Critical Markup applies a consistent five-dimension rubric — factual accuracy, source diversity, editorial neutrality, comprehensiveness/context, transparency — to articles from across the political spectrum. The rubric scores how a piece is constructed, not whether its conclusions are correct.
How it works
A scheduled crawler picks recent articles from each source's homepage, fetches the full text via a JS-rendering proxy stack with archive.ph fallback for hard-walled targets, and runs each through Anthropic's Claude API with a fixed methodology prompt. The structured analysis is stored alongside the article extract; the public site shows ratings, the one-line summary, the full critique, and a prominent link back to the original.
The rubric
- Factual accuracy — verifiable claims (numbers, dates, quotes), specificity
- Source diversity — ratio and type of voices quoted on the central question
- Editorial neutrality — word choice, framing-as-fact, attribution patterns
- Comprehensiveness/context — what's missing that a reader would want
- Transparency — bylines, datelines, source affiliations, beat disclosure
Fair use
Each analysis page contains commentary and criticism on the article (17 U.S.C. § 107) and quotes specific lines for the explicit purpose of demonstrating framing or sourcing patterns. Articles aren't republished in full; readers click through to the original at the source.
If a publication objects to inclusion despite all this, the system honors a removal request as a matter of practice. Email critical-markup@criticalmarkup.org.
What this is not
- Not a fact-checker — we don't adjudicate truth claims; factual accuracy is one dimension among five.
- Not a partisan watchdog — the same rubric runs on Jacobin and National Review, NYT and WSJ.
- Not a replacement for reading the article — we link to and encourage reading the original.
- Not a recommendation engine — readers decide what to read; we surface patterns.