, what if instead of rolling a 1d20 for the above encounter list, we rolled a 1d2 for left side/right side? Then, we just go down the list. The first left-side encounter is always the pointing root finger, then the phantasmal wizard, then the gust of wind, etc. We always get the ghouls, then the ogre, then the crab spider, etc.

This is, from the player’s perspective, exactly the same. Rolling a 1, then an 11, then a 2, then a 12 was already possible. If we weren’t okay with that sequence to begin with, it needs to not be possible.

Once viewed from that light, we can think about better sequences. For example, which leads to more interesting play?

1. Fire beetles (#14), then the foraging gnomes (#15), then the ghouls (#11)

2. Mutagenic Ogre (#12), then the sacrificial gnome (#20), then pink trogs (#19)?

The second sequence seems way more interesting to me, but both sequences are equally likely.