About

Sources and licences

A source’s licence is a check the code makes, not a paragraph in a policy. Where a source permits facts but not prose, this site can render its dates and names and cannot render its sentences.

The distinction that governs everything here

UN documents and official records — resolutions, meeting records, verbatim records, and public information material not offered for sale — are in the public domain worldwide under administrative instruction ST/AI/189/Add.9/Rev.2 of 17 September 1987. They can be reproduced freely.

The text of un.org itself is not. Its terms reserve all rights and require permission in writing. We link to it and cite facts from it; we never mirror it.

Notes on each

Certificate Transparency logs

Subdomain discovery for the un.org family. CT is public by construction -- the logs exist to be read -- and observing it is passive. Carry util-ct's bound honestly: 'first seen by us' is a true statement, 'first issued' is not.

European Parliament Open Data Portal

The licence took establishing and the route matters. The Parliament's own legal notice at www.europarl.europa.eu/legal-notice/en/ answers 202 with a zero-byte body to a plain HTTP client -- from the cluster and from a domestic connection alike, so it is not an egress block, it simply does not serve to a machine. data.europarl.europa.eu/en/legal-notice is a JavaScript shell with no text in it either. The answer came from the route built for machines. The EU open data portal lists the dataset 'Members of the European Parliament (MEPs)', publisher European Parliament, with licence COM_REUSE. That code resolves at publications.europa.eu/resource/authority/licence/COM_REUSE to the European Commission reuse notice: 'reuse is authorised, provided the source is acknowledged', implementing the Decision of 12 December 2011 (2011/833/EU). A first-party, machine-readable declaration attached to this exact dataset, which is a better answer than the prose page would have been. Attribution is a condition, not a courtesy: every page built from this names the European Parliament.

HTTP response headers

CSP, HSTS, referrer policy, permissions policy, cookie flags, server banner, and the TLS parameters of the connection that carried them. ONE ordinary GET per host per cycle, honouring robots.txt and the stated Crawl-delay. That is a page view, not a scan, and the difference is the request count.

Machine-written summary of public-domain sources

Grounded only in sources whose own licence permits reproduction -- in practice UN documents under ST/AI/189. A summary of an all-rights-reserved article is a derivative of that article, so those are never passed to the model.

Permanent Missions to the United Nations (the Blue Book)

The authoritative roster of accredited diplomats. Blue Book Online at bluebook.e-delegate.un.org is an Angular SPA whose API host (edelegate.unmeetings.org) does not resolve publicly and whose own paths answer 403 -- it is NOT a public data source and must not be scraped around. The usable path is the published PDF series ST/PLS/SER.A/*, which is an official document and therefore public domain. Spouses are listed in the source and are deliberately not ingested: they are private individuals who hold no office.

Public DNS

A, AAAA, MX, NS, TXT (SPF), _dmarc, DNSSEC. Reading published DNS records is not scanning: the records are published to be resolved.

Public object-store index

Lookup against a third-party index of buckets that are ALREADY public. Passive: we query an index, we never enumerate a bucket and never fetch an object. Facts recorded are the bucket name, the index's own counts, and first/last seen. The index itself is not redistributed.

Published image metadata

EXIF/XMP/IPTC read from images the UN has itself published. The metadata is measured and reported; the image is not redistributed, because UN Photo is separately rights-reserved and sits outside ST/AI/189/Add.9/Rev.2. Store the fields, never the file.

SERP API (Serper)

Discovery only: which UN hosts and pages exist, and which questions people actually ask. Result snippets are the search engine's, not ours, and are never rendered -- they are used to decide what to fetch from the primary source.

Third-party page found by search

Discovered via the SERP source. Fetched to record that the page exists, what it is titled and when we saw it; the body is hashed and measured, never stored. Promoting a domain to a source with reproduce_text=True is a deliberate act requiring an actual licence, one domain at a time.

UN General Assembly ideal point estimates

Estimated positions on the dominant dimension of General Assembly voting, per state per session, 1946 onward, with credible intervals. Two requests: the dataset listing, then the estimates file resolved from it by name -- Dataverse mints a new file id on every deposit, so pinning one would break at the next release. The raw roll-call record is not published here; searched 2026-09-03 and this dataset carries only derived estimates.

UN General Debate Corpus (Jankin, Baturo & Dasandi)

Two free layers, which is rare: CC0 on the compilation, and the statements are UN official records that are public domain under ST/AI/189 anyway. So the text may be reproduced in full and is the only large body the summariser may be given without the licence gate having to be argued. NOT uniform, and the corpus says so: sessions 1-78 are official UN transcripts, session 79 is delegation PDFs with OCR and GPT-4o translation, and session 80 is Whisper transcription of the simultaneous interpretation because the UN published no validated transcripts for 2025. `text_provenance` carries the tier onto every row and every page.

UN Meetings Coverage and Press Releases

Public information material informing the public about UN activities and not offered for sale, so it falls inside ST/AI/189/Add.9/Rev.2. Meeting summaries name who spoke and what they said, which is what makes a diplomat page more than a directory entry.

UN Peace and Security Data Hub

DPPA/DPO datasets behind the SPA at psdata.un.org, served from api.psdata.un.org/public/data/<ID>/<format> with metadata at /public/metadata/<ID>. Reachable from the cluster -- measured 2026-08-31, unlike www.un.org and peacekeeping.un.org, whose CloudFront blocks datacenter ranges (peacekeeping.un.org 403s its own robots.txt even from a domestic line, which is why this hub is the route). Whole-dataset downloads: vetoes ~136 KB, uniformed-personnel contributions ~45 MB -- one GET each, written as full snapshots, never crawled.

UN Research Guides (Dag Hammarskjold Library)

The membership record: each member state's admission resolution, the original 51 with their signature and ratification dates, and elected terms on the Security Council, ECOSOC and the Human Rights Council. Facts only -- the guide's own prose is not reproduced. robots.txt names 24 crawlers it refuses, all of them bulk AI and search crawlers; we are none of those, we identify ourselves, we honour the stated delay, and the whole role is seven requests.

UN Statistics Division SDG Indicators API

Goals, targets, indicators, series and per-country observations.

UN World Population Prospects

The published bulk files, located through the manifest the downloads page itself loads (wpp/assets/downloads.json) -- the portal's /data API answers 401 to every request however small, measured 2026-09-01 with one location and one year, so the open metadata routes exist but the data rides the ~17 MB TotalPopulationBySex CSV instead: two requests, whole dataset, the psdata shape. PopTotal is in thousands; a floor demanding a state over a billion people is what makes that scaling a checked fact. Medium variant, five-year steps.

UNHCR Refugee Data Finder

Population totals at api.unhcr.org/population/v1/: refugees, asylum-seekers, IDPs and stateless people by country of asylum and of origin, per year. Reachable from the cluster -- measured 2026-09-01, plain nginx, unlike www.unhcr.org, which Cloudflare 403s from datacenter ranges. Responses carry ISO3, so rows join the member-state spine without name matching.

United Nations Digital Library

Bibliographic spine: resolutions, meeting records, voting records, speeches. Enumerate via sitemap_index.xml.gz (142 sub-sitemaps, ~23.7k URLs each). Per-record MARCXML at /record/<id>?of=xm. The OAI-PMH endpoint at /oai2d answers Identify/ListSets/ListMetadataFormats but throttles ListIdentifiers and ListRecords with a bare 503 'Retry after 2 seconds' -- treat OAI as a supplement, never as the primary path.

United Nations Security Council Consolidated List

One XML document, regenerated daily, served via a redirect to an Azure blob with a short-lived SAS token -- follow redirects and do not cache the signed URL. Names living people under sanction: it is public by design and by Security Council decision, but it is still the sharpest personal data on the site. Never join it to any other person-shaped dataset.

United Nations Treaty Collection

MTDSG status tables: per treaty, per party, signature and ratification dates and declarations. Slow to respond and has timed out on first contact -- budget generous timeouts and retry rather than raising the rate.

United Nations website

Structure, navigation, entity names, office holders, dates. Never mirror page copy. Crawl-delay 10 is stated in robots.txt and is honoured as a commitment.

Wikidata

The entity spine: member states, organs, agencies, office holders and their term dates, with stable QIDs to resolve against. Coverage of permanent representatives is real but patchy -- 553 people across 111 country-specific positions, measured 2026-08-25 -- so it corroborates the Blue Book, it does not replace it.

Wikimedia Commons

Images only, and every file is judged on its own metadata. Two independent gates, because copyright status and permission to use are different questions: `Copyrighted`/`License` decide the first, `Restrictions` the second. Measured 2026-08-26, every national flag on Commons carries Restrictions: insignia while being public domain for copyright -- and so does the UN flag, which GA resolution 92(I) and Article 6ter of the Paris Convention put out of reach whatever its copyright status.

Wikipedia

Share-alike is contagious: any page that reproduces Wikipedia prose verbatim must itself carry CC BY-SA 4.0. The site therefore uses Wikipedia for the lead-sentence summary only, clearly delimited and licensed in place, and never blends it into first-party prose.

World Bank Open Data

Verified 2026-08-26 at datacatalog.worldbank.org/public-licenses: CC BY 4.0 is the default for datasets the World Bank produces, with additional mandatory terms covering DISPUTE RESOLUTION -- mediation then UNCITRAL arbitration -- and not use. Some datasets carry other licences and are labelled; anything added here must be checked rather than assumed to inherit the default. api.worldbank.org publishes no robots.txt (404, so no stated restrictions under RFC 9309) and is built for programmatic use, so the rate is our courtesy rather than a stated limit. Note that many WDI series are compiled from elsewhere -- the population figures come from the UN Population Division -- and source_organization is stored whole so a page can say so.

worldbody.org (derived from the corpus)

Counts and joins computed from tables already collected -- the General Debate concordance is the first. Makes no request. What is derived is ours; what it was derived FROM keeps its own licence and its own citation, so a page built on the debate corpus still credits the corpus.