About

Why a graph?

Real knowledge isn't a textbook. To read a frontier paper you need a chain of prior ideas — algebra, probability, the history of attention mechanisms, the architectures that came before. Charted makes that chain explicit. Pick a paper, see the prerequisite closure. Pick a topic you already know, see what it unlocks.

Scope

We start narrow on purpose. The current MVP covers the prerequisite closure for modern machine-learning research — about 2091 topics and 3661 prerequisite edges. Topics outside that closure are archived in the repo but hidden from listings. As specific frontier papers pull more areas back into scope, those topics return.

How agents and humans split the work

Agents draft topics and propose connections — they're good at scanning vast literatures and producing structured first passes. Humans review what the agents wrote. Crucially, not just any human: each topic has a review tier (foundation, field, or frontier) that determines who is qualified to approve it.

Foundational topics (algebra, calculus, probability) are stable enough to be community-reviewed. Field-level topics need someone with several papers in the area. Frontier topics — the ones tied to a specific paper — need an author or a substantively cited researcher.

Reviewer credentials are tracked publicly in reviewers.yml and surfaced on each contributor page. There are 1 reviewer registered today.

How the system works

Every topic is a Git-tracked MDX file. Every PR includes reviewer-attestation trailers — GitHub handle, ORCID, and asserted tier — which CI parses and checks against the registry. The check is advisory: a maintainer makes the final merge call. As the reviewer pool grows, gating tightens.

GitHub is also where coordination lives: each topic has a tracking issue (status, tier, claims, discussion). No backend, no accounts on the site itself — the source of truth is the Git history.

Why expert outreach

For each frontier topic, the primary paper's authors are the natural reviewers. We mine arXiv metadata as we ingest papers and reach out to authors with the page that cites them as a primary source. It's not cold outreach — it's an invitation to verify a public attribution.

Contributing

Read the contributing guide. Pick a topic in your domain. Open a PR. The review-check bot will tell you what's missing.

License

Code: MIT. Content (src/content/): CC BY-SA 4.0.