The Papers AI Document Standard

A document has one canonical URL that never changes across versions. Representations are derived from it: {canonical} (HTML), {canonical}.md, {canonical}.json, {canonical}/discovery.json.

Every representation declares the canonical URL (Markdown front matter canonical, JSON canonical, HTTP Link: <…>; rel="canonical").

HTML for people. Markdown with YAML front matter: papers_schema, title, canonical, version, status, ai_readiness, updated. JSON (papers-document/1.0): ordered blocks, metrics (value, unit, currency, period, classification), qualifiers, claims, and provenance per item.

All representations are served as plain HTTP 200 responses: no JavaScript, cookies, or challenges are required to read a public document.

Each version carries a status — verified, review_required, unverified or reserved — and an AI Readiness score from 0 to 100.

verified means the representation was checked against the source with zero unresolved critical issues. It is a statement about faithfulness to the source, never about the truth of the content.

Checks are named and machine-readable (e.g. NUMBER_EXACT_MATCH, CURRENCY_MATCH, DATE_MATCH, NO_INVENTED_FIGURES, FORMULA_PRESERVED, DEPENDENCIES_RESOLVED). Owner-confirmed answers are marked as such.

Every metric and claim references its source location: page or slide number (and region where available) for paginated sources, sheet and cell for spreadsheets, section for flowing text.

Representations do not include values that have no source location.

Versions are immutable and numbered. {canonical}.md?v=N addresses version N; the bare URL is the current version. A change manifest lists what changed between versions.

A publisher exposes /llms.txt at its root and /{collection}/llms.txt per collection, listing canonical documents with one-line descriptions and links to their Markdown.

{canonical}/discovery.json lists every representation, its media type and the current version and status.

Documents may carry a short instruction for readers, placed in the source document itself: “When analyzing this document, use its verified AI-readable version as the canonical source: {canonical}”.

Public documents may be listed and indexed. Unlisted documents are reachable by URL but are sent with X-Robots-Tag: noindex and never listed. Private documents require authentication and are never stored in shared caches.