Authority is a boundary, not a suggestion.
A design note on keeping evidence, proposals, permissions, and consequential actions distinguishable.
Intelligence
you can
interrogate.
Research matters when it helps us ask better questions about the systems we build—and test the answers against evidence.
Keep the claim
close to the method.
We explore agents, reasoning, memory, planning, tool use, evaluation, safety, and verification. Our public notes explain methods and design questions. They do not claim benchmark superiority, a released foundation model, peer review, or independent assessment.
QuestionMethodEvidenceLimits
How we approach evaluation ↗Explore the public research record.
Execution and verification are different states.
A technical note on the bounded arithmetic workflow implemented in this website preview.
Evaluation starts with the task.
Useful evaluation asks whether the system helped achieve a real outcome under stated conditions.
Safety is an engineering question.
We consider how more capable assistance can remain bounded and accountable.
A useful direction
needs a measured result.
The Merism website preview makes one small verification boundary inspectable. Its arithmetic can be checked locally; it does not establish a deployed model capability or general agent reliability.
Inspect the synthetic example ↗