Forecasting
How accurately can people and models express uncertainty, and when does aggregation improve the result?
Research
How should human and machine judgement be measured?
Our research focuses on forecasting, calibration, evaluation, question quality and the design of decision systems that can learn from resolved outcomes.
Calibration
Does stated confidence line up with what actually happens, across time and across different domains?
Human-AI teams
When does AI assistance improve judgement, and when does it introduce correlated errors or misplaced confidence?
Evaluation design
How do scoring rules, rubrics, evidence cutoffs and question wording change what an evaluation actually measures?
Research standard
We distinguish observations from inference, preserve source provenance and prefer questions that can eventually be resolved against reality.