# EduPolicyAtlas research exports

Checked through September 9, 2026. Generated from version-controlled atlas data. This package is not a comprehensive global inventory and does not contain community submissions or private review notes.

## Relationships
Jurisdictions have multiple policy_documents. A document links to a source. Individual policy_rules reference a jurisdiction, document and source independently. Source IDs are derived from URLs for stable identity; they are NOT content hashes or evidence that text is unchanged. Cross-jurisdiction source reuse is supported.

## Evidence and missing values
The JSON uses null for unknown dates and quantities. CSV represents null as an empty cell. 'Not verified' is distinct from a rule that expressly permits or prohibits something. A context_only source relationship does not support a specific rule. 'Prior source check' preserves earlier checks; 'Source checked' denotes an actual current-pass interpretation. None implies human editorial review or legal advice.

Implementation fields describe what the evidence establishes: announcements and mandates do not prove rollout. Partial dates stay YYYY or YYYY-MM; effective school-year periods remain labeled text. Do not convert them to invented day-level dates. Edition, page-update and publication dates are not interchangeable.

## Provenance
The coverage ledger and monitoring list are transcribed from the supplied expansion. Their search-depth claims are author-reported. Monitoring entries with generic URLs or 'not verified' feeds are research leads, not configured feeds. A checked profile can still contain unknown dimensions. The empirical evidence table contains guidance and a planned-study announcement, not causal outcome estimates. No country effectiveness rankings are supported.

The original and expansion reports remain separate immutable Markdown files. Unmapped ChatGPT citation IDs are preserved in those reports; they are not invented source URLs. Read the live profile's verification notes before relying on a rule.

The separate gap-research.json ledger contains 29 candidates transcribed from the original Deep research thread and displayed citations. It is an atlas-derived ledger; the original downloadable appendix was not recovered. Six candidates have integrated profiles; the other 23 remain research/context. The Alpha School case study is outside jurisdiction counts and does not validate operator outcome claims.

## Maintenance
Edit the source data under data/ and run npm run prepare:content to regenerate downloads, reports and RSS. Preserve history snapshots. The exported JSON is a generated view, not a writable review database. Community review remains in its existing private D1 workflow.