Serenity Forge

At Serenity Forge, I worked closely with the Producer and Creative Director to develop a innovative narrative system that would algorithmically maximize player engagement. In practice, this meant developing a robust database of measureable player stats, analyzing how different clouds of data corresponded with particular user personas, and authoring content to cater to different player types. (This title is still unannounced.)

​

(If you would like to see an NDA--safe version of this work, please click here!)

​

I'm sure this sounds like a slurry of of buzzwords so let's break it down with an example (while respecting my NDA!):

​

Imagine we have a player named Artemis who, after a few scenes of creeping dread, returns to her character's hotel room to find... a body! Artemis is invested in the story and quickly begins navigating her character around the room and interacting with every interactable. Artemis reads all the documents she finds, carefully considers what it means, and realizes it points to a secret safe in the corner. Since Artemis has read everything, she is able to input the code with no help from the game and finds a secret clue that leads her to the next scene -- a dramatic confrontation with the criminal! Artemis' character debates with the criminal about the philosophy of crime.

​

Imagine now that we have a second character named Phil. Phil is just here to have a good time and has spent the last few scenes playing all the mini-games and trying to jump on tables. As Phil walks into the hotel room, they find the body! But their character audibly explains who it is and why it's significant -- Phil has not been paying attention. Phil walks into the body and finds it has physics and then kicks it around the room a little. Because of how many tables Phil has jumped on, the game allows Phil to jump out of the open hotel window to the street below. There, they catch a glimpse of the criminal. Then, the game cuts to the next scene -- a confrontation with the criminal! But in Phil's confrontation, it's a gameplay sequence where Phil has to use the game's platforming to run down the criminal. Phil's character punches the criminal before they can talk, because Phil would think this is funny.

​

As you can see, these are very different players. How can we tell what each of these players' like? Consider the following measurable questions:

• Time: How long has the player been playing the game? How long did the player spend in a screen with a document? How long did the player spend on top of tables?

• Frequency: How frequently did the player press the jump button? Or the sprint button?

• Reading Preference: Did actually scroll to the bottom of each document? Or did they open them to get rid of the notification, and then close them? Did they spend long enough looking at a document to actually read it? Do they act on the information found in documents without additional support?

• Story Investment: Does the player spam the "next dialogue button" during cutscenes? Do they turn around at the beginning of each level and go backwards? Does the player control their character to act realistically for the situation? Does the player play the game like they're trying to win or like they're roleplaying a story? (There's nothing wrong with either answer -- but those different players want different things out of their game!)

These are just some of the statistics that would set Phil and Artemis apart -- not to mention every other player.

​

This sort of system isn't designed to make the game an entirely different game for each player, however -- it's something more akin to fine--tuning it to appeal to all members of a general demographic's playstyles. Both Phil and Artemis bought this murder--mystery game, after all!

​

Now, you may ask: How do you write content for such a game? The method we used was to make a golden path through the level first and then plug-in alternative pieces of content around it. Whether this content would appear or not was determined by probability according to broad statistics (such as "Reading Preference" or "Story Investment"). And these categorical statistics were measured by the relative weightings of many smaller statistics (as you can see demonstrated in the list above.) We implemented this numerical changes with Logic Driver flow charts in Unreal Engine 5.

​

While the success of this system is that it was fully fleshed out and accounted for, its biggest downside was the amount of content it required to be as flexible as we wanted. But while working on it, I developed highly organized systems in Notion to write, tag, and organize all these pieces of content and statistics -- as well as a better understanding of how to pitch, design, and catalogue modular content for the other designers, programmers, and artists on the team. Since this was by far my longest assignment yet, I really had the opportunity to dig into this design system and its implementation -- both in our game and in other games like it.