HPMOR Knowledge Base — Rationality Cinema Pipeline

Current Goal

Selecting scenes for a 4–5 minute first demo video from the Azkaban escape arc (chapters 52–59). Building a knowledge base of HPMOR's most cinematic moments — scored, shot-split, and ready for Veo generation.

Status

9 chapters extracted (ch1, ch52–59). All scenes scored on a weighted rubric (Visual×3 + Self-Contained×2 + Character×2 + Dialogue + Action + Emotion). All video prompts split into individual shots (≤80 words each).

Chapter	Title	Scenes	Shots	Top Score
1	A Day of Very Low Probability	2	9	70
52	The Stanford Prison Experiment, Pt 2	4	17	90
53	The Stanford Prison Experiment, Pt 3	4	14	88
54	The Stanford Prison Experiment, Pt 4	6	17	92
55	The Stanford Prison Experiment, Pt 5	5	15	92
56	The Stanford Prison Experiment, Pt 6	6	14	82
57	TSPE, Constrained Cognition, Pt 7	4	13	85
58	TSPE, Constrained Cognition, Pt 8	4	13	95
59	TSPE, Curiosity, Pt 9	8	22	88

Total: 43 scenes, 134 shots ready for generation.

Method

Extraction follows METHOD.md — each chapter is processed individually by an AI agent that reads the raw text, extracts characters/locations/magic/scenes, scores every scene, and writes shot-level video prompts for Veo.

Iterative improvement loop: extract → lint → critique → update method → repeat. Each chapter refined the extraction schema.

What's Been Built

Character blocks — copy-pasteable visual descriptions for Veo consistency
Location visual blocks — lighting, color palette, camera angles
Magic catalog — visually representable spells with HPMOR-specific details
Scene scoring — weighted rubric, 1-100 per scene
Shot-level prompts — one camera angle + one action + one lighting state per prompt
Lint tool — tools/lint_chapter.py validates all schema constraints

Source

jbnicolai/hpmor — plaintext export of Harry Potter and the Methods of Rationality by Eliezer Yudkowsky. 122 chapters, ~660,000 words.

Pipeline

raw/chapter/*.txt → processed/chapter-{N}.json → compiled/scenes/ranked.md → Veo

index.md