Suggestible Judges: data, code and manuscript

Data, code and manuscript for the paper Suggestible Judges Show Asymmetric Deference to Negative Labels in Large Language Model Adjudication. Three LLM judges (Claude Opus 4.7, GPT-5.5, Gemini 3.5 Flash) adjudicated annotator disagreements on SEC 10-K knowledge-graph edges with and without annotator labels in the prompt, including a randomly assigned synthetic label. The archive contains every per-pair verdict log, the disagreement pairs with both annotators' labels, the experiment harness, all analysis scripts, and the LaTeX source and compiled manuscript. Code is MIT-licensed; text and data are CC BY 4.0; ChaosNLI items keep their CC BY-NC 4.0 license. The full annotation sets, which are the evaluation data of a separate manuscript, are withheld; aggregate counts are included.