01
Structural Invariant Evaluation
Methods for identifying and testing properties that should remain preserved across model transformations and outputs.
Research
What must remain stable when intelligent systems transform information.
Our research examines the structural properties of AI systems across transformation, interaction, evaluation, and coordination. It is the foundation for our evaluation and systems work.
Areas
Four connected lines of research.
02
Agents & Coordination
Research into agent behavior, interaction, delegation, coordination, and emergent system-level capabilities.
03
Language & Interface Regimes
Investigation of language, interaction structures, and the interfaces through which humans and intelligent systems act.
04
Applied Data Systems
Computational methods for extracting, organizing, testing, and interpreting complex information across research domains.
Method
Research that ends in a test.
We treat reliability as a structural question: which properties of information, meaning, and behavior should survive when a system summarizes, translates, reasons, or acts?
Each line of research aims at something measurable: a property that can be stated precisely, tested on real systems, and used to decide whether a system can be trusted with a task.
Collaboration
Work with us on research.
We welcome conversations with research groups, institutions, and organizations working on related questions in AI evaluation, agents, and reliability.