Blind test
Also known as: Blind validation · Blind test · Two-send blind test · Out-of-time
A test design in which Zarv computes the score without knowing which cases had the event, so the result cannot have been tuned to the answer key.
At Zarv
2 sends: Zarv delivers the scores before it receives the outcome labels.
A score test is valid only if one thing holds: whoever computes the score cannot know the answer. If the outcome labels — who had a claim, a default, a cancellation — are at hand when scoring, there is no way to prove afterward that the score was not, consciously or not, tuned to them.
A blind test guarantees this with two sends. On the first, you send only the registration data, with no outcome indicator, and Zarv returns the scores. On the second, with the scores already delivered and frozen, you send the outcome labels for the match. The order is what shuts the door: the score exists before the answer key arrives.
Frequently asked questions
Why does a score test need to be blind?
So the result is trustworthy: if whoever computes the score knows in advance which cases had the event, there is no way to guarantee the score was not tuned to the answer key. A blind test separates the two moments into two sends.