README.md

web-render-10k (partial)

Scrape-phase corpus from web_screenshot_ng (rescrape_to_mhtml.py): MHTML archives + raw viewport screenshots.

This snapshot is an in-progress partial upload (1612 pages scraped, 6779 domains skipped at pack time). The full target corpus is ~8.5k URLs from scrape_manifest.jsonl.

Layout

Each corpus_part_NNN.zip contains page directories:

<page_id>/
  page.mhtml          # captured MHTML
  shot_1440x900.png   # raw desktop screenshot
  shot_390x844.png    # raw mobile screenshot
  meta.json           # capture metadata

• manifest_uploaded.csv — URL metadata for all pages in this snapshot
• parts.json + corpus_part_NNN.manifest.json — which pages live in each zip
• Extract all corpus_part_*.zip into one directory to rebuild the page tree

Pipeline

1. Scrape (this data): rescrape_to_mhtml.py → page.mhtml
2. Render (later): render_clean.py --source mhtml → shot_clean_*.png