Model Registry

A model registry is the authoritative record of every production model an institution operates, maintained at the level of detail required to distinguish genuinely distinct models from duplicate entries describing the same underlying model twice: target definition, filter criteria, sample size, and validated performance metric together define a model's signature, and registration asserts signature uniqueness, meaning two entries sharing an identical signature are one model, not two. A properly maintained registry also separates true, production predictive models from governance artifacts, reference tables, and retired experiments that may appear in an institution's broader model count but do not themselves generate predictions used to price capital, since conflating these categories inflates a headline model count in a way careful diligence will catch immediately. This platform's own registry maintains this distinction explicitly, citing its production predictive model count separately from its governance artifacts and reference tables, because citing the larger, undifferentiated total as though it represented the predictive fleet overstates that fleet's actual size by a meaningful multiple, an overstatement any careful diligence process will identify quickly once it asks for the registry itself, at the signature level, rather than accepting a summary total presented without that underlying detail attached, since the detail is where the inflation always hides.