Species Exposure Bundles

This dataset supports "From raw occurrence records to accessible knowledge: Transforming biodiversity infrastructure through evidence-manifesting design" (Ecological Informatics). It contains three components. (1) Supplementary_Table_S2.xlsx — complete transcripts of all 120 language-model responses (10 species × 4 models × 3 retrieval conditions) with the corresponding expert evaluation sheets and dimension-level scores. (2) Statistics_deposit.R — the analysis script reproducing every quantity reported in the paper (Table 2, Table 3, failure and perfect-score rates, coefficients of variation, mixed-effects model, pairwise contrasts, Wilcoxon tests, and the dimension-level sensitivity subsets). It expects the workbook above in the same folder. (3) WoC_SEB.zip — the Species Exposure Bundles for the ten test species, archived byte-for-byte from the cheCkOVER snapshot dated 29 January 2026, the evidence state served throughout the February 2026 testing window. MANIFEST.json records per-file SHA-256 checksums, retrieval times (UTC), and the version and snapshot fields as read; checksums.txt verifies the archive with sha256sum -c. Notes on the bundle metadata: version (1.0.0) denotes the analytical stage of the bundle (baseline); snapshotUsed.snapshot_name ("v0.9") is an internal label of the artefact package and not a separate version; generatedBy.pipelineVersion contains an unpopulated placeholder in this snapshot; created and updated are stamped at retrieval, so the checksums are those of the archived retrievals at the recorded times. Bundles were assembled by the endpoint at request time and not stored during the experiment; the transcripts in (1) are the record of the assembled content as each model consumed it. The World of Crayfish® platform (https://world.crayfish.ro) is live and curated continuously, so values served today differ from those in this snapshot.本数据集支持论文《From raw occurrence records to accessible knowledge: Transforming biodiversity infrastructure through evidence-manifesting design》（发表于*Ecological Informatics*），包含三个组成部分。 (1) Supplementary_Table_S2.xlsx —— 包含全部120份语言模型生成回复的完整转录文本（10个物种×4个模型×3种检索条件），附带对应的专家评估表及维度层级得分。 (2) Statistics_deposit.R —— 用于复现论文中所有统计量的分析脚本（包括表2、表3、失败率与满分率、变异系数、混合效应模型、成对对比检验、Wilcoxon秩和检验以及维度层级敏感性子集）。该脚本需依赖同目录下的上述工作簿文件。 (3) WoC_SEB.zip —— 针对10个测试物种的物种暴露包（Species Exposure Bundles），该归档文件是从2026年1月29日的cheCkOVER快照逐字节归档而来，为2026年2月测试窗口全程使用的证据状态。其中MANIFEST.json记录了每个文件的SHA-256校验和、检索时间（协调世界时UTC）以及读取到的版本与快照字段；checksums.txt可通过`sha256sum -c`命令验证该归档文件的完整性。关于该包的元数据说明：版本号（1.0.0）代表该包的分析阶段（基线版本）；snapshotUsed.snapshot_name("v0.9")为产物包的内部标签，并非独立版本；generatedBy.pipelineVersion在本快照中为未填充的占位符；created与updated字段为检索时加盖的时间戳，因此校验和对应记录时刻的归档检索结果。该物种暴露包由请求时的接口端点实时组装，实验期间未进行存储；(1)中的转录文本即为每个模型调用时组装内容的完整记录。World of Crayfish®平台（https://world.crayfish.ro）处于持续更新运营状态，因此当前提供的数据与本快照中的数值存在差异。