Project Overton
Home
Analyze an Article PRO History PRO
Rubric
Reports SOON
How It Works Analysis Types Scoring Models Bibliography Glossary
Anchor Library SOON Anchor Names SOON
About Project Overton FAQ & Challenges
Data Privacy
Membership & Pricing
Log in Get access
Scoring Models
Five independent models process the same observations differently. Where they agree is confidence.

Scoring Models

Every article is analyzed through a shared set of analytical lenses. Five independent scoring models then process those observations differently. Where they agree, confidence is high. Where they disagree, the disagreement is the finding.

Model What it does When to use it
Elimination DEFAULT Starts with all possible scores on the table, then removes the ones the evidence rules out. What's left is a range. The width of the range tells you how ambiguous the article is. Contradictions are flagged, not averaged away. Every analysis. Shows you where the evidence converges AND where it conflicts. The range width tells you how much the article is saying two things at once.
Tripwire Checks which analytical detectors fire when they read the article. Like a panel of sensors — each one either trips or it doesn't. The pattern of which sensors fired maps to a score. Cross-validation. When Tripwire and Elimination agree, the score is robust. When they disagree, the article is doing something unusual.
Accumulation Weighs every piece of evidence by how strong it is, how prominent it is, and how central the lens is to the axis. The weight of evidence determines the score. Articles with rich, layered signal across many dimensions. Shows exactly how much evidence supports each score level — you can see where the weight concentrates.
Fingerprint Compares the article's analytical pattern against 80+ verified governance records to find the closest match. "This article reads like the world of X." Historical context. Answers: "which real-world governance reality does this article assume?" The match gets more precise as the anchor library grows.
Debate Makes the strongest case for the highest possible score, then the strongest case for the lowest. The gap between them is the finding. Contested pieces where you want to understand the tension. Shows the strongest case for the highest score AND the lowest — both arguments visible, both backed by evidence from the text.

All five models analyze the same evidence independently. When multiple models converge on the same score, confidence is high. When they disagree, the disagreement itself reveals something about the article. Pro and Research users can run all five and compare results side by side.