Replication materials for "Cited but Not Used: Documented Adoption and Data Accessibility in 40 Non-English NLP Benchmarks"

Replication materials for a study of 40 non-English NLP benchmarks measuringthree properties usually collapsed into one: how often each is cited, whatciting authors document doing with it, and whether its data can still beobtained and its licence determined. Contents: the labelled citation corpus (487 records); the benchmark codingtable; the accessibility and licensing audit with the exact revision inspectedfor each benchmark; three reliability instruments with their completed returns;and five scripts that recompute every figure reported in the paper. Each scriptchecks its output against the value printed in the manuscript and fails on amismatch. See README.txt. Requires Python 3 with openpyxl, numpy and scipy. No network access needed.