K-EndangeredBench: A Benchmark Dataset and Build Pipeline for Evaluating AI-Generated Content on Korean Endangered Species

Record set pairing Korean national endangered-species designations (Class I/II under Annex 1 of the Enforcement Rule of the Wildlife Protection and Management Act, as amended 2022-12-09) with IUCN Red List categories, GBIF taxonomic and occurrence facts, and BonV Lab ecology-card narratives, for measuring factual accuracy and hallucination in AI-generated science education content. Includes the 282-taxon statutory census in the same schema. Version 1.0.9 keeps the record values of v1.0.6 to v1.0.8. It states the behaviour of the IUCN country-list endpoint as observed rather than inferred, narrows the statement on current Phoxinus assessments to what was checked, identifies the Catalogue of Life release behind the Dryophytes suweonensis synonymy (Annual Checklist 2023, DOI 10.48580/dfsr), corrects the geographic wording of the Dryophytes relation, removes an internal enumeration value and a duplicated assessment from iucn_derived.json (with packaging checks for both), revises the Korean guide and the IUCN value check script. Version 1.0.8 corrected the account of the superseded Phoxinus phoxinus assessments in the IUCN country lists (with their scope and category), traces the GBIF Backbone synonymy of Dryophytes suweonensis to its source checklists, marks categories of earlier IUCN systems in the assessment histories, records the SHA256 of every cached response in the manifest with a script that lets a reader with an IUCN token check the IUCN values by assessment identifier, ships the Annex 1 PDF of the version in force at the currency check, and revises the Korean warning messages and guide. Version 1.0.7 added the lookups behind the hand decisions to the cache manifest. From earlier versions it records where each concept relation came from (rule, range check or hand override), flags the hand-mapped IUCN taxon of the Amur tiger, and keeps from earlier versions the IUCN counterparts of taxa whose Global assessment belongs to another species concept, the census CSV, the classes of the 25 renames of the 2022 annex amendment and the held-out validation of the reference scorer. The deposit contains the data archive (CC BY-NC 4.0) and a snapshot of the build code (MIT) with the inputs and outputs of the paper's example application.