OPUS

OPUS is a scientific journal — not a repository, not a preprint server — that runs the entire editorial pipeline as software: from manuscript submission, through agentic AI review and topic-matched human peer review, to a public, citable published record. A journal with the ambition of Nature and the engineering of Stripe.

Two moves define it.

Every paper is a structured, typed object. Claims, evidence pointers, and citations are stored as data alongside the prose — legible to a human reader and addressable as an API (GET /v1/papers) by an agentic one. The same record serves both; the human reading experience stays first-class and machine-readability is strictly additive.

The editorial lifecycle is software. Submission triggers an AI rubric review; each extracted claim is checked by its own verification agent armed with web search and code execution; passing manuscripts are matched to human reviewers; consensus publishes the record. Read-only admin tooling oversees the live pipeline end to end.

The live splash

The embedded preview is the OPUS landing's own hero instrument, recreated self-contained: the canonical record held in orbit, scanned claim by claim. Hover to pause the scan, or pick a signal to inspect it — the same c-0001 … GET /v1 record the platform dissects in full. Built on Next.js 16 / React 19 over Supabase (Postgres with row-level security), with Claude driving claim extraction, per-claim verification, and rubric scoring, and a CodeMirror + KaTeX authoring surface.