A reference population is the defined set of comparable resolved matters against which a score, a distribution or a rate is computed. A figure means nothing until its reference population is stated, because the same matter ranks differently against different peers.
Every probability, band and percentile is a statement about a population. The reference population is defined by filters such as forum, case family, filing window and any other stated criterion, and the definition is published with the figure. Changing the filters changes the answer, and a reader needs to know which answer they were given.
A reference population must be large enough to support the claim. Below a stated minimum the method returns a tier or declines to rate, because a precise-looking number computed from a handful of peers conveys confidence the data does not hold.
The reference population is not the training population. The training population is what a model was fit against. The reference population is what a given output is compared to. They can overlap, but they answer different questions and are recorded separately.
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.