{"schema_version":"1.0","measurements":{"unit_of_analysis":"A reported measurement or a separately identified evaluation observation.","numeric_values":"Only explicitly normalized observations have result_value, baseline_value, and unit. Null does not mean zero.","missing_information":"Unknown design, sample, and uncertainty remain null. A record-level reported measure is not a new evaluation.","review":"review_state is inherited from the underlying measure. extraction_basis identifies a new source review when one occurred.","qualifications":"field_qualifications retains each evaluation field's availability, review state, and public qualified answer. An unresolved answer is not promoted to a confirmed value."},"crosswalks":{"relationship":"Exact retained benefit classifications and explicitly inferred system-task correspondences are separate rows.","task_rules":"Conservative phrase matching in the stated system, purpose, and AI role. The basis excerpt is retained for inspection.","limits":"An inferred correspondence does not establish equivalent risk or harm. Missing categories are not negative classifications.","sources":["https://trustible.ai/post/a-framework-for-measuring-the-benefits-of-ai/","https://doi.org/10.1787/cb6d9eca-en"]}}