The thesis
Around 150 new AI papers hit arXiv every day. Every paper-tracking tool on the market stops at the same step — they rephrase the abstract and call it a day. That answers whatthe paper says. It doesn't answer the question a builder, operator, or investor is actually asking:
Which of these papers will actually matter 90 days from now?
The method
Paper Radar doesn't summarize. It tracks whether each paper is shipping into the real world. For every paper it measures:
- GitHub reposthat reference the paper's arXiv ID in their README or code.
- Citation count and velocity from Semantic Scholar — not just total, but week-over-week rate of change.
- Time decay— recent traction is weighted more than stale citations (λ = 1/30 days, exponential).
Those signals collapse into a single Traction Score per paper, plus a categorical badge:
- 🚀ShippingCode exists. Multiple GitHub repos already build on it.
- 📈ClimbingCitation velocity is rising. Researchers are picking it up.
- 💤QuietPublished but no notable signal yet. Most papers live here.
- 🎭HypeHeavy social buzz, no shipping signal. The counter-signal — deferred until Twitter/X data is wired up.
What you're looking at
The feed lists papers from arXiv categories cs.AI, cs.LG, and cs.CL— essentially the modern LLM/agent research firehose. Ingestion runs daily at 06:00 UTC. Traction scoring runs every 2 hours. The top movers each day get a 2-paragraph narrative from Claude Sonnet 4.6 — one paragraph on what the paper claims, one on what's actually shipping because of it.
Subscribe
The full ranked feed is available as RSS at /api/rss. Weekly email digest is on the roadmap.
Stack
- Next.js 16 (App Router) on Vercel
- Supabase Postgres + pgvector
- arXiv, GitHub, Semantic Scholar APIs
- Claude Sonnet 4.6 for narratives
Built by Naren Krishna. Public, read-only, no accounts.