Thanks for playing Condition Unknown! I hope you enjoyed it. If you haven't had a chance to play it yet, you might want to do so before reading this post. It's spoiler-free, so I won't ruin how the systems work, but you might enjoy the game more before reading this nonetheless.

Exploring Generative Forensics

Last year I made Nothing Beside Remains, in which you explore a ruined village and try to figure out what killed it (this may already sound familiar). If you've not played it, it only takes a few minutes, and you might enjoy contrasting the two experiences. The idea was to procedurally generate a village via simulation, simulate it until it collapsed, and then have the player try to understand the systems by observing the end result. Unfortunately, I didn't have time to actually build a simulation, so the generation of the village was very abstract. In the developer notes for NBR, I wrote:

Ultimately, I didn't have the time to fully explore these ideas here, and so it's ended up extremely hard to really intuit a village's fate from the information provided. 

This left me with two goals for this year. I wanted to build a similar game about "generative forensics" - seeing the endpoint of a simulation and trying to understand the generative system that produced it. But unlike last year I wanted to use a more detailed simulation of events, to see what new affordances and design challenges it would offer. The other goal was to see if I could challenge the player to actually draw concrete conclusions from the evidence the game offered. Last year I wrote:

Another important aspect of the game is that it doesn't confirm or deny your findings, there's no way to even input 'findings' into the game, nor to return to a village later or show it to anyone else.

This was a really important part of NBR, but in the year since I've thought a lot about the long-term project idea I have. Is it possible to get the player to derive real knowledge from a simulation's end state? What are the challenges in shaping or guaranteeing that knowledge? What kind of knowledge can they derive? There's definitely value (and poetry) in not evaluating the player's knowledge, but I wanted to experiment with the other side of things too.

Condition Unknown explores both the idea of a deeper simulation, and the idea of extracting information from it. The research station's events are simulated, using little (or not so little) agent systems representing the staff and other things in the station like fire or... other things. That means that logs found by the player during gameplay represent real events, with timestamps that reveal their actual order in the simulation. And rather than solely exploring, the player may - if they want to - record a report summarising how they think each staff member died, and submit it. Submitting the report ends the game instantly, meaning they can't go back and correct mistakes, which keeps a little bit of the mystery intact.

Procedural Information Games

Games like Return of the Obra Dinn, Heaven's Vault and Her Story are what Tom Francis calls 'information games'. Tom describes them thusly:

An information game is a game where the goal is to acquire information, and the way you do it is to use information you've already gained and reason about it, deduce things from it, come up with theories and use those theories to go looking for more information.

Another rule of thumb for identifying these games is that you tend to end up using a pen and paper to write something down, and that thing is not a code or a map. One of the things that makes information games so exciting and clever is that the player typically ends up doing very complicated things, like making educated guesses, following hunches, making emotional arguments, drawing conclusions based on social and cultural knowledge - things that computers have a lot of trouble understanding or reasoning about. The genius of most information games is that the computer never has to understand or reason about any of these things. That part of the game takes place outside the digital, in the player's mind or in their notebook. The game itself always has a much more elegant interface: type words into a search box, choose from a list of possible translations, pick from a list of names and causes of death.

One of the things that I think makes these games stand out is that information is normally picked up in tiny bits and pieces, rather than in big chunks. In a lot of mystery games, including point-and-click adventures or research projects that try to generate mysteries, we find information is deposited in huge binary chunks. Before, you didn't know that the killer had black hair, then you found this thing that tells you they did have black hair, and now you have this as an established fact. Westerado is a good example of this, and AI researchers also like this approach because we can do interesting and cool reasoning about this kind of information to design mysteries up front. Information games have that information trickle in, and similarly often let you advance in trickles too. There are a few big gotcha moments in Obra Dinn, but it's mostly characterised by small realisations - noticing a facial expression, an object being casually handled by someone, a number on a wall or a relationship between two people. Nothing in the Obra Dinn guarantees the story is solvable - some people could play it for hours without finishing it, while others might instantly see every important piece of information. There is no AI system ensuring its story is solvable behind the scenes.

This design format is perfect for procedural generation via simulation, I think. Instead of worrying about chains of deduction and guaranteeing player information in big chunks, we focus instead of making a rich simulation of events that creates an interesting end state. In this simulation, we ensure that there are systems that offer up information to the player, but in small doses rather than huge binary facts. If someone shoots at something in Condition Unknown that means they had direct line of sight to that thing, it means they had a gun, it means they believed the thing they were shooting at was dangerous. The game itself does not need to understand all of these implications, but the player might, and each one is a tiny information crumb that can be added to their notebook.

I talk more about these ideas in the Spoiler Notes, and specifically how Condition Unknown's systems overcorrect and try to make sure there's always a way to figure out the truth. I won't go into it too much here in case you want the mystery preserved, but suffice it to say that I think this is a really promising use of procedural generation.

Inspiration

For Nothing Beside Remains I talked about inspiration, and of course this game also shares many of the same inspirations, especially Return of the Obra Dinn, The Thing, and Tom Francis' thoughts on information. Condition Unknown is another prototype on the way to something bigger, but that bigger thing is unlikely to materialise for many years! I really hope you enjoyed playing it though.

More from me on Twitter: @mtrc

More writing, research and games on my personal site