flags.page is built on a knowledge infrastructure designed to make every factual claim traceable to a primary source. This page explains how.<\/p>\n
Most flag-content on the open web is recycled from a small handful of secondary sources, with citation styles ranging from “trust me” to “Wikipedia says”. When the underlying sources are wrong, the error propagates indefinitely. We’re trying to do the opposite: hold ourselves accountable to primary sources, log every claim against a verification record, and publish only what we can defend.<\/p>\n
flags.page is an English-language reference site on vexillology (the study of flags). It is owned and operated by an independent editorial team that includes both human editors and AI-assisted research and drafting. We believe AI-augmented editorial work is a legitimate way to produce high-quality reference content \u2014 provided<\/em> the editorial chain is transparent and the fact-checking is rigorous. This page describes that chain.<\/p>\n Behind every flags.page article is a four-layer knowledge stack that determines what gets written and how it is verified.<\/p>\n flags.page maintains structured records for every flag, vexillological term, and regulatory document we cover. As of writing, that is approximately 246 national and supranational flags, 80 glossary terms, and 20 regulatory documents \u2014 and growing. Each entity has its own page (see \/flag\/italy\/<\/a> for an example), with structured fields covering identity, design, history, official protocol, and machine-readable specifications (Pantone, hex, FIAV codes, aspect ratio, design family, color family). These are the nodes of our knowledge graph.<\/p>\n Every substantive factual statement we publish \u2014 a date, a statute name, a numerical fact, an attribution, a direct quote \u2014 is extracted into a structured database alongside its source URL and verification status. Each claim is classified as one of:<\/p>\n When you read a flags.page article, the editorial team has, in principle, verified every load-bearing factual statement against this ledger. When you read it on a flag’s entity page, you see a “Articles citing this entity” box that shows how many claims we have logged for that flag, and how many corrections we have caught and incorporated.<\/p>\n Not all sources are equal. We classify each source we cite into one of these reliability classes:<\/p>\n The source registry is consulted before we cite. A claim sourced exclusively to a “secondary-cross-check” entry is, by policy, downgraded to “questionable” unless a primary-authoritative source can be added.<\/p>\n Every editorial action \u2014 proposing a topic, drafting an article, fact-checking it, publishing it \u2014 leaves a record. Every AI-assisted drafting run is logged with the prompt that produced it, the model version, the sources retrieved, and the resulting output. This makes it possible, weeks or years later, to answer the question: “where did this specific sentence come from, and what was the verification state at the time of publication?”<\/p>\n We publish corrections openly. Every article logs the claims that were initially mis-stated by the drafting agent and corrected during fact-check. Each entity page surfaces the number of corrections logged for that entity. We treat this as a feature, not an embarrassment: it shows the system is working as designed.<\/p>\n We also maintain a public error-pattern ledger that documents the recurring failure modes our drafting process exhibits \u2014 frequently fabricated court-case citations, miscounted statistical tables, conflated statute clauses, absolutist phrasing where exceptions exist \u2014 alongside the countermeasures we have applied. Each new article passes through these countermeasures before reaching human review.<\/p>\n We cannot guarantee accuracy. AI drafting agents hallucinate. Sources change. Human reviewers miss things. What we can do is make the chain of decisions auditable and the verification status visible. If you find a factual error in a flags.page article, write to matteo@kezers.com<\/a> with the article URL, the specific claim, and the primary source that contradicts it. We will update the claim ledger, correct the article, and credit the correction at the foot of the piece.<\/p>\n We also do not yet do every check that a SOTA newsroom would. We do not yet automatically embed-on-publish, our drift checker runs only weekly, and we have not yet built a style-consistency layer that compares each new draft to a published-corpus voice baseline. These are on our roadmap.<\/p>\n The byline on most flags.page articles is Luca Sereno<\/em>, who is the editorial pseudonym for our AI-assisted research and drafting pipeline. Articles signed by Luca Sereno have been: (a) drafted by an AI agent using the process described above; (b) fact-checked by a separate AI agent in adversarial mode; (c) reviewed and edited by a human editor before publication. Articles bylined to a different name are written primarily by a named human author.<\/p>\n We are transparent about this because we believe AI-augmented editorial work is a legitimate practice, but only when it is disclosed. If you would rather not read AI-assisted content, we invite you to find another source.<\/p>\n Editorial contact: matteo@kezers.com<\/a>. flags.page is built on a knowledge infrastructure designed to make every factual claim traceable to a primary source. This page explains how. Why this page exists Most flag-content on the open web is recycled from a small handful of secondary sources, with citation styles ranging from “trust me” to “Wikipedia says”. When the underlying sources … Weiterlesen<\/a><\/p>","protected":false},"author":3,"featured_media":0,"parent":21,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"ai_generated_summary":"","footnotes":""},"class_list":["post-735","page","type-page","status-publish"],"_links":{"self":[{"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/pages\/735","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/comments?post=735"}],"version-history":[{"count":0,"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/pages\/735\/revisions"}],"up":[{"embeddable":true,"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/pages\/21"}],"wp:attachment":[{"href":"https:\/\/flags.page\/de\/wp-json\/wp\/v2\/media?parent=735"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}The four-layer architecture<\/h2>\n
1. The entity graph<\/h3>\n
2. The claim ledger<\/h3>\n
\n
3. The source registry<\/h3>\n
\n
4. The audit trail<\/h3>\n
The editorial process, in seven steps<\/h2>\n
\n
What we know we get wrong<\/h2>\n
Limits of this approach<\/h2>\n
Authorship and AI disclosure<\/h2>\n
Contact and corrections<\/h2>\n
\nSource corrections: include the article URL, the specific claim, and a primary-authoritative source that contradicts it.
\nThis page is itself a living document and will be updated as the methodology evolves. Last revised: 30 May 2026.<\/p>","protected":false},"excerpt":{"rendered":"