Engineering & Research Blog

Narrative AI Architecture

Technical essays on state machines, Detailed Outline Control (DOC), and long-context LLM storytelling.

Problem Analysis 10 min read • Published Oct 2026

Why AI Novels Fall Apart at Chapter 4 (And How We Solved It)

A post-mortem on the 3 naive approaches powering most consumer AI wrappers: Sliding Window context, Vector Semantic Search (RAG), and Recursive Summarization. Learn why temporal truth breaks down, and how an Event-Sourced Dynamic State Engine guarantees zero character amnesia.

Architecture & Benchmarks Read Article

Academic Research 12 min read • Published Oct 2026

From Re3 to DOC: The Science of Detailed Outline Control

An in-depth breakdown of Kevin Yang et al.'s landmark paper (UC Berkeley / Meta AI). We explore why high-level outlines lead to wandering plots, why Re3 struggled with creative drift, and how DOC's granular beat contracts produced +28.2% outline relevance in human blind tests.

Research Attribution: alphaXiv 2212.10077 Read Article

System Design 8 min read • Published Oct 2026

Inside the MythMaker Dual-Pass Engine: Decoupling Prose from State

Why forcing an LLM to generate 2,000 words of prose inside a JSON structure degrades prose quality and causes token starvation. Discover how our Pass A (Literary Novelist) and Pass B (Analytical Observer) achieve unmatched style while maintaining complete state tracking.

DeepSeek-V3 Prefix Caching Read Article