Evaluation data and analysis code — zero-marginal-cost RAG advisory system

This deposit contains the evaluation set, human-validated answer keys, paired experiment outputs, and analysis code supporting the paper "Evaluating Determinants of Answer Quality in a Zero-Marginal-Cost RAG Advisory System." It enables independent verification of the paper's quantitative claims, including the +21.3-point gap between single-reference and human-validated multi-reference answer-key grading (95% CI [12.8, 29.8], n=94) and the zero-marginal-cost model comparison. The full production application is in a private repository, available from the author on reasonable request. See README.md and MANIFEST.md inside the archive for a file-by-file map to each result. Data licensed CC BY 4.0; code licensed MIT.