What Paperraft measures
Paperraft tracks research in efficient inference and AI agents, then counts the independent repositories that implement each paper. Author-owned repositories are separated using a published heuristic.
The score rewards independent implementation before mass attention. Papers above 1,000 stars or 200 citations are excluded because they no longer represent an early research signal.
What it does not measure
Paperraft does not reproduce experimental results. Performance gains come from the paper and are labeled as author-reported. The system does not explain why implementations appear or predict which papers will grow.
Archive status
1406 papers across 37 editions. The database, scoring code, and exclusion rules are versioned in the same repository.