Methodology & Sources
Modern Politics is a solo-maintained, AI-assisted project. This page describes the process that actually produces the site, states how much of the site that process has covered so far, and lists what it does not do. The figures below were counted from the project’s own files on the date given; the repository is not yet open for you to recount them yourself, which is noted among the limitations.
The site has two kinds of content, and they are held to different standards today. The narrative — the long-form story and the statistics pages — is written from named sources. The encyclopedia — the entity and timeline pages that make up the bulk of the site — is not yet systematically cited. Those are stated separately below rather than averaged into a claim that would be true of neither.
Who Makes This
One person: Germán Greiner. See Contributors. There is no editorial board, no fact-checking desk, and no legal review. Drafting and review are performed with AI assistance, orchestrated from a set of agent and skill definitions kept in the repository under apps/agents/ and .claude/skills/.
How a Page Gets Made
- Discovery — Candidate topics come from coverage analysis against the existing knowledge base (
modpol coverage,modpol status,modpol inventory) combined with web search for recent events. There is no archival crawl, no library-database subscription, and no FOIA pipeline. - Drafting — A page is drafted by an AI agent working from web sources, following the editorial standards in the repository’s content writing guide: neutral tone, no political labeling, no editorial commentary, and explicit date precision (a year-only date is published as a year, not as January 1st).
- Review — The draft goes through an automated multi-agent review described in the next section. Findings are either fixed or explicitly adjudicated; the outcome is recorded per file.
- Publication — Content is committed to a Git repository, so every change to every page is recorded in that history. Notable updates are summarized on the Blog / Updates page.
What the Automated Review Checks
Each file under review is read in parallel by five specialized checkers:
- Relevance — does this page belong on the site, and is it in the right collection?
- Bias — does the text favor a political side?
- Factual accuracy — are dates, numbers, names, and causal claims plausible and precise?
- Tone — is there editorializing, exaggeration, or advocacy?
- Labeling — does the text apply political labels rather than describing actions?
What the review does not do, and this is the important limitation: the checkers read only the repository. They have no web access and cannot fetch a cited source. A “factual accuracy” pass flags claims that appear wrong or unsupported against the reviewing model’s own knowledge — it does not verify the page against the documents the page cites, because for most pages there are no cited documents to verify against. Review is a consistency and plausibility check, not source verification.
What the Review Badge Means
Entity pages, timeline events, and story sections carry a badge reporting one thing only: how far that file has got through the automated review described above. There are four states.
| Badge | Meaning |
|---|---|
| Reviewed | The file completed the automated review with no unresolved findings, and has not been edited since. The tooltip gives the date of that review. |
| Findings open | The file was reviewed and at least one finding is still unresolved. |
| Needs re-review | The file was reviewed, but its text has changed since. The recorded result describes wording that is no longer on the page, so it is no longer reported as reviewed. |
| Not yet reviewed | The file has not been through the review pipeline. This is a statement about coverage, not a defect in the page. |
Needs re-review exists because a review result is only true of the exact text that was reviewed. When a file is reviewed, a checksum of its contents is recorded alongside the outcome; when the site is built, that checksum is compared against the file as it then stands. A page that no longer matches drops back to Needs re-review automatically. It is not a claim that anything is wrong with the page — only that the earlier review no longer covers it.
The badge is not a claim that the page’s assertions have been checked against a source. Because the checkers cannot fetch a cited source, no badge on this site can carry that meaning today. The word “verified” is deliberately reserved for a future state gated on citation coverage.
Review coverage, counted from the repository when this page was built on 21 August 2026:
| Section | Reviewed clean | Needs re-review | Total |
|---|---|---|---|
| Story subsections | 13 | 74 | 87 |
| Encyclopedia entries (all collections) | 182 | 114 | 773 |
The remaining collections — media, places, organizations, platforms, narratives, symbols/memes, technology, ideas, and subcultures — have not been through a review pass yet, and most timeline events have not either. Their pages show as Not yet reviewed, which is accurate.
Sourcing, by Section
All citations on the site use a single shared citation format defined once in the repository and generated into the site’s schemas. Where a page carries sources, they render as a numbered list at the bottom of the page; the story compiles its subsection sources into one bibliography.
Coverage as of 21 August 2026:
| Section | Pages with citations | Total |
|---|---|---|
| Story subsections | 74 | 87 |
| Statistics datasets | 13 | 13 |
| Encyclopedia entries | 169 | 773 |
The story and the statistics pages are sourced. The encyclopedia is not. That gap is the largest open issue on the site and it is not hidden here because it is visible to anyone who opens a page.
Where This Is Going
The citation field is now defined on every collection, so any entity or timeline page can carry sources. It is optional, and coverage is being ratcheted collection by collection through backfill work tracked in the repository rather than enforced site-wide at once. Until a collection is backfilled, its pages should be read as an organized summary of publicly reported material, not as a cited reference work. Most entity pages link to a corresponding Wikipedia article, which is a starting point for a reader who wants to go further — not a substitute for a citation.
Known Limitations
- Citation coverage. 169 of 773 encyclopedia pages carry citations. Backfill is in progress and incomplete.
- Review coverage. 195 of 860 reviewable files have passed review and are unchanged since. A further 188 passed review but have been edited since and are awaiting a fresh pass; they show as Needs re-review. The badge tells you which is which.
- Review depth. No automated or human process verifies page claims against external sources. Nothing on this site has been fact-checked in the sense a newsroom means it.
- Link rot. Citations record the date a URL was last checked. There is no automated revalidation and no private archive of captured pages. Links will rot and no process currently catches it.
- Recency bias. The timeline holds 99 events: 4 from the 1990s, 11 from the 2000s, 41 from the 2010s, and 43 from the 2020s. Recent years are far better covered than early ones. This reflects what is easy to find and to source, not the relative importance of those periods.
- Single perspective. One maintainer and one family of AI models draft and review the material. Systematic blind spots are likely and there is no second reader positioned to catch them.
- The repository is not public yet. The content files, the review agent definitions, and the recorded review state all live in a Git repository that is currently private. That means the figures on this page cannot be independently recounted, and the site’s writing — though openly licensed — cannot yet be forked. Opening the repository is an open item, not a completed one.
- No correction SLA. Corrections are welcome and are read, but this is not staffed work and no response time is promised.
Licensing and Derived Material
Two questions come up about where this material comes from and what you may do with it. Both are answered here rather than left to inference.
The writing
Most entity pages carry a link to a corresponding Wikipedia article, and Wikipedia’s prose is licensed CC BY-SA 4.0 — a copyleft license with attribution and share-alike obligations. Facts themselves are not copyrightable, and independent synthesis of sourced facts carries no obligation; close paraphrase of another writer’s expression does.
Our position is that entries here are the former: they are written from the underlying sources, organized around this project’s own framing of platforms, movements, and events, and are not restatements of any single article’s structure or wording.
We do not ask you to take that on trust. The site’s own writing is published under CC BY-SA 4.0 — the same license Wikipedia uses. If a passage is more derivative than we judge it to be, the license the material is already under is the one that would apply anyway. Adopting it costs this project nothing, since it wants to be cited and reused, and it removes the question rather than arguing it.
The images
Photographs and logos are not covered by that license. Each is hosted elsewhere, most often Wikimedia Commons, and retains the license its own creator chose. For every image we record the author, the source, the license, and a link to the source page, and we display that credit beneath the image on the page that uses it — CC BY and CC BY-SA both require attribution at the point of use, not merely somewhere in a data file. Automated checks in the repository fail the build if an image requiring credit is missing any of those fields.
When reusing an image, follow the license shown for that image.
The software
The code that builds the site is separate and is released under the MIT License. Full details are in the repository’s licensing summary.
Data Ethics
We do not scrape private data or bypass access controls. The datasets behind the statistics pages are public or openly licensed, and each one names its sources on the page. Before a dataset is used, the terms of service of its source are checked and recorded in the repository. Where a page would need to identify a private individual to make its point, we prefer to leave the detail out.
Corrections and Contributions
Every page has a feedback button. It files an issue in the project’s issue tracker recording the page URL, the category you pick, and what you wrote — nothing else, and no identifying information unless you put it in the description yourself. Corrections, missing sources, and broken links are all worth reporting, and the more specific the better: a page URL and the sentence at issue is enough. Because this is one person’s project, there is no promised turnaround; reports are read and worked through as time allows.
If you want to contribute directly rather than report, see Suggest an Edit.