About

Y Recombinator is Hacker News, filtered through Hacker News. Every item here is an HN thread or comment that other HN commenters went out of their way to link and recommend: “read this”, “best explanation I’ve seen”, “this aged well”. Items are ranked by how many different people did that, from 2007 to today.

It’s modeled on Andrej Karpathy’s HackerTube, which does the same thing for YouTube videos.

How it was made

1. Read all 50.4 million HN items: the open-index/hacker-news dump, topped up with the HN API and Algolia for everything after August 21, 2026.
2. Found 320,822 comments that link to another HN item (397,309 links in all).
3. Dropped the obvious non-recommendations with plain rules: links to your own posts (80,306), links within the same thread (31,978), moderator “Related:” lists and “previous discussion” pointers (109,219), dupe and moderation notices (7,901), bare links with no comment (33,293), and dead or deleted items (8,216).
4. Kept items that at least two different people linked, then had Claude Sonnet read all 37,260 remaining comments and decide why each person posted the link: a genuine recommendation, notable in hindsight, or just a reference, a criticism, a dupe notice, self-promotion or spam. 5,067 of the 41,033 links were recommendations.
5. Ranked items by distinct recommenders (an item’s own author never counts). Items only one person recommended are included when that endorsement was clear or emphatic (“worth reading”, “best explanation I’ve seen”); they always rank below anything two or more people recommended. Set aside HN administration (moderator comments, site-policy threads, hiring threads), and sorted the remaining 2,013 items (1,733 threads, 280 comments; 726 recommended by two or more people) into 119 topics in 12 sections. 2,621 different people recommended something here.

Every title, snippet and quote is verbatim from HN. Topic names and the “aged well” / “aged poorly” tags on comments are model output.

Caveats

• The classifier is good, not perfect. The full label set is in the data download if you want to check its work.
• Only direct links count. Someone who praises a thread without linking it isn’t counted.
• Older items have had more years to collect recommendations, so the all-time lists lean old. Use the year filter, or recent.

Privacy

No accounts, cookies, analytics or third-party requests. Likes, hides and topic picks for For You stay in your browser’s local storage and never leave it.

Keyboard

Press ? anywhere, or see the list.

Not affiliated with Y Combinator or Hacker News. Comments and titles belong to their HN authors.