GEO-LINE-X: A Cross-Model Behavioral Scoring Dataset for LLM Evaluation Research

GEO-LINE-X is a pilot-scale dataset for studying how evaluator identity and user-directive framing affect behavioral measurements of large language model outputs. The release includes the stimulus set, raw response corpora from three models, consolidated behavioral scoring records, a directive key, the scoring rubric, and a reproducibility script. It supports research on LLM-as-judge reliability, directive-conditioned behavioral measurement, and evaluation-environment effects. It is not intended for ranking model capability or making population-level claims without further replication.