Almost every data product in this market now ships a score. Very few of them ship the thing that makes a score meaningful, which is a record of whether it was right.
Ranking answers a smaller question
Ranking says: of these two rows, this one first. That is genuinely useful, and it is also entirely relative. A ranked list of a thousand poor opportunities still has a top ten, and it will look exactly like a ranked list of a thousand good ones.
Scoring should answer a larger one: how likely is this, on its own terms, to become something? That question has an answer that can be checked, which is precisely why so few products want to be held to it.
A score you cannot be wrong about is not a prediction. It is a presentation.
What a calibrated score requires
- An outcome. Somebody has to record what actually happened, including the ones that went nowhere.
- A stable definition of the score at the time it was given, so a later change to the model does not quietly rewrite its own history.
- Enough volume in one market that the comparison means something. A score calibrated nationally can be wrong in every county at once.
This is the part that cannot be bought in, because it is not a data source. It is a loop: discover, understand, prioritize, act, measure, learn. Every pass through it makes the next score slightly less of a guess.
Not every pulse becomes an opportunity, and a system that pretends otherwise is easy to build and impossible to trust. The distinction between the two is the entire product.
