Reasoning
How commitments are represented, repaired, forgotten, and verified.
Current questions
How commitments are represented, repaired, forgotten, and verified.
When learned operators transfer, and what breaks when the setting changes.
Machine learning for scientific measurements where artifacts can mimic signal.
Evaluation and routing for language tasks where disagreement is part of the data.
Structure in neural wiring, with reconstruction choices treated as evidence.
Benchmarks and audits that separate capability, calibration, and measurement bias.