Distribird: benchmark data and evaluation results

Benchmark datasets, evaluation code and results supporting the manuscript Distribird: Literature-Informed Prior Distribution Design for Bayesian Model Calibration. Distribird constructs prior distributions for Bayesian model calibration from published literature. Given a parameter name, description and domain context, a multi-agent pipeline searches the literature, extracts and weights reported values, and fits a probability distribution by AIC model selection. This deposit contains the benchmark used to evaluate it: 24 parameters across 12 use cases in ten scientific domains, run on three local open-weight language models and compared against a single-prompt baseline. Contents: the twelve benchmark datasets with the fetch logic recording each original source; the parameter and model definitions; the run scripts and report builders; and the evaluation output, comprising per-prior provenance, scored comparisons, cost statistics, per-use-case tables, figures and rendered reports. See README.md for a full description and a mapping from each manuscript table and figure to its result files. All primary datasets are third-party and publicly available from their original sources, which are cited in the manuscript and encoded as URLs or API calls in data/download.py.