subjectId
inScope
Contamination-resistant competitive-programming problems collected
on a rolling basis from LeetCode / AtCoder / CodeForces. Python,
C++, Java solutions evaluated by hidden test cases. Includes
code-generation, self-repair, code-execution, and test-prediction
sub-tasks.
outOfScope
Repository-scale software-engineering tasks (use SWE-bench), agentic
tool-use, ML-engineering tasks, natural-language reasoning, and
benchmark snapshots before the contamination cutoff.
outOfScopeReasonIds