The official verified eval suite for the data-agent RL pipeline. 366 Harbor-format
data-analysis tasks, each with an LLM-assigned difficulty label (L1–L5), a Kaggle
dataset dependency, and a tested reward function.
💡 Browse this dataset in your browser — click the badge above or open
AdithyaSK/harbor-visualiser
to inspect every task's spec, instruction, environment, tests, and difficulty.
