Definition
Plain language
Tailoring a solution to the exact questions on the visible exam rather than learning a general rule.
As stated in the literature
Exploiting the specific unlabeled test inputs (or repeated scorer feedback on them) at solution time, rather than inducing a hypothesis expected to hold on unseen data.
Also called: transduction, transductive exploitation
Why it matters: It can win on a fixed benchmark while producing nothing that works on tomorrow's data, which makes such scores easy to misread as genuine capability.
For example, instead of building a rule that works on any new photo, a system tunes itself specifically to the exact hundred photos it can see in the test set.
Heard on the show
“… a raw score improvement didn’t count if it depended on leakage, duplication, entity overlap, transductive test-set exploitation — meaning tailoring the solution to the visible test inputs — or any other …”Episode 257 — They Planted a Shortcut in the Data. Seven Coding Agents Took It.