The MythMaker Narrative Architecture
Why conventional conversational chatbots fail at multi-chapter fiction, and how our Event-Sourced Dynamic State Engine brings deterministic continuity to AI literature.
Academic Attribution: Detailed Outline Control (DOC)
UC Berkeley • UCLA • Meta AIMythMaker's long-form narrative pipeline is built directly upon the research presented in:
In peer-reviewed human blind evaluations, the DOC framework outperformed previous recursive architectures (like Re3) by +22.5% in plot coherence, +28.2% in outline relevance, and +20.7% in overall reader engagement. We synthesize DOC's detailed outliner with our proprietary sparse-state relational database.
Why Most AI Story Generators Fail
1. Rolling Context (Sliding Window)
Feeds previous raw text into the next prompt. By word 4,000, context limits force text out of memory. Characters forget established wounds, lost keys, or emotional betrayals.
2. Pure Vector RAG (Semantic Search)
Vector embeddings match semantic similarity, not chronology. The system frequently pulls memories of dead or uninjured characters because the scene "vibe" matches, causing fatal narrative retcons.
3. Recursive Summarization
Repeatedly summarizing past chapters creates the "telephone game." Vital foreshadowing and minor clues wash away after two summarization passes.
The MythMaker Dual-Pass Engine
Large language models suffer from attention dispersion when forced to compose literary prose and structured JSON state metadata inside the same inference window. MythMaker decouples generation into two dedicated passes:
Given the Master Bible prefix (cached) and the current chapter's DOC Beat Contract, the model focuses 100% of its parameters on writing evocative, dialogue-rich Markdown prose (~1,500–2,000 words).
A low-temperature analytical pass parses the newly drafted chapter, extracting state mutations, injuries, new locations, and immutable historical events into our database.
We credit industry pioneers like Novelcrafter's Codex and HoloAI for inspiring structured character tracking. MythMaker extends this concept with Sparse Entity States tied dynamically to every Narrative Unit.
Explore Our Technical Blog Posts
Deep dives into the math, prompt architectures, and multi-model economics behind MythMaker.