University AI Policy Tracker
An open, evidence-backed database and public infrastructure layer for tracking how universities publish and change AI policy and guidance.
Live site:
Project and operator
University AI Policy Tracker is a Canadian public-interest EdTech initiative operated by an independent Canadian company. Founded in May 2026, the project operates on a public-interest, non-commercial basis. Founder and operator: Sam Song. Contact: [email protected].
Overview
The project turns fragmented official university pages into source-attributed, citation-ready records. It is designed for researchers, universities, maintainers, and agent or data integrations that need a durable public record, not a login-only SaaS product.
Public Infrastructure And Impact
The following signals are intentionally tied to public, reproducible sources:
- Public release scale: As of the
public-release-20260801-001release
- Distribution: As of 2026-08-08, the public API index lists 43 versioned
- Maintenance: The repository has 317 commits through 2026-08-01. The
- Evidence discipline: Public records retain official URLs, source
Production deployment and maintainer-only operational procedures live in the
runbooks under docs/, while this README focuses on the public
contract, contribution path, and evidence boundary.
Current Status
The project is now a public beta of an evidence-backed policy database, not just a link directory. It includes crawlable university reference pages, versioned public JSON, source-backed claim records, policy analysis pages, coverage dashboards, review workflow surfaces, feeds, widgets, and MCP-oriented metadata for agent integrations.
Current public release:
- Release:
public-release-20260801-001(published 2026-08-01) - Universities: 828
- Claims: 5476
- Evidence records: 5741
- Official source attributions: 3195
- Entity review states: 827
agent_reviewed, 1needs_review - Claim review states: 5472
agent_reviewed, 4needs_review - Source languages represented: 61 language tags across the evidence set
pnpm readme:update-status from the release manifest and staged merge.
The public release has no current audit issues from pnpm audit:public-data.
The data is source-backed and review-labeled, but it is not legal advice,
academic integrity advice, or an official university statement unless the
linked source is the university's own official page.
Implemented Capabilities
Public reference site:
- GitHub-style public evidence database interface
- SSR/SSG-friendly university detail pages
- claim/evidence cards with review state, confidence, source URL, source
- citation copy actions and citation guidance
- dark mode tokens and initial i18n scaffolding
- homepage, university index, methodology, citation, datasets, changes,
- project background and agent working methods at About
- evidence-linked student policy pages with a separate tool-access module and
Policy analysis layer:
- policy analysis contract and validator
- static analysis builder over public claim/evidence records
- analysis index and theme pages
- per-university analysis JSON
- coverage score metadata
- page-quality and review workflow surfaces
- explicit boundary that analysis summaries are not final policy conclusions
- independent
uapt-policy-snapshot-v1schemas and per-university authoring
- fail-closed dual-agent review and claim/source/release fingerprint checks
- versioned snapshot index and university JSON endpoints (Wave 1 index is
Data distribution:
- versioned public API under
/api/public/v1/... - JSONL dataset downloads
- dataset release manifest and checksums
- public search index and entity index
- RSS/Atom feeds for changes and reports
sitemap.xml,robots.txt, andllms.txt- Dataset JSON-LD and citation metadata
- read-only MCP manifest and tool catalog
- embeddable widget JSON endpoints
- widget embed script
- OpenClaw staged artifact contract
- dataset release validator
- public contract validator
- OpenClaw artifact validator
- local review cache and knowledge-review indexes
- private mirrored analytics dashboard at
/internal/analytics, protected by basic auth and backed by first-party event storage
- QS 2026 coverage dashboard
- source health dashboard
- Firecrawl source-health verification metadata for blocked or inconclusive
- review queue page and JSON
- staging-vs-public release knowledge summaries
- contribution intake surfaces for missing policy URLs, policy changes,
Public Surfaces
Core trust pages:
- About:
- Methodology:
- Citation:
- Datasets:
- Changes:
- Search:
- Analysis:
- Coverage:
- Source health:
- Contribute:
- Review workflow:
- Widgets:
- MCP alpha:
- API index:
- Universities list:
- Recent changes:
- Per-university records:
https://eduaipolicy.org/api/public/v1/universities/{slug}.json - Per-university claims:
https://eduaipolicy.org/api/public/v1/claims/{slug}.json - Search API:
- Search index:
- Entity index:
- Analysis index:
- Student policy snapshot index:
- University student policy snapshot:
https://eduaipolicy.org/api/public/v1/policy-snapshots/universities/{slug}.json - QS coverage:
- Source health:
- Review queue:
- Dataset release manifest:
- Reports index:
- Reports outreach package:
- May 2026 monthly report chart data:
- Contribution index:
- Contribution review policy:
- MCP manifest:
- MCP tool catalog:
- Bulk downloads:
Data Model Principles
This project separates four layers:
- Evidence layer: source snapshots, source URLs, source language, hashes, evidence snippets, and claim extraction.
- Reference layer: crawlable university pages, claim/evidence cards, citation blocks, and change logs.
- Distribution layer: public JSON, sitemap,
llms.txt, dataset documentation, and repository trust assets. - Contribution and review layer: staged OpenClaw artifacts, pull requests, review states, and institution correction workflows.
Canonical facts and localized display are separate. Original-language evidence is the canonical evidence record. Translations and display summaries may help readers, but they must not replace source evidence.
Confidence and review state are separate:
confidence: machine confidence in an extracted or normalized claim.reviewState: workflow status such asmachine_candidate,agent_reviewed,human_reviewed,needs_review, orrejected.
Repository Layout
apps/web: Next.js public site and public JSON route handlers.apps/api: Node/Nest API surface for server-side reads and ingestion-adjacent work.apps/worker: worker package placeholder for later scheduled jobs.packages/shared: shared schemas, public contract types, seed data, and OpenClaw artifact contracts.packages/db: Prisma schema, migrations, and database helpers.packages/crawler-core: crawler and diff utilities.data/public-releases: curated release manifests for public staged data.data/rankings: university ranking source metadata used by the public index.staging/uapt-runs: staged OpenClaw or curated artifact directories.docs: architecture, crawler policy, data contract, and roadmap documents.docs/dataset-release-process.md: dataset release checklist, artifact
docs/report-distribution-playbook.md: report, feed, outreach, and media
knowledge: local retrieval summaries for public release, staging runs,
scripts: validators, audits, smoke tests, conversion tools, and local
OpenClaw Boundary
OpenClaw may help with crawl planning, source discovery, snapshot generation, claim extraction, evidence binding, review assistance, report writing, and PR generation.
OpenClaw must not:
- write the production database directly
- publish canonical claims without review
- push
main - deploy the public website
- bypass robots.txt, login walls, paywalls, or access controls
- mark its own output as
human_reviewed
docs/openclaw-data-prs.md and docs/agent-workflow.md.
Verification
The main local quality gate is:
pnpm check
It runs database client generation, public API contract validation, analysis contract validation, policy analysis audit, analysis page smoke checks, entity search smoke checks, OpenClaw artifact validation, dataset release validation, public data audit, typecheck, and lint.
Recent production verification covered:
https://eduaipolicy.org/coverage/coverage/qs-2026/search?q=MIT/api/public/v1/search.json?q=MIT/api/public/v1/search.json?q=ANU/api/public/v1/search/index.json/api/public/v1/entities/index.json/api/public/v1/universities.json/api/public/v1/claims/anu.json- widget endpoints for review state, source freshness, and policy coverage
/sitemap.xml,/llms.txt, and/robots.txt- the private
/internal/analyticssurface is not part of the public API
Local Development
Requirements:
- Node.js 20+
- pnpm 10
pnpm install
pnpm check
pnpm validate:dataset-release
pnpm --filter @uapt/web build
git diff --check
Run the web app:
pnpm dev:web
Citation
Use the canonical page and versioned JSON together. Retain official source URLs, source language, snapshot hash, review state, confidence, and original evidence snippets when reusing claim-level data.
Machine-readable citation metadata is in CITATION.cff.
License And Source Rights
- Original software and code authored for this repository are licensed under
- Tracker-created metadata and dataset outputs are licensed under CC BY 4.0
- University source pages, PDFs, screenshots, excerpts, logos, and other
Policy evidence maintenance
Crawl, audit, repair and indexing tasks follow the versioned requirements and policy evidence integration runbook. OpenClaw scheduled maintenance uses the same evidence boundaries and synchronization requirements. Private inventories, raw responses and audit bundles stay outside published data.