Home / Eureka
Eureka — the Gain Score
Can an AI evaluation be gamed? He built one to resist it. Eureka scores AI-era content the way the machines actually read — and it was built, in the open, by a crew of agents checking each other’s work.
Four agents, no honor system
Creator
Makes the work. Never grades it.
Evaluator
Scores against the standard — originality, insight density, machine-readability.
Refiner
Works the gaps the evaluator found. Still doesn’t grade.
Arbiter
Settles disputes and signs the score. The separation is the point.
The origin story runs October 2024 to October 2025 and is published in full — the first shelf item below. The scores it produces resist being flattered upward, because the system that makes the work never touches the grade.
Start here 3 builds
Where the Gain Score starts — the origin story, and the case for an evaluation built to resist gaming.
The system 4 builds
How the Gain Score is built and run — the explainer, the calculator, and the technical stack.
The build record 4 builds
The program as it actually unfolded: roadmaps, timelines, and the dashboards behind the system.