Paired MSA-Saudi Arabic Tool-Use Test Set

A paired test set for controlled evaluation of tool use in Modern Standard Arabic and Saudi Arabic. The release contains 150 records forming 75 meaning-matched MSA–Saudi Arabic pairs across five deterministic mock tools, together with tool definitions, fixtures, answer contracts, schemas, and checksums. Version 1.0.3 is the confirmed evaluation snapshot used in the accompanying study. Its 150 released request texts match the evaluated input package exactly; no request text was changed after the evaluated snapshot was frozen. The public package is a privacy-minimized representation of that same benchmark, with records SHA-256 1239ae0cdf24f6343d574c9b04477259650221a1f1bfa932b38c7f139af41e34. Two different assigned human request authors produced the realizations, one for MSA and one for Saudi Arabic. Two independent human reviewers accepted the final pairs. Because each variety has one request author, author style is confounded with language variety. The public package excludes personal names, identity mappings, individual review records, and working materials.