Procedural Generative Mystery Game; does this already exist?

Hi all! This is a fairly broad question, forgive me!

I’d had an idea recently of a mystery game concept that wasn’t about finding the right clues, but instead having an abundance of information that you have to sift through and decide on which ones are the most pertinent.

What I want to avoid is making the mystery experience have only one solution, because in that situation, the player isn’t really deducing the mystery, so much as they’re deducing the solution the designer could imagine, if that makes sense.

I also want to create the experience of having an abundance of information but not having up-front knowledge of it’s importance… that’s the deducing part which I’d love to leave up to the player.

Points of inspiration are Scents and Semiosis by Sam Kobo Ashwell and Dwarf Fortress, where in both of these the player encounters procedural generated text that feels rich and full of meaning. Also Ultima Ratio Regum which recently had an update where riddles are generated that have actual solutions inside the game world.

I don’t necessarily need to hit the detail level of those last two, which is enormous, but I would like to approximate the experience.

I recently discovered Shadows of Doubt which initially seems like it’s exactly what I want to do, while also simulating a small city! But my impression there is that while it is constantly generating murder mysteries, the amount of clues are still relatively small: you find the ‘right’ pieces and solve the murder.

I suppose the difference here would be that instead of finding the ‘right’ pieces, the gameplay I’m striving for would be ‘building the case,’ and there might be a minimum threshold necessary for success, but the goal is to build the strongest case, not the minimum viable one, if that makes sense.

(I’ve got other questions, I’m working on, such as if Inform7 is the best choice for this, and how to implement and build the procgen stuff, but I also want to make sure I’m not accidentally making the same thing as someone else!)


I’ve also had a similar kind of gameplay in mind for a project, and would love to know if anyone has tried it in IF before. :grin:


As far as i know, there are several mystery games with randomly chosen perpetrator/murder weapon (like Art of Murder and I think maybe Moonmist) which requires some generative content.

However, I’m not aware of murder mysteries that have a threshold system like you’ve described. Most that I’ve seen fall into one of four categories (quoting this from a different post i wrote a while ago):

1-Have a standard puzzle game that happens to be about murder mystery, with solving the puzzles leading to solving the mystery. This is like Ballyhoo.
2-Modelling evidence and clues in-game, which have to be combined to form a solution. This is how Erstwhile works, and most of my mysteries.
3-Collecting evidence through puzzles and conversation and then having a quiz at the end (where you have to accuse the right person). This is how Toby’s Nose works.
4-Collecting physical evidence and showing it to someone, being able to make an arrest when you have enough evidence.

But all of these have one solution, usually. I have seen some “quiz type” games have different levels of success (like the Sherlock Holmes consulting detective game).

So having multiple solutions that you need a threshold to pass does sound new and interesting. It could be an extension of any of categories 2-4, and i think it could be really fun.


I’m afraid there are a couple of problems with such an approach. This isn’t supposed to mean it’s useless to try, just to keep in mind:

  1. Completionist players. Although they are in reality very rare, there are a lot of people who maintain they have such an inclination. They will naturally object to alternative game elements which exclude each other. That is, they will want the option to build a case against every suspect and then complain they were only able to frame a single of them.
  2. Impatient players. These are the majority (also of the self-professed completionists) and they loathe complex puzzles and what they percieve to be red herrings.

Whatever you do, you might want to contemplate how you’re going to sell your game to these two groups.


This is the part that seems most questionable to me, but you might look to the TTRPG space and talk about GMing mystery games.

How do you imagine that working, and what are you trying to accomplish by it, and how are you communicating that to the player? Because it could work really well in some cases, with some players, with some expectations. But the opposite end of “the player is only deducing the solution the designer imagined” is perhaps “the designer is just agreeing with whatever solution the player chooses,” which some players also find uninteresting, if not outright condescending.

And… I mean, any real-world mystery is going to have a single “solution.” There’s only going to be one way that the thing actually happened, though the detective could accuse the correct criminal for the wrong reasons, or convict the wrong person, or whatever.

On the other hand, if the game (especially in tabletop role-play) is more about the players collaborating to tell a story of a tense investigation, then you probably really want to give them some say in what the story is.

So I tend to think that the players who like deduction are going to want there to be a single solution for them to find. The big failure mode I worry about with multiple solutions is what happens if players see through it? What if they go “hey, I could build a pretty strong case against any of these four people” so I guess I just get to choose who did the crime? Is that part of the experience you want to give people?

Hrmph. Yeah. I just don’t see how you have multiple solutions without either alienating a good fraction of people who like deduction, or making the game explicitly about choosing who committed the crime in this story instead of only deducing it.

The “building a case” part and minimum threshold vs. strongest case stuff makes way more sense to me though. I think the question here is how you make your case to the game? The Case of the Thinky Game Jam linked to some stuff about this. But it feels like there’s a spectrum, from “the game doesn’t care, you just have to get the answer right” (like Toby’s Nose), to “you have to prove to the game that you have enough evidence” (probably lots of games).

Dunno. If the game is judging the strength of the player’s case, doesn’t that get back to players only being able to present the kinds of evidence that the designer thought of?

But allowing people to have different strengths of cases seems kind of normal: we do that with story all the time. No author expects every reader to pick up on every nuance. And even with the puzzle code for murder mystery novels (around Agatha Christie’s time? I think?) where the reader is supposed to be able to solve the mystery in theory when the detective says “Aha! I know who did it” but then there’s usually a bunch of the book after that where they’re trying to get enough evidence for a confession or whatever, and that gives the reader time to get more clues to figure it out (or, y’know, some of us just read them as fun stories, too).

You’re not the same person as anyone else, so it won’t end up exactly the same game as anyone else’s, and also:


Yeah, agreed with everything Josh said – I think with a procgen approach vs. a curated narrative, the risk is that the game-y elements of the investigation could overwhelm the thematic and narrative ones. Which could be fine; logic puzzles with a light murder-mystery theming can certainly be fun. But it might be worth thinking about ways to split the difference, so that it’s not just flavor text but also some mechanics that are different when you are investigating an arson vs. an art theft, or where the culprit is a somewhat-sympathetic victim of blackmail vs. a well-concealed sociopath, or whatever.

The other thing that occurs to me is that it’d be important to have a clear, but probably complex, rubric for what counts as a “strong case”. In the real world, this is largely governed by the legal system: a strong case is one that can be proved in court. So that means you’re not just looking for witness testimony, but testimony from witnesses who are going to be sympathetic to a jury and don’t have obvious motives to lie, or not just physical evidence, but physical evidence that was obtained pursuant to a lawful search. That might be fiddlier than what you want to put into a game, but again, it’s probably going to be more engaging to have some more qualitative system like that than a generic scoring system where each piece of evidence is just some points and you win if you get above the threshold.


Really wonderful feedback and ideas, everyone! This is enormously helpful.

@JoshGrams I love how you dug into the question, but I should make one thing clear… my concept isn’t about ‘agreeing with whatever the player chooses.’ It’s more about the investigative process not being a single-path process. Find the fingerprint, check the database, voila, mystery solved, every time.

It’s more about building the case, and structuring the gameplay loop so that building that case isn’t a process of just following my logic as a designer for a single problem. (Though this procgen idea does have the corollary problem of still having some kind of associative logic involved, and communicated to the player, so that the they know how the game works.) It’s not even so much about replayability, though I think that would create a benefit.

The procgen element is more about creating an abundance to sift through, instead of singular clues that slide into place like dominoes.

So the mystery would have a solution or answer, but the gameplay is about the work involved in convicting / convincing.

That agency you describe about letting the player collaborate in telling the story; my hope is that the abundance of information will give them agency by allowing them to choose the different areas to sift through. Almost a mystery equivalent of an open-world sim, where you can accomplish goals through stealth, combat, negotiation, etc. etc. The end goal is still present, but the approach is their own. (I.e. pursuing alibis to the expense of forensic evidence, for example.)

(Which is also why I’m thinking procgen for abundance! If I want for them to have so much detail that it’s not about taking it all in, and to have that sim quality, procgen seems necessary.)

I hope that clarifies my goals! It’s also a good reminder to make sure the player understands those elements before they start.

And great reminder about the cake!


Another option is to keep everything in a “quantum superposition” until the player investigates it. There’s one famous game which has the old coin-weighing puzzle (there are twelve coins and a balance scale, you can only use the balance three times, figure out which one is the fake), but doesn’t choose in advance which coin is which—instead, every time you use the balance it chooses the result that provides you with the least information, and if there’s any uncertainty left when you choose the answer, your guess will always be wrong.

In other words, the game starts with evidence exculpating (or incriminating) every single suspect, and the game quietly removes any conflicting evidence as you go. You dusted for fingerprints on the picture frame before the knife? The fingerprints on the knife disappear, or change to match the new evidence you have.


@DeusIrae Great points!

Different mechanics for different kinds of cases are a great idea, and how that’s reflected in the rubric for a ‘strong case.’

This has been the thing I’ve had a harder time figuring out, and part of my explorations around procgen tools, Inform 7, etc. I feel like this is the necessary direction for the kind of game I want to make, but it means that it’s not just creating an abundance of details, but finding ways to associate them to the mystery at varying levels of priority.

Maybe it would have something to do with whatever the inciting element of the mystery is: if there’s a murder, then things like the weapon and eyewitness testimony would have an associated higher priority, but also might be ‘harder’ to find… whereas a lot of secondary evidence and testimony might take it over the line, but there’s more to accumulate.

(Which could create an interesting, different gameplay loop. You could Dark-Souls it and hunt for the strongest clues with more difficulty, or ‘grind’ the weaker clues into a still-strong case.)


Oh, that’s an interesting way to go along too!

I feel like I do want the ‘answer’ to be established so that the player knows it’s out there, somewhere… but this mechanic and idea is really interesting and could also be part of a time-based sense of urgency.

All possible investigative paths are available, but as you progress, some of the elements change in the back end to close off that path.

But I do think there’s something really valuable about that 'quantum superposition idea. Maybe that can be a way to structure how the procgen works by not just building a huge mass of associative clues up front, it might be quantum at the level of which kinds of clues build at which levels of investigation.


I think Inform 7/10 would be pretty good for implementing a system like this, actually – you can create relations between objects (like say, clues and suspects) that could indicate which create suspicion about who, as well as including a prioritization level.

This makes sense – this could also be thematic, where the player can just pull together a sufficiently large heap of circumstantial evidence to convict several suspects, and therefore reach an ending, but might be uncertain about whether they’ve fully cracked the case even if they do get the right suspect.


So… a Prosecutorial simulator? :grin:


Ah, I see, so “avoid making the mystery experience have only one solution” means you might find and use different clues to build your case for the same mystery: you’re not going to necessarily find all of them. That makes a lot more sense.


Actually, now I really think you should look at Toby’s Nose: it doesn’t do any of the judging how strong your case is, you just have to accuse the right person, but it has boatloads of nested detail, much of it irrelevant. And no support for whether you’re on the right track: all the deduction happens in the player’s head.


This is a half-thought because my brain’s still on holiday but I feel like there’s something worth exploring in regards to playing with player biases here? A wealth of evidence could lead to an overwhelm, and when humans don’t know which way to go we tend to lean on our biases (whether consciously or not!), and I think there could be a really interesting element that brings the player’s biases to the surface there that’s really made possible by increasing the amount of evidence available.


(Also this really makes me want to finish a piece I started years ago inspired by Sarah Krasnostein’s The Trauma Cleaner, where you’re figuring out what happened by going through a hoarder’s house and cleaning it up after they died!)


This is Spellbreaker, right?
Another good one where this happens is The Hitchhiker’s Guide to the Galaxy, which has you pick up twelve items for the end, but if you miss even one, it will always be that missed one. You have to have all twelve for it to be “randomised”.


Spellbreaker, yeah. It’s a very awkward puzzle in many respects, because it doesn’t play by the rules of the rest of the game, or of the fictional world—SAVE is disabled, “piles of cubes” are (openable!) containers that cannot be created or destroyed (even when empty, it’s still an “empty pile”), the relevant spell functions differently in that room than in the rest of the game, there’s no logical reason why the real magical artifact should be less magical than the fakes…but, it’s an interesting case study in this sort of “quantum puzzle” design.


There is a “generative forensics” approach, when game procedurally generates a world and events in it, and you as a player have to deduce what happened - i.e. solve the mystery. This talk goes into details on how this is made:


Thank you for this!