CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

This is a static academic paper project page for “Astrolabe” — a reinforcement learning framework for distilled autoregressive video models. It is a pure HTML/CSS/JS site with no build step, package manager, or framework.

Development

Open index.html directly in a browser, or serve it locally:

python -m http.server 8080

There are no build, lint, or test commands.

Architecture

Three source files:

index.html — page structure and content (paper title, authors, abstract, method, gallery sections, citation)
styles.css — all styling; uses CSS custom properties --astro-start, --astro-end, --astro-mid for the brand gradient
script.js — all interactivity

script.js key patterns

Lazy video loading: Videos use data-src instead of src. A global IntersectionObserver (window._videoObserver) swaps in the real src when a video scrolls into view (with an 800px pre-load margin). Any dynamically created video must call window._videoObserver.observe(video) after appending it to the DOM.
Dynamic grid population: buildDynamicGrid(sectionId, data) is a generic helper. The three video sections (#short-video, #multi-prompt, #multi-prompt-scene-cut) are populated from CASUAL_FORCING_* filename arrays on DOMContentLoaded. The #comparisons section is populated from MERGED_COMPARISON_VIDEOS and uses VIDEO_PROMPTS (keyed by numeric ID without leading zeros) for hover overlays.
Speed controls: Each .speed-controls group scopes its speed buttons to the nearest .gallery-section ancestor.

Asset layout

assets/
  logo.svg
  pipeline.png                        # method overview figure
  videos/
    casual_forcing/                   # 2-way side-by-side comparison clips
    comparisons/astrolabe/            # per-method clips (unused by current JS)
    comparisons/base/                 # per-method clips (unused by current JS)
    comparisons_merged/               # 4-method merged clips used by #comparisons section