The Myst Graph, 4: What's Next?

“Structurally, narrative shares the characteristics of the sentence without ever being reducible to the simple sum of its sentences: a narrative is a long sentence, just as every constative sentence is in a way the rough outline of a short narrative.” Drawing on Roland Barthes’ observation that narrative is more than the sum of its parts, the Myst Graph project explores a similar principle. It demonstrates that the relationship between game world views (the HyperCard cards, individually; the “sentences” of the game) transcends their mere aggregation: The whole (the narrative of the game) is greater than the part (the collection of views and their own puzzles). There is an emergent narrative property at play, one that can only be captured, understood, and analyzed by considering the network-level properties (clustering, re-arrangement of edges, etc.).

My goal with this project has been to reveal a new understanding of Myst by studying it structurally, treating its components as elements within a larger system. I believe that structural analysis offers a powerful tool for video game archaeology, though its applicability is limited to specific contexts—particularly games, and, more generally, interactive media, inherently structured as a graph. The Manhole, released in 1988 by the Millers, could perhaps be a natural application for this approach, mirroring Myst’s foundation in interlinked cards. This approach could be applied to any sufficiently complex application built on HyperCard—a technology created by Bill Atkinson, whose recent passing is a significant loss. Representing other games as graphs will inevitably prove more complex; the original structure of Myst facilitated a relatively natural graph creation. To apply such an approach to free-movement games, nodes must shift from static views (files) to map locations, introducing challenges in automatically identifying edges based on how game engines encode location reachability.

In this article, concluding the series of articles dedicated to the Myst Graph, let’s now turn our attention to potential improvements. As mentioned in my previous article, further development of DeMystify offers significant potential. Several improvements are possible: first, encoding the logic of the HyperTalk scripts into the graph’s edges—for example, indicating that a connection to node 1234 only occurs when button A is pressed—would add valuable detail. A recent initiative by Andrew Plotkin, inspired by this project, makes Myst’s scripts publicly available, which would greatly facilitate this effort. While displaying all these edge constraints in a static graph representation would be impractical due to label size, as the logic would be intricated for some of the edges, they could still be leveraged for other analyses (see below). Second, extracting and processing the areas of views that connect to others could potentially aid speedrunning efforts; encoding distances from a reference point (like the screen center) to clickable areas as edge weights might offer marginal benefits. For example, if the player needs to reach a level, encoding the distance to that lever as an edge weight could potentially shave off fractions of a second. However, given the established speedrunning techniques and well-known shortest paths, any practical gains would likely be minimal. Third, following a suggestion from a reader, an interactive rendering of the graph—replacing nodes with actual view images—would greatly enhance its usability. Users could select two different views and see the most optimal path, see how groups of nodes are interconnected, etc. Finally, and more simply, using the DOT-encoded graph, anyone could suggest alternative representations to the current one by tweaking the graph’s general configuration and using a different layout engine.

Of course, the Myst Graph could be combined with artificial intelligence in multiple ways. One original idea could be to “retrofy” Myst by transforming it into a text adventure game. An AI model could be used to textually describe each view, and the Graph could be leveraged to ensure that the inner logic of the game is preserved. In other words, make a 90’s game look like a 70’s game using 20’s technology. To process the natural language inputs, a LLM model could naturally be used. On this topic, it makes me wonder, outside the scope of this project, how feasible it would be to create an LLM model able to use a graph as a system prompt or even build the model upon the graph, constraining it not by a list of rules, but by the structure of a graph. That will probably constitute my next project for this website.

I am delighted to see that Myst Graph has generated interest, particularly from Cyan, the company behind Myst, and through an enthusiastic discussion on Hacker News. Initially, I envisioned a more niche audience for this project. I hope to see further creative works built upon this foundation.