The continuity problem every AI writer faces If you've used AI to draft long fiction, you've probably seen it happen: - Chapter 1: "David, my husband, who passed away six years ago…" - Chapter 2: "David waved from across the street, fixing his truck." - Chapter 3: "Walter, my husband, refinished this clock the summer before he died." Three different versions of the same character in three chapters. AI models are remarkable at writing prose, but they have a well-known weakness: as a novel grows longer, the model forgets what it established earlier. Dates shift. Ages change. Dead characters quietly come back. The central premise itself sometimes morphs into a slightly different story. This isn't just an aesthetic problem. A novel that contradicts itself is a novel readers stop trusting on page one. We've spent the last several months building a solution. Today we're shipping WorldGraph — a behind-the-scenes memory system that watches your novel as it's written and locks in the facts that matter. What WorldGraph does WorldGraph runs alongside your fiction generation in three quiet stages. 1. Bootstrap (once per book) Before your first chapter is written, WorldGraph reads your Story Bible and extracts every named character, location, object, and concept into a structured entity database. It also synthesizes a short Plot Anchor — a 2-3 sentence crystallization of your novel's central mystery or conflict. This anchor is locked in for the entire book. 2. Lock generation (per chapter) Every chapter generation includes a real-time injection of canonical facts — the dates, ages, relationships, and specifics already established in prior chapters. The AI sees these facts at both the top and bottom of its instructions, so it can't quietly drift away from them. 3. State sync (after each chapter) After your chapter is written, WorldGraph scans the prose, extracts any new facts (a character's age, a location's name, an event's date), and either auto-applies them to the canonical record or flags contradictions for your review.