A promotion gate is the fixed review a model must pass before it is used in client work, covering discrimination, calibration, sufficiency of data and leakage checks on a holdout it never saw. A model that fails stays visibly experimental.
The gate is applied to each model individually against a published standard. It is not a one-time approval of a method. Every model is tested on the period after its training window, and the result is recorded in the registry whether it passes or not.
A gate that never rejects anything is not a gate. Models that fail remain labeled as experimental, stub or failed, and a model that later fails review is demoted with the reason recorded. A result that looks too good is treated as suspect, because the usual cause is information from the outcome reaching the inputs.
Passing the gate does not make a model right on any one matter. It establishes that, as a population-level method, it met the stated bar on data it had not seen.
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.