For the complete documentation index, see llms.txt.
Skip to main content

Methodology

This page describes the process behind every substantive page on Model Diplomat — from research brief to published page to scheduled refresh. It is meant to be auditable: if a claim on our site is wrong, this document tells you how it got there and what we will do about it.

Last updated: May 1, 2026

1. Topic selection

Topics are drawn from the twelve domains that make up our curriculum spine (Competitions, Political Systems, Global Affairs, Law & Rights, Economics, Media, History, Leaders, Rhetoric, Regions, Religion, Tech & Power). Within each domain, we prioritize topics with documented reader demand — search-trend data, our own query logs, and the syllabi of major IR programs.

2. Research

Each page begins with a sourced research brief: primary documents (treaties, court decisions, official statistics, government publications), peer-reviewed scholarship where available, and reputable secondary analysis for context. Atlas, our AI assistant, is used to surface candidate sources; every source that appears in the published page has been opened and read by a human.

3. Drafting

Drafts are produced in one of three ways, disclosed per page: (a) human-only, (b) AI-assisted with a named human editor, or (c) AI-drafted with a named human editor and a named subject-matter reviewer. AI-drafted pages on YMYL topics (geopolitics, civic institutions, elections, government trust) do not ship without a human edit ratio of at least 15% and a subject-matter reviewer.

4. Expert review

Pages on the eleven substantive pillars are reviewed by a named subject-matter expert before publication. The reviewer's name, credentials, and review date appear on the page. We track the diff between the AI draft and the reviewed version; a reviewer who consistently ships unedited drafts is removed from the rotation.

5. Citation density

Target: one citation per 150–200 words on YMYL pages, with primary sources preferred. Inline statistics, named entities, and direct quotations are all sourced. We do not use unsourced "according to experts" framings.

6. Publication and indexing

Pages are published with structured data (Article, Person, Organization, BreadcrumbList, and where applicable FAQPage, Course, Quiz, Event) and a canonical URL. The datePublished reflects the original publication date; the dateModified reflects only substantive editorial passes (≥500 new words, statistic refreshes, or claim corrections), never automated bumps.

7. Refresh cadence

Tier-1 pillar pages are refreshed every 60–90 days. Evergreen pages are refreshed quarterly. Daily briefings carry a visible data cut-off timestamp. When a page is materially changed, a brief changelog entry appears at the bottom.

8. Corrections

See /corrections for the public log. See /editorial-policy for what we count as a correction versus a typo.