Evidence and limits
Finding: Participants underperformed the risk model, could not reliably judge performance, and interacted with advice in racially disparate ways.
Impact: Strengthen subgroup and interaction-effect evaluation. Reject fair stand-alone model metrics as proof of a fair combined system.
Boundary: Lay simulations do not establish judge behavior. Next: which workflow traces reveal disparate interaction before deployment?
Flags: constructed decision setting; subgroup transfer
Record State: Catalog review recorded; not a performance endorsement
What was reviewed: Structure and relevance screened; not a full evaluation