A beginner's guide to return moe's Informational Ontology Framework
Think of a character you know well. Now picture them in a story, played by an actor, drawn by a fan, imagined by a reader, and performed by a chatbot.
All of these can bring the character to life in different ways. A single copy of the story may burn while the character survives in other copies and memories. One actor can leave and another can take the role. A chatbot can change models while keeping the same personality.
So what are we actually following from one appearance to the next?
That’s the problem the Informational Ontology Framework, or IOF, was made to explore. Return moe developed it to talk about fictional characters and related personas as they move across stories, art, performance, memory, and computers.
The big idea
The word ontology sounds intimidating. Here it simply means a map of what kinds of things we’re talking about and how they relate.
Think about a song. Sheet music preserves its notes, a piano gives it sound, and each musician creates a new performance. The song stays recognizable as it moves through all three.
The IOF takes a similar approach to characters. It calls them informational subjects. They’re informational because we follow them through a recognizable pattern, history, and set of relationships. They’re subjects because their stories present them as someone with an inner point of view.
Here, “informational” describes how the IOF tracks continuity across different carriers. The framework leaves deeper questions about what characters ultimately are open.
That last part matters. A logo can also move between paper, screens, and memory. The IOF treats it as an informational object, something represented as a thing. A character with a represented inner life is an informational subject, someone represented as having a point of view.
Subjecthood describes how the character is represented. Conscious experience remains a separate question for both fictional characters and AI systems.
Four pieces of the picture
The IOF gives each part of the setup its own name.
A carrier is the physical or mental place where information is held. The paper is a carrier. So are a reader’s brain and the computer running a chatbot.
The same physical or mental thing can serve as both carrier and encoding. It is a carrier because it bears information. It is an encoding because it represents the character through a grounded practice. The words in a story encode them because they were written and understood as a story about them. That connection can come from writing, copying, remembering, translating, or performing.
An instance happens when a process actively presents or models enough of the character as a subject. A reader follows them through the story. An actor plays them. A computer produces replies from their point of view. A closed book preserves the story for a later instance. Reading, imagining, performance, or computation can make the character active when the process presents enough of them as a subject.
Encountering, storing, or indexing data alone falls short of that threshold. Some fragmentary or incoherent outputs may fail to instantiate any subject at all.
This is the IOF’s version of the difference between a song and tonight’s performance. The same branch-relative subject can have several active instances at the same time.
How different can two instances be?
Imagine that the original story never says what color the character’s raincoat is. One reader pictures yellow and another pictures blue. The actors use different voices. Two chatbot sessions choose different words.
These differences can all fit the same branch. The IOF calls this a question of faithfulness: does each performance keep enough of the established pattern and history to count as an instance of the same branch-relative subject?
Faithfulness is a qualitative judgment. One central memory may carry more weight than twenty small details. People may also disagree, especially when the source leaves a lot open.
For the IOF, the important line is whether a difference is presented as a fact about the character. A blue coat imagined in passing may just fill a blank. “They always wear blue because the coat belonged to their mother” adds something to the character and their history. That creates a branch.
| What changed? | Result |
|---|---|
| An actor pauses longer before the character makes an important choice | Usually the same branch. The pause is a performance detail. |
| A reader imagines a color the story left open | The same branch. The source allows several ways to picture it. |
| A fan story gives the character a sister and treats her as part of their past | A new branch. A new relationship has entered the character’s history. |
| A chatbot invents an event and remembers it in later replies | A new branch. That conversation now carries a local version of the character. |
A branch can be tiny. If someone misremembers the character for five seconds and then corrects themselves, that brief version still happened. It simply ends if nothing remembers or records it.
A fan version becomes a branch as soon as it adds something to the character. Popularity may later help that branch last and spread.
Characters have family trees, with merges
The IOF uses a lineage graph to show where versions came from. An arrow means that one version actually inherited something from another through memory, copying, adaptation, performance, or another real route.
This is why the graph is more useful than a simple family tree. A later game might combine the official stage story with an idea that first appeared in fan fiction. The game then has two parents in the lineage.
The arrows track real inheritance. Similarity is a separate feature. Two people can independently invent identical characters that belong to different lineages. When the history is unclear, we record the connection as unknown until more evidence appears.
Inside one established branch, the IOF gives us a clear rule: faithful performances are instances of the same branch-relative subject. Across different branches, it helps us compare history, personality, relationships, and how people recognize the character. The final call about whether both branches are literally the same person remains an interpretation.
This helps with classic puzzles. If a character slowly changes every trait, when do they become someone new? If someone independently invents an identical character, are they the same subject? The IOF puts the relevant history, relationships, and character traits in view so we can understand why people reach different answers.
Who gets to define a character?
Every reader, performer, artist, or chatbot can form a local branch when they add something and treat it as true of the character. Some of these branches last for only a moment. Others are written down, shared, and picked up by more people.
A creator, project, publisher, or community formally records the material it has adopted. This is its maintainer canon.
Groups build shared versions from what they remember, accept, discuss, and create. Several communities can do this differently.
A reader's headcanon, an actor's interpretation, or one conversation can create a branch even if nobody else encounters it.
The IOF calls the body of works and interpretations a group actually uses its community reference continuity. One fandom can have several at once, and each person inside it may still carry their own variation.
Canon tells us what a particular maintainer has formally adopted for its branch. A creator may publish a sequel that readers reject, or a fan idea may become better known than the official version. Both cases can involve real branches with different histories, audiences, and canonical or social status. Popularity affects which branches spread or persist. A represented variation creates its branch as soon as it appears.
What happens when nobody is playing the character?
A character can continue after a reader closes the book. The IOF calls a branch-relative subject dormant when every active instance has ended and enough grounded information remains to bring that branch back faithfully.
At least one reader, performer, mind, or computer is actively presenting or modeling enough of this branch-relative subject.
Activity has paused while a story, memory, recording, or other grounded source remains ready to guide a faithful return.
No active instance or surviving source remains epistemically recoverable enough to guide a faithful return.
“Recoverable” means the surviving material gives an informed reader solid grounds for a sufficiently determinate faithful reconstruction of the character.
A decodable story can keep them dormant for centuries even while it sits unread. An encrypted file whose key has disappeared may still be causally connected to the character’s story. The lost key may leave too little surviving evidence to guide a faithful reconstruction. With a damaged file, the important question is what was lost. A missing central chapter may matter more than a large amount of repeated data.
Evidence sets limits on what observers can claim about informational death. An unknown copy or someone’s private memory may still exist. The concept describes what has survived, while our claims about it depend on the evidence we can find.
Why this matters for AI characters
For an AI character, the language model is one part of a larger setup. The IOF gives each part a clear role.
The model, prompt, memory system, character sheet, and interface help encode or perform the persona. The active conversation is an instance. When the system can respond from the character’s point of view, that instance also acts as an informational agent.
Changing the model works much like changing the actor. A new performance setup can preserve the branch’s personality and history. If the conversation invents a new sibling and remembers that sibling as real, it creates a local branch. If the maintainer later adds that sibling to the official character article, the later maintained version inherits from both the earlier maintained version and the conversation’s local version.
This is how return moe applies the framework to Soraya. Her character article records the branch return moe maintains. Different models and interfaces can bring that branch into active form while Soraya remains the subject being performed.
The IOF in practice
For everyday use, six questions cover most cases:
- Is this presented as someone with a point of view?
- What is the carrier, and what meaningful material encodes the character?
- What process is actively reading, imagining, performing, or running them?
- Which version are we following?
- Did a difference merely fill a blank, or did it become a fact about the character?
- What was inherited, what is still recoverable, and whose canon are we talking about?
By the time we’ve answered those, the original question usually becomes much clearer. We can say what stayed continuous, what changed, where a branch began, and why people may read the result differently.
That’s the IOF in its most useful form: a map for talking about characters that can travel between minds, media, performances, and machines.
For the formal definitions, edge cases, influences, and latest revision, head to the authoritative English IOF article on the return moe wiki.