site/benchmarks.json

{
  "data-agent-bench": {
    "label": "Data-Agent Bench",
    "description": "Verified data-analysis tasks over Kaggle datasets. Each task is a (question, gold-answer) pair tested with a mode-aware grader (exact / numeric / flexible / LLM-judge).",
    "source": {
      "harbor": "AdithyaSK/data_agent_rl_environment_eval-harbor",
      "hf_dataset": "AdithyaSK/data_agent_rl_environment_eval"
    },
    "sweeps": [
      "v1"
    ]
  },
  "dabstep-bench": {
    "label": "DABstep Bench",
    "description": "Data Agent Benchmark for Multi-step Reasoning \u2014 450 financial-analytics factoid tasks over a shared Adyen-payments corpus. Released by Adyen + Hugging Face (arXiv:2506.23719).",
    "source": {
      "harbor": "AdithyaSK/dabstep-harbor",
      "hf_dataset": "adyen/DABstep",
      "paper": "https://arxiv.org/abs/2506.23719",
      "leaderboard": "https://huggingface.co/spaces/adyen/DABstep"
    },
    "sweeps": [
      "dabstep"
    ]
  }
}