What is Fabula?

Fabula is a bookkeeping system for stories. It tracks the state of your narrative—characters, locations, events, what each character knows, what the reader knows—and tells you when something doesn't add up.

Write a scene where two characters have a conversation? Fabula checks: are they in the same place? Do they know each other? Introduce a mysterious letter? Fabula remembers: that letter needs to be opened eventually, or you've got a loose thread.

The Human-AI-Algorithm Triad

Fabula acts as the "algorithm/bookkeeper" in a collaborative storytelling system:

The AI generates, the human guides, and Fabula ensures nothing breaks.

Why Prolog?

Stories are fundamentally about relationships and rules:

• If character A knows a secret, and A tells B, then B knows the secret
• If the gun appears in Act 1, it must fire by Act 3 (Chekhov's constraint)
• If two characters are enemies, their dialogue should reflect tension

Prolog's logic programming paradigm naturally expresses these constraints. Fabula uses a metainterpreter pattern—Prolog reasoning about Prolog—to:

1. Track narrative state as facts
2. Express story rules as logical implications
3. Query for violations and loose threads
4. Surface consequences of plot decisions

Core Concepts

Story State

% Characters exist in locations
at(alice, library, chapter_1).
at(bob, garden, chapter_1).

% Characters possess items
has(alice, mysterious_letter).

% Knowledge tracking
knows(alice, secret_password).
knows_about(reader, mysterious_letter).  % Reader has seen this

Narrative Operations

% Moving a character
move(Character, From, To, Chapter) :-
    at(Character, From, Chapter),
    retract(at(Character, From, Chapter)),
    assert(at(Character, To, Chapter)).

% Conversation requires co-location
can_converse(A, B, Chapter) :-
    at(A, Location, Chapter),
    at(B, Location, Chapter),
    A \= B.

Constraint Validation

% Chekhov's Gun: introduced items must be used
loose_thread(Item) :-
    introduced(Item, Chapter),
    \+ used(Item, _),
    current_chapter(Current),
    Current > Chapter + 2.

% Knowledge consistency
plot_hole(Character, Info) :-
    uses_knowledge(Character, Info, Scene),
    \+ knows(Character, Info, Scene).

Workflow

1. Author writes scene (or AI generates draft)
2. Scene is parsed into narrative operations
3. Fabula validates operations against current state
4. Violations surfaced → author fixes or acknowledges
5. State updated → ready for next scene

Current Status

Fabula is in active development. Current focus:

• Core metainterpreter for state tracking
• Basic constraint library (co-location, knowledge, item tracking)
• Integration protocol for LLM-generated scenes

Related Research

This project emerged from experiments on procedural story generation, specifically investigating how constraints affect LLM creativity. See:

• Thematic Passphrases - How seeds influence story generation
• Why LLMs Name Characters "Jack" - Absorption points in narrative generation