Data

Everything behind this site, so you (or your agent) can build your own version: yrecombinator-data.zip (7 MB).

What’s inside

  • items.jsonl: one line per item on the site. Rank, HN id, kind (thread or comment), title or text, author, date, score, URL, topic, section, number of distinct recommenders, hindsight tag.
  • recommendations.jsonl: every comment counted as a recommendation. Comment id, author, date, target item, label (rec, aged_well, aged_poorly), strength 1–3, verbatim quote, and the comment’s HTML as HN serves it.
  • labels.csv: the classifier’s verdict on all 41,033 candidate links, including the ones it rejected.
  • links.csv: all 397,309 comment→HN item links we found, with the rule that dropped each one (if any).
  • topics.json: the sections and topics.
  • prompts/: the exact instructions the classifier and the topic sorter were given.
  • README.md: field-by-field schema and method notes.

License

The underlying HN data comes from open-index/hacker-news (ODC-BY 1.0) and the official HN API. Comments and titles belong to their HN authors. Our labels, topics and rankings are released under ODC-BY 1.0 too; please credit Y Recombinator.