What review controls
Review gates whether submitted sources, corrections, translations, and analysis metadata can affect public tracker records.
Review workflow
Contribution review before publication
Review turns submissions into source-checked tasks before anything can affect public claim/evidence records.
What remains non-final
Machine-candidate and needs-review records remain visible workflow states, not final policy conclusions.
What contributors submit
Contributors submit official source URLs, corrections, and evidence context; publication still requires validation and review.
Analysis profile review
Page quality does not approve derived analysis.
Analysis profiles remain machine_candidate until reviewers confirm evidence binding, not-mentioned reasoning, source-language preservation, page quality gates, and no-advice boundaries.
Server-rendered public content
Analysis conclusions, caveats, review state, evidence counts, and public JSON links must be visible in HTML.
Theme evidence threshold
Theme analysis pages are generated only when at least five records have source-backed evidence for the dimension.
Review state boundary
Review state must be visible and must remain separate from confidence.
Original-language evidence
Source-language evidence remains canonical; localized display cannot replace it.
Coverage score caveat
Coverage scores must be described only as breadth of public source-backed coverage, not quality, compliance, legality, safety, or ranking.
No-advice boundary
Pages must state that tracker analysis is not legal advice, academic integrity advice, or an official university statement.
Versioned public JSON
Public analysis metadata must use /api/public/v1/... endpoints.
Reviewer checklist
- Confirm every non-empty analysis dimension links to public claim IDs and source URLs.
- Confirm not_mentioned means absence of current tracker evidence, not permission, prohibition, or proof that no policy exists.
- Confirm original-language evidence and source language are retained in the basis records.
- Confirm confidence remains secondary to review state.
- Confirm coverage-score text avoids quality, safety, compliance, legal adequacy, or ranking language.
- Confirm public JSON URLs are versioned under /api/public/v1/.
- Confirm the page includes no legal advice and no academic integrity advice.
Review queues
Each queue has its own publication gate.
Source discovery review
Verify that suggested URLs are public, attributable, source-language labeled, and relevant to university AI policy.
A source can be staged only after reviewers confirm provenance, rights caveats, source language, and crawlability.
Crawl failure review
Separate inaccessible, blocked, redirected, no-policy, and weak-source cases before extraction.
Failure records can be published only as labeled status metadata, not as policy conclusions.
Claim/evidence review
Check whether a proposed claim is supported by short original-language evidence and source attribution.
Claims remain candidate records until evidence, confidence, review state, and citation fields pass review.
Analysis profile review
Review deterministic policy dimensions, coverage-score caveats, not-mentioned reasoning, basis claim IDs, and page-quality gates before analysis metadata graduates beyond machine-candidate status.
Analysis profiles remain machine_candidate until reviewers confirm source-backed dimensions, original-language evidence, review-state separation, and no-advice boundaries.
Translation review
Review localized display summaries without replacing source-language evidence.
Translation changes can affect helper display only; original evidence remains canonical.
Institution correction review
Handle official corrections, metadata fixes, attribution disputes, and canonical page adjustments.
Corrections must preserve audit history and cite the official or attributable evidence used.
Course submission review
Moderate course-level AI policy evidence, privacy concerns, copyright limits, and source context.
Course records reuse claim/evidence and remain pending until moderation and rights checks pass.
Abuse and moderation review
Reject personal attacks, private data, doxxing, unsupported accusations, and full copyrighted materials.
Rejected or unsafe submissions are not converted into public facts.
Safeguards
Issue intake with publication safeguards.
Privacy
- Do not submit private student information.
- Do not submit non-public instructor personal data.
- Course submissions should include only public course context or short evidence excerpts needed for review.
- Institutional role claims may be noted, but public review still requires source evidence.
Copyright
- Do not paste full copyrighted syllabi, PDFs, LMS pages, or source pages.
- Use short original-language excerpts only when necessary for evidence review.
- Tracker metadata can be open licensed, but source documents retain their original rights.
- Raw source materials should not be added to Git unless explicitly approved.
Moderation
- No doxxing, harassment, personal attacks, or unsupported accusations.
- No requests for legal advice or academic integrity advice.
- No login-wall, paywall, robots, or access-control bypass requests.
- Candidate records must stay labeled until reviewed.
Publication gates
Review state and confidence remain separate.
- Submission creates a review task, not a canonical fact.
- Source review checks officialness, language, access, and rights.
- Claim/evidence review checks short original-language evidence and attribution.
- Course submissions pass moderation before any course entity can be published.
- Institution corrections preserve audit history and cite supporting sources.
- Only reviewed outputs can graduate into public claim/evidence records.
Review policy JSON
Read-only metadata for contributors and agents.
Contribution review policy
Read-only contribution review policy, queue definitions, safeguards, and publication gates.
Review queue
Unpromoted staging run queue with validation status, source breadth, detected slugs, and recommended next action.