The Navigator introduction and first observations are by Guillaume Lethuillier. The systematic study that follows is by Guillaume Lethuillier and Twitch, and is also published on Twitch’s website (which contains additional images by combining animated diffs, highlighted diffs, and pairs of original images).
Disclaimer: This article contains spoilers applicable to both versions of Riven.
Card labels such as Temple:155 have a thick underline, in the color of their stack, and open that card in the graph browser.
Just how stable is Riven? [Version of the game: Riven 1.0, the original English five-CD release (Windows), September 1997. Its card numbers are used throughout, as in the Riven Graph.] It is well known that Gehn’s unorthodox writing methods invariably produced Ages that eventually fell apart, and Riven in particular is demonstrably close to its inevitable collapse. But do we see any symptoms of this in-game? Besides the endings, Riven appears perfectly normal to an untrained eye. But what if we look closer? What if there are anomalies in plain sight that we simply never noticed before?
To address this lack of tooling, I developed an application to generate a Riven navigator (simply called “Navigator” in this article) that decomposes the game world into its constituent cards. Each card contains metadata (card name, etc.), its embedded images, the location of its “hotspots” (that is: the clickable areas within the game), and their associated scripts.
Decoding these scripts was still ongoing work during this first exploration in September 2025. Therefore, a full analysis of relationships between cards was not yet possible. I kept the Navigator closed source during development and have since open-sourced it.
Even without full script analysis, the Navigator made it possible to explore the images, hotspots and script sizes of individual cards. The following observations are from that first exploration.
Unlike Myst, where each card typically embeds a single view (though some cards include partial images for visual effects), Riven allows a card to reference multiple images. For example, consider this card from Temple Island, embedding 20 images (if you squint, you can spot very subtle differences between these images):
This means that in Riven, a card generally represents a location, except for special cards like the menu or those representing books. In contrast, a Myst card typically represents a single view. Furthermore, a location in Riven can have different appearances depending on player actions: reaching card X might show a closed door, while pulling a lever (without changing locations) could open it. Both views are handled by the same Riven card (X), whereas Myst would require two separate cards: card X for the “door-closed” view and card Y for the “door-open” view.
Based on this distinction, I can start by identifying the cards with the most images associated with them. For consistency with the Myst Graph project, I use the same naming convention in these first observations: “{stack name}:{card number}”. The joint study below shortens the island name to a letter: for example, Jungle:154 becomes J154.
The entrance to the Gate Room, a room that rotates, requires no fewer than 24 images to represent all unique views resulting from different combinations of actions: (wall, screen door opened, lever up), (wall, screen door closed, lever down), (room visible (no wall), screen door opened, lever up), (room visible, screen door opened, lever down), (room visible, screen door closed, lever up), etc.
In this example, the image on the left (180 Tgr S1 Bdc Ld. 40) shows the screen door closed in front of the wall, while the image on the right (180 Tgr S1 Ido Ld. 40) represents it open, also in front of the wall. The middle image highlights the differences between these two views, with red pixels indicating exactly where the images differ. This diff image is generated by the Navigator, which automatically compares all unique pairs of images within a single card. This technique allows for the detection of even subtle variations. For instance, while the following images may seem identical at first glance, there is a slight difference in scene lighting:
The telescope views, shown above, result from the combination of multiple factors (bridge up/down, manhole closed/open, reflection of the light on the dome, lever up/down, telescope up/down, etc.).
As stated above, Riven cards contain hotspots. They represent areas of images with which the player can interact, either to perform actions from a given location (e.g., pull a lever, push a button), or go to another card (change view, move). The Riven Navigator outlines these areas with rectangles. When clicking a hotspot rectangle, the Navigator shows the script associated with that hotspot, facilitating the analysis of each action.
Here is an example of an image with typical hotspots (usually, there are hotspots on the left and right so that the player can turn their head):
Temple:155: the telescope seen from a doorway, with its hotspots outlined.
The spinning book domes require many hotspots, as each segment of the slider is a hotspot.
This one is interesting. Card Gehn:85 depicts the top of Gehn’s nightstand. From here, there are only three interactive options: zoom out (hotspot 1), open his journal (hotspots 2 to 24), or examine the metal device (hotspot 25, go to new card). The unusually high number of hotspots for the journal is a consequence of the Mohawk engine’s hotspot implementation. Mohawk only supports defining non-rotated rectangular hotspots. To achieve full coverage of the slanted journal cover, the developers were forced to approximate its shape with a series of smaller, overlapping hotspots, each linked to the same action.
Card Boiler:480 employs a similar technique but works in reverse to disable interactive areas. For example, hotspot 4 takes the player to the boiler, but it is partially obscured by rocks in the image. To prevent accidental clicks on those non-interactive parts, hotspots 5 and 6 are used: they essentially do nothing (NOPs) and cover the area of hotspot 4 that overlaps with the rock formations. The same approach is applied to the left side of the boiler.
Diffs: Images With Minimal Variations (Close-to-Identical)#
Riven’s development was remarkably meticulous, and a fascinating way to appreciate that detail is by examining pairs of near-identical images within the game files. These “diffs” (images with only slight variations) reveal how systematically the 3D scenes were rendered and updated to reflect even the smallest changes in gameplay state. Often, these differences are so subtle they are almost invisible without a direct comparison.
Here are a few examples showcasing this impressive attention to detail. (To detect them, focus on the red pixels in the middle gray image.)
This example perfectly illustrates Riven’s systematic rendering approach. The images depict the boiler from a distance, focusing on its front gauge (or level indicator) positioned just right of the door. One image shows the gauge empty, while the other displays it full. The change is incredibly subtle, almost imperceptible to the naked eye, with only 9 differing pixels.
Animated comparison (GIF):
(In the high-speed animated GIF above, the change of the gauge is hard to see.)
Looking at Gehn’s lab from the boiler reveals another small but significant variation. This diff highlights whether the ladder hatch is open or closed. It is a logical update reflecting player interaction, rendered with impressive precision.
Boiler Island seems to be a hotspot (ha!) for these nuanced changes. Like Boiler:477, this image focuses on the boiler’s front gauge at a distance. The game logically tracks whether it is full or empty, and this is reflected in the rendering despite the difficulty of visually discerning the difference.
Inside the dome, a subtle change occurs depending on whether the bridge is up or down. This variation is particularly challenging to spot due to the dark environment.
In one location, the contrast between the submarine and the sky is slightly reduced (99.99%) because some parts of the submarine’s hull match the colors of the landscape on the right.
This one is particularly interesting, as it is an easter egg (the “walk in the sky”). Again, a contrast against the sky, except that the sky is below us!
While reviewing these early diffs, I caught some interesting, minor variations. I had not conducted a systematic study yet but wanted to share these observations and invite readers’ thoughts on their potential meaning. The joint study below follows up on these early questions, including the apparent patch in the rebels’ tunnel.
Shadows and light cast by external elements have a minor impact on the lighting on the doors of Temple:304 and Temple:526. It is a second-order effect, but noticeable, nonetheless.
Looking at Boiler:200 and Boiler:230, the maglev has a faint presence. In one location, it is theoretically visible through the stairs, even if not directly seen. In the other, while partially visible on the right side of the frame, there is an intriguing effect on how the stair framework supports (the metal tubes) are rendered: they appear brighter when the maglev is present.
I also spotted what seemed to be “patches” in some comparisons, like this one in Jungle:194 within the rebels’ tunnel (a rectangular shape). These patches look as though images have been layered on top of the original view rather than being fully re-rendered.
In Temple:396, I was still trying to understand the nature of that slanted line. Smaller patches are also present there. [(Update September 2025) On the Myst, Obduction, and Beyond Discord server, user “Twitch” (drrg) offers a credible theory: the slanted line could be a narrow beam of light filtering down from above while the elevator is in its lowered position.]
You can look up the elevator shaft only when the elevator is down to see this. But when the elevator is up there’s no gap.
. . . I also wonder why the button looks different between these two shots considering that the button doesn’t ever permanently change states.
. . . I think the button is straight up in the wrong place in one of these renders.
At that stage, I could compare script sizes rather than their content, since I did not yet have a robust way to decode the binary data into readable scripts.
This location contains complex logic (complexity here correlates with data size, which holds true in this specific context). Its CARD resource contains 6,390 bytes, including a four-byte header before the scripts. The byte counts below use this same measure.
Generally, there is a correlation between the number of images per card and script complexity. However, Prison:32 stands out as an exception. It contains just over 1 KB of data but only includes one image. I did not recall ever encountering this situation in-game (that is: within the closed elevator).
These initial observations were the most readily apparent patterns, rather than a systematic analysis. My next step was to properly decode the scripts, with the goal of understanding the relationship between cards and potentially even viewing them as interconnected nodes. In parallel, the visual differences opened another line of investigation: Twitch and I began examining the alternate renders systematically. Where the Myst Graph revealed development insights through card relationships, the Riven Navigator uncovers them through visual differences.
Since renders took significant amounts of time in the 90s, and different worldstates were often rendered at different stages in development, many renders have subtle differences beyond the changes caused directly by the player. Most of these are virtually impossible to notice due to the worldstate changes being triggered in separate locations from where the actual differences are visible. But when you compare card differences in-place, changing the worldstate without moving the camera, they can become much more obvious. Such tools reveal many places where tweaks to the 3D scene were made between renders, resulting in changes to the world far beyond the changes being made by the player.
The open-source Navigator provides a methodical way to analyze the unique differences between images composing a given card. For instance, when a card contains three images (A, B, and C), the Navigator generates three unique comparative analyses: A versus B, A versus C, and B versus C. These differential comparisons, which we call “diffs”, are visualized through two complementary modes:
Pixel differential rendering: By default, the tool combines both images into a black-and-white composite, with red pixels highlighting mismatches between the two views. In many cases, the differences visible here will be caused by changes to the dithering or the color table, resulting in false positives, but quite often it can highlight small but very real changes that would otherwise be hard to spot.
Animated comparison: Reviewers can toggle a rapid succession mode that alternates between the two images (similar to an animated GIF), making differences easily visible as points of motion in an image that is otherwise mostly unchanged.
Typically, a card will have multiple diffs. The number of differences follows the triangular number sequence: 1 render: 0 diffs, 2 renders: 1 diff, 3 renders: 3 diffs, 4 renders: 6, 5 renders: 10, and so on (sequence A000217 on OEIS).
Example of pixel highlighting: the spider chair activated/deactivated, from Temple Island, card T11.
The same diff, as an animation.
We then began systematically examining these differences, expecting to find rendering glitches tied to game logic. This was a fun and almost philosophical experience: we could call up different moments in time from the same place, without having to move around to change the worldstate. It is like wearing magic glasses that represent the same location from different points in time. This article, and the follow-up articles, explore these findings in depth.
We use the following terminology throughout these articles:
ANOMALY: An “anomaly” is a visible rendering quirk. These can be caused by many things, including changes to the underlying world geometry that occurred between renders, mismatched splices, or other errors. Many anomalies are visible from multiple cards. Many anomalies require specific worldstates in order to be visible.
CARD: Anywhere the player can stand is a “card”. A card can have many renders assigned to it depending on how many different worldstates are visible from any given spot. Walking or turning moves between cards, but different versions of one spot will share the same card.
We label cards with a letter and a number. The number is that of the card in the original English 1.0 CD release of the game. Different islands can share numbers, so we prefix each card label with a letter corresponding to the island’s label within the datafiles: B for Boiler, G for Survey (“garden”), J for Jungle, O for 233rd (“office”), P for Prison, R for Tay (“rebel”), and T for Temple. For example, card J127 (pictured at the top of this article) refers to card 127 within Jungle Island. Each label opens the card in the Riven Graph.
Note: card numbers change between versions of the game and between languages. The study of the renders was first carried out on the English 10th Anniversary DVD, version 1.2, whose numbers are those of Twitch’s articles and of the filenames produced by the asset extraction tool Riveal by Ron Hayter, which was used heavily in the research and preparation of these articles. For Boiler, Survey, 233rd Age, Prison and the menu, the DVD numbers cited here are the CD numbers; for Temple and from the middle of Jungle, they are not. I matched every picture and every file name cited here with the pictures of the 1.0 discs to obtain the numbers of this article; the table at the end gives both. We have not found any difference in the actual content of the renders between the two editions.
WORLDSTATE: The “worldstate” refers to the state of all the player-actionable things, such as the position of a lever or an elevator. Whenever the player performs an action or the story progresses, the worldstate changes.
PATCH: A “patch” refers to a smaller picture or video that is placed on top of the background image in real-time by the engine. Patches are often used in places where many different worldstates are visible, to prevent the need for a huge number of renders covering every possible configuration. Patches are also used for the large number of very small videos that don’t cover the entire screen, such as when a spinning dome is visible in the distance. Patches are always rectangular, and are stored as separate images/videos within the datafiles, to be overlaid in real-time within the game.
SPLICE: A “splice” refers to when the image data of a render has been copied from another image or another render. These are sometimes used when only part of an image needs to be re-rendered, but a patch would not be appropriate or desirable for whatever reason. Splices are edits which exist within the finished render itself, and are not separate files like patches are.
FMV: An acronym for “full motion video”, referring to the many videos used throughout the game. Almost any time you see motion, it’s an FMV. (Rare exceptions include the water rippling effects all over the game, and the various minuscule flying bugs that appear in places like Jungle Island.)
OCCLUSION: In computer graphics, “occlusion” refers to when one object blocks the view of another object. When one object is in the foreground and blocking another object in the background, the background object is “occluded” by the foreground object. Several anomalies involve objects blocking one another in ways that are incorrect based on where they actually exist in 3D space. These are problems with occlusion.
An example of a “Splice”: the rectangular edges are easily visible in the differential comparison of J191.
Jungle Island is a location very familiar to many of us, but it has a large variety of anomalies, making it suitable as our first point of examination.
Several anomalies are associated with the state of the spinning dome (this is a recurring theme—every single spinning dome in the game has anomalies). In cards J121, J124, J133, J154, and J422, the two semicircular “hinges” that sit at the base of the dome are occluded incorrectly, making them fully or partially obstructed by other geometry depending on whether or not the dome is flipped. In J124, you can also see the hinge switch from high to low polygon count.
J121: the semicircular hinge at the right of the dome’s base.
J124: the hinge at the lower left of the dome’s base.
J133: the hinge at the dome’s base, near the lamp.
J154: the base of the dome, behind the leafy branch.
In card J130, the base of the dome visibly moves when the dome is flipped.
Card J154 is also the location of one of the Spyder easter egg renders. A direct comparison with a normal J154 card reveals that the camera is in a different spot. Also, the furthest of the spherical lamps in the easter egg has been raised/lit like many of the others along this walkway.
J154, the normal render and the easter egg render: the stairway is seen from a different camera position.
In cards J110, J119, and J127, the foliage changes (quite drastically in J127) when the dome is flipped. J119 also has some leaves incorrectly occluded by parts of the dome when closed.
J127: the foliage of the trees on the right and at the upper left.
At the top of the wahrk gallows, in cards J178 and J265, the doorjamb to the prison changes depending on whether the door is open or closed. In card J265 the entire jamb changes, and in J178 the differential comparison makes it obvious where the images were spliced around the jamb.
J178, differential comparison: the red area stops in a ring at the doorjamb.
There are two patches applied when you’re standing in card J181, one for the submarine and one for submarine dock 5 (the schoolhouse).
When the submarine is raised out of the water, the submarine overlaps the railing in the foreground of J181.
Furthermore, there is a patch that shows dock 5 extended, which is correctly used to show an extended dock 5 in J181 when the prison door is open. However, the render of the closed prison door already has an extended dock 5, and there is no patch to show it retracted. As a result, if you retract dock 5 before opening the prison door, you can see the schoolhouse dock retract the moment the prison door finishes its opening FMV.
J181: the submarine, in the distance, overlaps the handrail in the foreground.
J181: dock 5, the small pale shape in the lagoon at the lower left, vanishes once the door is open.
Inside the tunnels, differential comparison revealed a render that seems out of place. If you turn on exactly 3 of the lights in the tunnel, when you’re walking back toward the ocean, there is one render assigned to card J194 that’s actually a render from the perspective of J196. In the differentials, this looks like the tunnel changing shape between worldstates. In-game, this makes it seem like one of the steps toward the ocean darkens the lighting without moving forward at all. The incorrect render may be from earlier in development, as it has noticeably egregious banding on the dim/gray parts of the image, which isn’t a problem on the correct version visible from J196, or anywhere else in the tunnel. It’s likely card J194 should be completely black when 3 lights are on, as it’s barely lit (with a correct render) with 4 lights on.
J194: the misplaced render, with three lights on, then the correct render with four lights on.
The misplaced render of J194, then the render of J196, both with three lights on.
Walking down the tunnel toward the ocean, with three lights on.
Each of the 25 totems in the Tay book room can be raised or lowered individually. When you’re standing in the center of the room, individual image patches are used to allow arbitrary totems to be shown in the downward state, and renders with Tay unlocked are made with the correct totems already lowered. However, the zoomed-in views on almost all of the 25 totems have all of the totems in the same state as the totem you’re currently zoomed in on. (Exceptions: J217 and J225)
Two anomalies are visible as a result: Firstly, and more obviously, when zoomed in on a totem, the visible edges of the 2 adjacent totems will be raised or lowered based on the state of the totem you’re currently zoomed in on. Secondly, in many of the totem views, a reflection of the room is visible on the water, which shows every other totem in the room raised or lowered in the reflection.
J221, differential comparison: the adjacent totems in the bottom corners, and the totems reflected on the water at the right.
J221: the totems reflected on the water, at the right.
J220: the slivers of the adjacent totems in the bottom corners.
J222: the adjacent totems in the bottom corners, and the reflection at the right.
J223: the edges of the adjacent totems in the bottom corners.
Also, for some reason, a few of the totem renders have bright blue pixels on the extreme right or left side:
J225: two bright blue pixels on the right edge, in both renders.
J231: one bright blue pixel on the right edge of the lowered render.
J233: one bright blue pixel on the right edge of the lowered render.
J239: bright blue pixels on the right edge, in both renders.
Furthermore, some changes to the lighting occur. A shadow changes visibly in card J241 based on the totem state. In cards J211 and J213, all of the shadows cast by the totems disappear when the Tay book is unlocked.
J241: a shadow changes with the state of the totem.
At the big wahrk head staircase, at card J376, some bright white bits in the jungle disappear when the wahrk is open.
J376: a bright white spot among the trees at the far left.
Also, the angle of the camera in card J382 changes when the wahrk is open. Some foliage also changes, and the lighting overall becomes slightly brighter.
J382: the framing of the wahrk head, and the foliage around it.
At the maglev leading to Temple Island, at card J808, the rocks on the far left are different based on whether the maglev is present or missing. Also, some of the details on the concrete are slightly moved.
A few anomalies depend on whether the sunners are present. In card J621, the rocks on the right change when the sunners are absent. Furthermore, in cards J621, J627, and J629, the horizon changes visibly based on the presence of the sunners. (In card J629, the mismatched horizon is only present if you enter the card while the sunners are awake/startled, probably to match the FMV.)
J621: the horizon line, and spots of the cliff on the right.
In card J527, the shape of the background mountains changes to match the FMV when you open the submarine hatch.
J527: the outline of the mountains in the background.
In card J297, if dock 4 is retracted, a red spot appears on the rocks when the submarine is turned around. It also appears some of the rocks change, but it’s unclear which of these changes are just reflections of the submarine.
J297: a small red spot on the pale rock, to the right of the submarine.
In card J450, when the gallows shortcut ladder is still raised, submarine dock 2 will be completely missing when retracted. (The extended dock is shown using a patch.) Furthermore, when dock 2 is visible in any worldstate, it becomes apparent that the submarine dock is on the wrong side of the walkway, a quirk also visible in J448.
J450: on the distant platform, a small dark structure is missing while the ladder is raised.
J448: the submarine dock, the railed platform in the distance beyond the walkway.
In cards J469, J471, J477, and J582, the diffs reveal that the shortcut ladder isn’t actually raised. The bottom half has merely been erased. Card J335 clearly shows that the ladder should be visible when raised.
J469, differential comparison: the only red pixels form the lower half of the shortcut ladder, at the upper left.
J469: the lower half of the shortcut ladder, under the cliff walkway at the upper left.
J471: the lower half of the shortcut ladder, under the cliff walkway.
J477: a short segment of the shortcut ladder, left of center.
J582: the lower half of the shortcut ladder, at the left.
J335: the lower rungs of the ladder, seen from above.
Finally, in card J460, a detached shadow probably belonging to the shortcut ladder is visible on the leftmost cliffside at all times.
J460: a thin dark line on the cliff at the far left, in both renders.
The cards cited in this article are numbered as in the original English 1.0 CD release. Twitch’s articles, and the first version of this one, use the numbers of the English 10th Anniversary DVD (version 1.2). The two numberings agree for the menu, Boiler, Survey, the 233rd Age and Prison Island, and for the first part of Jungle Island. They differ elsewhere, because cards were added or removed between the editions: for the cards cited, the Jungle numbers of the DVD are one lower than the CD’s from 264 to 628 and two lower at 806; the Temple numbers of the DVD are higher than the CD’s, by 1 to 17 depending on the place.
Jungle 110 to 241; Temple 11 and 130; every card cited of Boiler, Survey, the 233rd Age, the menu and Prison Island
The correspondences come from the pictures themselves. The renders of the DVD match those of the CD (the ones that could be compared byte for byte are identical), so each card cited here was found among the cards of the 1.0 discs by comparing its pictures; the file names that the article quotes, which begin with the DVD number (392 Tfmd.190), correspond to those of the CD (386_tfmd.190); and the sizes of the CARD resources that the first part of this article gives (6,390 bytes for Boiler:284, 5,770 for Temple:264, 3,666 for Temple:433, 3,450 for Temple:180, 3,304 for Jungle:328, 1,072 for Prison:32) are those of the CD cards. ScummVM’s source records one such pair, Temple:371 on the CD and 377 on the DVD, which agrees with the difference found here at that place. The screenshots of the Navigator show the DVD names and numbers, as the Navigator displayed them.