How records become source-backed claims

1. Source discovery

   Public university policy pages, teaching guidance, IT/security pages, procurement pages, PDFs, and source-attributed materials.

2. Crawl and snapshot

   Fetched source text receives a content hash for future change checks.

3. Claim extraction

   Claims keep confidence separate from review state.

4. Evidence binding

   Every public claim needs a source URL, snapshot hash, evidence snippet, source language, and rights caveat.

5. Change detection

   Changed hashes can create new snapshots, candidates, claim updates, and change records.

6. Multilingual source-first evidence

   Original-language evidence is canonical; translations are display helpers.

Claude, Codex and other AI tools support product development, official-source discovery and review, structured evidence extraction, translation drafts and policy audits. Our workflow has used Claude Code and Antigravity; tools vary by task.

HTTP fetches, browser tools and Firecrawl collect source material. Agents help organize and review it. A generated answer is not a university policy: published claims must remain tied to original quotations, source URLs, snapshot hashes and explicit review states.

Selected repository history

• Source checks and evidence re-sourcing
• Live-fetch and evidence audit tools

• QS 2026 currently remains the main crawl batching source for expanding public coverage.
• THE 2026, ARWU 2025, U.S. News 2025-2026, and CWTS Leiden 2025 are supported as ranking, index, and filter sources.
• CWTS Leiden 2025 is a derived metric order, not an overall global university rank.
• Different ranking years are not presented as one unified 2026 ranking.
• Ranking data does not create or override policy claims; visible claims still require official sources, evidence snippets, confidence, and review state.

• Every public claim needs a source URL and source snapshot hash.
• Evidence snippets stay short, necessary, and source-attributed.
• Original-language evidence is canonical; translation is display-only.
• Canonical facts and localized display must remain separate.
• Candidate claims must be labeled and must not be used as final summaries.
• Public pages and public JSON should be generated from the same promoted public release dataset.
• Official source documents retain their original rights and terms.

• University policies can be distributed across departments, courses, PDFs, and internal systems.
• Sparse records remain labeled as early coverage until reviewed evidence supports stronger conclusions.
• The tracker records source-check and claim-evidence metadata; it does not provide legal advice or academic integrity advice.

This tracker is not legal advice, not academic integrity advice, and not an official university statement unless a linked source is the university's own official page.