Procedural Generative Mystery Game; does this already exist?

I’ve seen a lot of stabs at this in the TTRPG space, and even the best tend to be pretty vague, outlining a situation in broad strokes, and very much meant as a creative aid for a human who would continue to have a lot of work to do to make something remotely playable.

I think Technoir comes the closest to anything I’ve seen of making something usable out of the box.

The Dreadful Secrets of Candlewick Manor is vaguer, but I find it more fun (and wholly underrated).

I’m not sure how much relevance they could have to IF design, but here are a couple more generated investigation games.

Noirlandia I’ve played, but I don’t remember much about other than use of a corkboard and red string as part of the game.

Brindlewood Bay takes the interesting tack of flinging significant-sounding clues without pre-defined meaning at the players and putting the burden on them to come up with a sensible narrative.

3 Likes

Thanks for these leads! I’m going to take a look at all of them and see what I can glean.

I’m a bit aware of Brindlewood Bay and that lack of pre-defined meaning is a sticking point for me. I do like the idea of adapting to player agency, but it seems to me a mystery that forms itself based on player decision removes the central appeal of mysteries and deduction: finding something out… it feels like it does need to be something before the players start looking, even if within a TTRPG space the DM might tweak and adapt the mystery for the sake of gameplay. ^

The Fun City Ventures crew did a one-shot of Brindlewood and although it was fun, it ended up feeling more like a pastiche of the mystery genre because of that element.

(I also acknowledge that I haven’t read those books and I’m sure there’s ways to make this work that are more subtle than my current understanding of the system!)

All of this to say: I do want there to be an actual thing that the players are engaging with, that has information to be engaged with and studied, and it’s more about the sheer volume of it requiring decisions to be made in terms of filtering and connecting it.

Which might require some kind of time limit built into the game… you can’t just sift through the data forever.

^ Now that I’m thinking about it, though… I wonder if a mix is possible. There’s a definite thing being investigated, that has generated clues, but certain clues or elements can adapt to the search.

It might also be a system of checkpoints or chapters: gather a certain amount of clues at a certain value level, and a new ‘key clue’ is found that adapts to the type of clues already found…

Red herrings are an inherent part of any mystery, and also one that is “gamified” - if you’ve played Clue/Cluedo there are potential weapons that are not used and their purpose is to verify they weren’t in the location/hands of the perpetrator and crime scene to narrow down the murderer/weapon/location.

Clue is the simplest form of a randomized game mystery and therefore lacks detail - a murder was committed in one of several rooms with one of several weapons by one of several people and a random card draw determines this. Details are not given, but sometimes emergent plot/story can pop up - the maid in the kitchen with the knife makes sense and you can imagine a scheming staff member offing her employer.

If you want a randomized mystery that changes and is replayable adding detail causes combinatorial explosion. If you watch the movie version of Clue they came up with three different ways the denouement changes based on set story beats - leaving several people out of important scenes when unseen plot points are happening. They considered a fourth ending but things got too complicated.

The trick is the base murder has to have enough open-ended threads so that any of the random possible suspects/methods can still apply until you reach a story point where it is customized for a particular outcome. This can be done easily mad-libs style like Clue “$murderer used $weapon in $location”.

I did some of this in Final Girl - basing it mostly on Clue - it was a slasher movie pastiche with several characters who could all be the murderer and could all be victims depending on RNG - the way to narrow it down was to find and identify bodies throughout a huge campsite. I found it easier to kind of hand wave the actual details of how the crime was done and load the story with horror movie tropes - masked killer, masked victims - victims triggered a movie flashback giving the player an advantage, each killer had a sort of stock reason they were the killer and because the actually mystery is vague it could all fit together in tinker-toy fashion, but the details had to kind of not focus on the mystery and be more of a boardgame type of thing - you had to progress the boardgame as busy-work for a clue to whittle down who the killer is not rather than working on who they were.

The more complicated you make the surrounding story, the more moving parts you have and unless well planned, it can be like writing entire different stories for every character. Which isn’t bad, but it can be difficult writing multiple entire plots for every outcome unless you have some gather points and crossover in the timeline/plot structure.

This can be simplified by having it sort of “modular” - once you get past a point the mystery is detailed and focused on the real killer. The risk of this is on replay it’s easy to give away - if it’s the “mechanic did it” ending the player realizes once they find blood on the wrench they’ve solved it without advancing through that entire module a second time once they recognize it. That requires potentially making more granular smaller modules that can still all fit together (maybe the weapon is a drill, but that could be the mechanic OR the gardener OR the carpenter potentially) but going too granular with randomized plot points that matter increases the amount of material the author must write and the amount that will ultimately go unused in a single play through.

I suppose procedural generation might help this, but it’s easy to fall into the “the ribbon was found in the chest, that means the debutante has been set up to be the killer”. What helps this is if the roll isn’t completely random, but perhaps details are fixed until the game is re-set and programmed not to choose the same outcome. Completionists will eventually realize there’s a potential ending for every character, so the mystery becomes less mysterious once they’ve been through six of eight suspects and know it’s one of the remaining two.

The other thing that may help is what Failbetter calls “fires in the desert” plot. Imagine a dark desert with several fires in the distance. The story only happens at the campfire points, but the player isn’t told what’s in between. With careful suggestive world building and intelligent omission of detail a larger story can be implied in the dark “between” the campfires using the player’s imagination to figure out connections that that needn’t be written. Each fire plot point just needs to be able to happen in any order and perhaps add a background element. This involves writing from a higher level viewpoint rather than the standard immersive point of view.

Vague example: Imagine two fires in the desert - a party of adventurers camping telling stories and offering food, and one party of adventurers stuck injured with their fire waiting for morning. If the character hits the healthy party first, they might talk of some of their friends missing and the player may find them later and point them in the direction of the good campsite for XP. If the player encounters the injured party first, they learn they are separated from their friends and when they find the healthy camp, they get XP for relaying information and location about them. The player had the same two encounters, but the story can work in any order. The sections walking in the dark between campfires might have had stuff happen - maybe there’s a dice roll that “injuries happened” - the player may improvise that someone fell in the dark or was stung by a scorpion (since those are logical world elements) but not need to be explicitly told. The fires are set plot points, the paths between are “Much later, during which this may or may not have happened” that can be left to player imagination.

I found the best way to figure this kind of stuff out is “how would I get this across in a physical boardgame?” Plot elements may be collected like cards and cumulatively add up to point to a particular ending that may not have been determined until the cards are drawn. IF engines with rules can do this well - essentially "if the player has seen the kitchen and the knife, it’s going to be either the Cook or the Maid or the Butler - and that can be determined later from a special “deck” that narrows it down, by drawing one of either “bloody cufflinks” “bloody apron” “bloody pepper shaker”

4 Likes

Also from the land of TTRPGs, this book, So You Want to Be a Gamemaster has fantastic advice on designing investigative scenarios. It’s based on decades of the author’s blogging on the subject; check out the Three Clue Rule and Node-based Scenario Design at Gamemastery 101 for a taste.

(It’s a bad title and worse cover; try not to be scared away by them.)

I also found Mapping the Investigation like a Dungeon especially useful.

4 Likes

Thank you for posting this, I’d heard about The Alexandrian (and the book) but not pursued either. Now I have and very glad I am too!

1 Like

Really appreciating a lot of this… it’s both examining the mechanics of the game and the aesthetics that it implies, suggests, and/or works against!

I think the Clue point is valuable, and how things can become emergent… the narrative the player constructs about why the maid would use the wrench seems exciting to me.

That’s also connected to the appeal of Scents and Semiosis by Ashwell, where the procedural work is intentionally asking you to evoke a context beyond the text generated.

Likewise, the ‘space between the fires’ as connected to that concept.

What I’m playing with in my head is trying to figure out how to both

a) make sure the thing being investigated has it’s own truth (as in, not just adapting to the player investigation, though I’m not ruling out a bit of it)

and b) make sure that the amount of detail really works at the level of implication and suggestion.

It could be something like: the system generates the ‘mystery,’ as well as the clues around it, with point values for clues closer to the truth of the mystery. The player assembles some amount of the clues, assigning their own point values based on their assessment of what they’ve found, and the endgame provides an ending that scales to the proximity between the two values. (That’s a VERY rough framework.)

The player would also know up front that the amount of data is meant to be ‘too big’ so it’s not about being completionist, it’s about both investigating and experiencing the tension between deduction and truth.

It also strikes me that a murder mystery might not be the best framework! It could be more interesting if it’s around something like politics or espionage.

1 Like

I hope this isn’t quite ‘necroposting’ or vanity posting… but…

I did recently finish building a prototype in Inform 7 based on the ideas I presented here, so if anyone who was interested in this discussion would like to try it out, I’d love your feedback!

Beta test request post over here: Procedural Evidence Text Game Prototype - Beta Testers Wanted

There was a game called Sleuth: Noir where almost every mystery you solved was procedurally generated. Sometimes they’d host games that used the same basic gameplay mechanics but were especially written for the event and the participants. The game eventually collapsed because the client was written in Flash and they never ported it elsewhere after the world gave up on Flash. There are still some aspects of its core gameplay that inspire me.

Going way back for this quote - I was going to say Moonmist but that’s basically a randomized Infocom game with four (?) different variations, but it’s not procedurally generated. One major complaint about this game is once you played every variation you could solve the mystery when you find the first clue to know what “version” of it you got in the RNG.

Procedurally generation of a mystery is kind of difficult unless you go the Clue/Cluedo route that the mystery is mostly the same (and often purposefully generic) but only specific key elements are randomized.

You might be able to quasi-QBN it where clues you find aren’t specific - talking to suspects just builds up a “You discovered a clue!” quality and you can grind that quality to 100 to be sure of the solution (which isn’t specific, just “you discovered enough clues to get it right; here’s the [randomized] story…”) or solving specific puzzles offer 5-10 clue points based on difficulty. Say if you found 60 clues of 100 you’d have a 6 in 10 chance the ending dice roll is successful and you solve it.

1 Like

I should also note, after reading through this whole thread again, it’s given me a lot of ideas to try out in the next iteration of the project, though I’ll keep using the underlying engine.

In case no one followed the beta-tester thread link I gave above, I made a prototype version of the basic game loop: investigating a new student at High School, determining if they are a Nerd or Jock based on the contents of their locker.

And along the lines of the ideas and issues raised in this thread: there is a correct answer, but because of the amount of and kind of evidence, it’s not an obvious conclusion… there is intentionally uncertainty every time you try to conclude an answer.

It’s also built with procedural generation around an ‘ideology’ mechanism that means that each student is somewhere on a Nerd/Jock spectrum, and the further to the edge of the spectrum the more conclusive the evidence.

I’d love feedback on that investigative gameplay loop from this particularly smart group of makers!

The itch link and password, if you want to give it a try:

Procedural Evidence Text Game Prototype by jason-mehmel - itch.io

password: studentlocker

I can imagine this particular investigative loop as both getting a bit more complicated, and also being in a larger overall game experience, with elements being drawn from the investigation but also some more curated experiences.

One interesting idea is that this ‘evidence search’ dynamic lends itself less to things that have clear information points, like a murder or a theft or some other decisive action. I’m thinking about investigative frames that lend themselves more to ideology, like the McCarthy HUAC trials, or religious investigation and persecution of possible heresies. (I’m against McCarthyism and religious persecution, if it needs being said!)

While waiting for feedback, I’m going to go through this thread again and pull out the specific bits from everyone that will inspire the next version!

Murder on the Zinderneuf is a detective mystery with procedural generation from way back. Supposedly some of Christopher Huang’s games have randomization too, but I’m not sure which (none of those I played).

An Act of Murder is one of them.