This site demonstrates a content-integrity suite native to Drupal — claim-level fact-checking, AI-writing likelihood and verbatim plagiarism search — with all inference running locally on AMD hardware: Google Gemma 4 served by Ollama on ROCm, on the Ryzen AI mini-PC hosting this very site (a 2B model extracts claims, a 4B model judges verdicts).

1. Open the Student essay node and its Content scan tab. Run the scan: claims are extracted and checked against the university's own indexed corpus (the four article nodes). The essay contains three fabricated claims — the scan finds them, with per-claim verdicts and analysis.
2. Browse Trusted sites (Content, filtered by type): the demo imports the full Media Bias/Fact Check dataset — 7,800+ sources with computed reputation and political-bias rating. Open nytimes.com to see the imported assessment side by side with a dated site editorial assessment: the institution, not the vendor, owns the trust decisions. Negative-reputation domains count against claims that only they echo.
3. Every model call is audited at /admin/reports/ai-router-decisions (latest first): local AMD models run at cost 0; complex chat requests can escalate to a premium cloud tier, each decision logged with its estimated cost.
4. Feel free to add other content to factcheck, edit the testing asset or the trusted sites and see the results.

Log in with the credentials provided in the submission (or admin/admin on a local docker compose boot — the whole stack, models included, provisions itself on first run).