midah/pairwise_analyses

该数据集包含软件和AI许可证之间的成对许可性判断，由三个LLM模型在v4提示下生成，并包括衍生的共识排序和Hasse图。数据集涵盖了两个语料库：93个许可证的Hugging Face Hub可选集和完整的747个许可证的SPDX + AI规范语料库。每个判断可能是A > B（许可证A比B更宽松）、A = B（A和B在范围上等效）或A < B（B比A更宽松）。共识是通过多数投票计算的，原始模型输出中保留了分歧。数据集还包括每个许可证的LLM提取特征。

Pairwise permissiveness verdicts across software and AI licenses, produced by three LLMs under the v4 prompt, plus the derived consensus ordering and Hasse diagram. Covers two corpora: the 93-license Hugging Face Hub-selectable set and the full 747-license SPDX + AI canonical corpus. Each verdict is one of: A > B (license A is strictly more permissive than B), A = B (A and B are equivalent in scope), or A < B (B is strictly more permissive than A). Consensus is computed by majority vote across models; ties and disagreement preserved in the raw per-model outputs. Includes LLM-extracted features per license.