CAD-bench/cad-bench-ed-2026-anonymous-tasks

该数据集仅包含CAD-bench的公共任务负载，CAD-bench是一个用于语言模型CAD代理的基于执行的基准测试。它是基准测试加载器使用的轻量级运行时数据集。每个任务目录包括：自然语言基准提示（prompt.txt）、任务元数据（task.toml）、用于验证和媒体生成的参考Build123D解决方案（gold.py）以及可选的夹具（如STEP文件或Blender模拟脚本）。数据集还包括tasks_manifest.json，记录每个任务的包哈希。该数据集故意不包含基准测试结果行、源存档或运行来源。这些内容在配套的完整审查工件中：CAD-bench/cad-bench-ed-2026-anonymous-full。

This dataset contains only the public task payloads for CAD-bench, an execution-based benchmark for language-model CAD agents. It is the lightweight runtime dataset used by the benchmark loader. Each task directory includes: the natural-language benchmark prompt (prompt.txt), task metadata (task.toml), a reference Build123D solution used for validation and media generation (gold.py), and optional fixtures such as STEP files or Blender simulation scripts. It also includes tasks_manifest.json, which records per-task bundle hashes. This dataset intentionally does not include benchmark result rows, source archives, or run provenance. Those are in the companion full reviewer artifact: CAD-bench/cad-bench-ed-2026-anonymous-full.