How it works

hackerrepos collects public comments from the hacker discussion archive that link a GitHub repository, then asks a language model (Claude Haiku) to label each mention: strong recommendation, recommendation, neutral reference, criticism, joke, or the author promoting their own project. Only the first two count.

Sources. Older comments come from the open-index/hacker-news dataset (ODC-By 1.0). Days after the dataset's last entry are filled in daily from the public Hacker News Search API by Algolia, using only comments that link github.com. Repository details come from the GitHub API.

Ranking is the number of distinct people who recommend a repo. Nobody counts twice for the same repo. Stars, forks, sponsorship and affiliate relationships are never inputs. Ties break on strong recommendations, then on the most recent recommendation.

Rising counts new recommenders in the last 30 days of data against the repo's earlier pace. Alternatives are pairs where one comment offers a repo as a replacement for another.

Quotes are checked to be verbatim text from the comment; if the model's quote does not match, we show none. Every quote links to the original comment.

Limits. Labels come from a model and can be wrong. Repos that were deleted or renamed away are dropped. Coverage is the date range shown on the data page.

This is an independent project and is not affiliated with Y Combinator.