A target definition is the precise statement of what a model predicts, including the event, the population, the period and the data source the label comes from. A model is only as clear as its target.
Vague targets produce vague claims. A well-formed target names the event, such as resolution within a window or a particular disposition class, the filter that defines which matters are in scope and the source and rule used to extract the label from the record.
Where a public record cannot reliably encode a result, no target is built on it. Prevailing party and case value are the standing examples. Naming a target after what the record supports, and not after what a user would like to know, is what keeps a model from overclaiming.
A target definition is not the model's output format. The output can be a probability, a band or a score, but each of those is a statement about the same defined target.
Working a file that turns on this term?
Working through a diligence process?
Institutional partners evaluating a position against this platform's outcome and duration models are welcome to reach out directly.