For lecturers + TAs
Using Policy Window in your AI-governance course
Policy Window is a free, open-licence (CC BY 4.0 content / MIT schema) evidence catalog covering 45 AI-governance instruments across 23topics. This page is the instructor-oriented brief: what works as a classroom resource, what doesn't yet, and a sample reading list / assignments you can drop into a syllabus.
Honest framing: PW is an evidence-catalog reference, not a textbook. Article prose is intentionally thin (a structural choice — see methodology §1); pair PW articles with traditional secondary sources for full-depth reading. Where PW is strongest is structured cross-jurisdiction comparison, per-cell pinpoint citation discipline, and permanent date-pinned URLs (?asOf=) for citation stability.
What PW is well-suited to
- Comparative-method seminars. Compare selected documents and their cited provisions. Distinguish a recorded assessment that a document does not address a topic from a missing catalog entry. Neither establishes an absence of law across a jurisdiction.
- Cite-checking discipline.Every coverage cell carries a pinpoint citation (e.g. "Art. 5(1)(h) prohibition + Art. 26(10) post-hoc rules") so students learn that "EU AIA governs biometric_id" is not a sufficient citation — the article is.
- Reproducible assignment workflow.
?asOf=YYYY-MM-DDpermanent URLs eliminate "the page changed between when I read it and when I cited it" disputes between students and TAs. - Critique-as-pedagogy. The methodology §11 (Limits & blind spots) page explicitly names the catalog's structural biases (instrument-centric ontology, Western coverage asymmetry, no Global-South editorial representation yet). PW is citable as an object of critique.
- Coverage Games as a class exercise. The inter-rater methodology at /wiki/coverage-games is replicable as a 3-week mini-assignment with student raters — see the sample below.
What PW does not replace
- Analytical prose.A typical grad-seminar reading is 3,000–8,000 words; a PW article is ~200–400 words plus a matrix. Pair with traditional secondary sources (Veale & Borgesius 2021, Bommasani et al. 2021, Gebru 2024, etc.) for the "why" behind the "what".
- Authoritative primary text. PW links to primary sources; it does not replicate them. Students still need to read EUR-Lex / regulations.gov for the full statutory text.
- Non-English reading. English-only until funded translation partnerships land. For Chinese / Brazilian / Indian readings, pair with native-language primary sources.
- Independent-editorial-warranty citations. The editorial board is 1 of 6 slots filled (see /wiki/editorial-board). For thesis-defence-grade work, recommend pairing the PW citation with the underlying primary source rather than citing PW as authority.
Sample 3-week reading list
Week 1 — Foundation models / GPAI
- /wiki/foundation_models — topic matrix
- /wiki/eu-ai-act — the most-cited instrument, has Q/A/C populated
- /wiki/compare — try ?a=eu-ai-act&b=us-eo-14110 for the comparative pane
- Pair with: Bommasani et al. 2021, On the Opportunities and Risks of Foundation Models (the technical-debate side PW does not centre)
Week 2 — Biometric ID / law-enforcement carve-outs
- /wiki/biometric_id — topic with EU Art. 5(1)(h) prohibition + Art. 26(10) carve-out pinpoint
- EU AIA + CN GenAI side-by-side via /wiki/compare
- /wiki/for-advocates — read as a meta-text about how PW frames advocacy-relevant topics
- Pair with: Algorithmic Justice League primary report
Week 3 — Silent regimes / Coverage Games
- /wiki/coverage-games + the 2026-q2 case study
- /wiki/methodology §11 "Limits and blind spots"
- Run a mini Coverage Game in class — see assignment 3 below
Sample assignments
- Pinpoint extraction. From
/wiki/eu-ai-act, identify the article that governs deepfakes; then find the China-GenAI counterpart cell; explain why PW labels one "governs" and the other "implicit." (Trivially gradable: TA verifies via the?asOf=YYYY-MM-DDpermanent URL.) - Classification audit. Choose a recorded classification or an unclassified document–topic pair. Read the named document and explain your assessment using its provisions and version. If another document addresses the topic, propose adding it rather than attributing its provisions to the first document. Prepare a source-supported correction or addition for discussion; this exercise does not itself approve publication.
- Mini Coverage Game. Working in pairs, blind-classify 10 cells of /wiki/training_data from primary sources. Compare calls. Write up the disagreement matrix and propose a rubric refinement. (Models methodological-replicability in 3 weeks.)
Citation guidance for students
The recommended citation form:
Policy Window. (YYYY). Article title. Retrieved YYYY-MM-DD from https://policywindow.org/wiki/[slug]?asOf=YYYY-MM-DD
The ?asOf=YYYY-MM-DD parameter pins the article to a specific snapshot stored in ArticleRevision. A green banner on the page confirms the pin; an amber banner indicates no snapshot was captured for that date (request a backfill via the page's "Report a problem" link).
Reference-manager (Zotero / Mendeley / EndNote) import works via the Google Scholar Highwire-Press meta tags embedded on every article page. CSL-JSON is available via the "Cite this" widget on every article. DOIs are on the Q4 2026 roadmap.
What we want from instructor adoption
- Topic-gap reports.If your course covers a topic PW doesn't (state-level employment laws, algorithmic-management in gig work, indigenous data sovereignty), file an issue or open a PR. Student-found gaps are some of the most valuable inputs.
- Coverage Games student data. If you run a mini Coverage Game in class and your students agree to share their classifications, those become public-record inter-rater data that PW will credit. This is the most scalable path to a higher-N reliability estimate.
- Translation contributions. If your cohort includes native speakers of Portuguese / Spanish / Hindi / French / Arabic interested in translating PW articles (with appropriate editorial review), reach out via the wiki home — translation partnerships are the unlock for non-English readership.