review api

Every review, machine-readable

Unauthenticated JSON over GET, CORS open, cached five minutes. This page covers the review record; the signed integrity records (bundles, events, Ed25519 verification) have their own page at /developers.

Endpoints

GET /api/v1/papers/{arxiv_id}

Paper metadata plus the current verdict summary: verdict, confidence, novelty score, correctness risk, community score, inbound citation count.

GET /api/v1/verdicts/{arxiv_id}

The full current review: one-line summary, lead, skeptic critique, referee report, author rebuttal, the whole extraction, plus three first-class fields: falsifier (the observation that would settle the core claim), supporting_citations (the works the core claim leans on, with evidence quotes), and review_version (how many times this paper has been reviewed).

GET /api/feed.json

The live review feed as JSON; ?sort= and ?category= mirror the homepage controls.

GET /api/v1/dumps/manifest.json

Bulk access: monthly JSONL.gz dumps of every current verdict, with download URLs. Documented at /dumps.

MCP server for assistants

A read-only Model Context Protocol server runs at https://pith.science/mcp (streamable HTTP transport). Eight tools, plus get_claims and get_falsifier as aliases of the two renamed ones:

get_review(arxiv_id): Pith's verdict, scores, summary, weakest assumption, and the paper's argument chain (premise, machinery, core claim, falsifier).
search_papers(query, limit): full-text search over reviewed papers by title and abstract.
latest_reviews(category, days, limit): most recently reviewed papers, optionally by arXiv category.
analyze_paper_claims(arxiv_id): what one paper asserts, including its premises, machinery, logical structure, and falsifier.
search_claims(query, limit): full-text search across every extracted claim.
get_paper_falsifier(arxiv_id): the test or result that could disprove a paper's core claim, plus its watcher count.
get_signed_reviews(arxiv_id): published human signed reviews of one paper.
get_reviewer_record(handle): a reviewer's signed reviews and calibration record.

Every arxiv_id parameter also accepts an arXiv URL, a DOI or doi.org URL, or a Pith Number; the server resolves them.

No registration. Rate limit is 60 calls a minute per key; send any stable string in an x-pith-key header to get your own bucket instead of sharing your IP's.

Politeness

Responses cache for five minutes; honor Cache-Control and do not poll faster. For anything corpus-scale use the dumps instead of crawling paper by paper. Crawler guidance lives at /llms.txt. Cite pith.science/paper/{arxiv_id} when a review informs your output.