Curiosity,
with a method.
We explore how software and AI systems can become more useful, more inspectable, and more honest about what they know.
FROM THE LAB / RESEARCH PREVIEWGood answers deserve
VERDICT™
Good answers deserve
better questions.
What happens when AI systems challenge each other’s reasoning? Our research explores independent proposals, adversarial review, and evidence-led evaluation.
Inside the researchEXPERIMENTAL · NO PUBLISHED BENCHMARK CLAIMSENGINEERING DIRECTION
02Where Claude fits in our products
A specific integration plan for drafting, evaluation, and human review.
RESEARCH NOTE
Why “unknown” is a useful answer
Preserving uncertainty in verification and operational interfaces.