Definition
Plain language
One graded chunk of a contest problem, usually an easier version with smaller inputs.
As stated in the literature
A scored partition of a problem defined by input constraints; solving only small-constraint subtasks yields partial score and reveals which algorithmic complexity class a solution reaches.
Also called: subtasks
Why it matters: It tells you not just that a solution fell short but which size of problem it can actually handle, which points directly at what needs to change.
For example, one problem may be split so that handling lists of up to 100 items is worth 30 points and handling lists of up to a million is worth the remaining 70.
Heard on the show
“The checklist verifier here breaks each task into weighted subtasks.”Episode 017 — When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers