Soxhlet oil-yield screening benchmark: data and code

Data and computational materials accompanying 'How much do published data add to five local measurements? A publication-held-out benchmark for Soxhlet oil-yield screening'. The deposit contains 197 literature condition-yield records from 22 publications and five feedstocks, with bibliographic mappings, model code, fixed settings, reference predictions and verified results. ESM_2_Model_Reproduction.zip reproduces the eleven-method comparison in Table 2 using eleven publication holdouts, five calibration measurements per holdout and fifteen seeds. ESM_7_Benchmark_Analysis.zip reproduces the calibration-label subgroups, Figure 2 and paired publication-bootstrap intervals from the saved predictions. Standalone CSV files duplicate the input data and source inventory in ESM_2. Each archive includes execution instructions and dependency versions. The separate five laboratory measurements and 27-condition design accompany the manuscript and are not included in this computational deposit. Version v1.2.0 adds open licenses and clarifies the tested software environment. All observations, feature values, class boundaries, model settings and reference predictions are unchanged. Earlier versions remain available through Zenodo version history. Python source code is licensed under MIT. Compiled data, analysis outputs and descriptive documentation are licensed under CC BY 4.0, to the extent of the authors’ rights. These grants permit reuse, adaptation and redistribution, including commercial use, subject to the respective license terms. RIGHTS.md and the accompanying license notices specify the file-type scope and attribution. Cited publications and third-party dependencies retain their own rights; source article PDFs and dependency packages are not redistributed. The reference environment is Python 3.12.14 with the exact versions in requirements.txt. Other software versions have not been certified to reproduce every saved prediction. Verification is stricter than agreement after rounding; any FAIL should be investigated using the generated comparison files. Funding:This work was supported by Zhejiang Provincial Natural Science Foundation of China (LQN26B060001), 2025 Yongjiang Talent Introduction Program funded by the Ningbo Education Bureau, Ningbo Commonweal Programme (2022S122), Ningbo Youth Science and Technology Programme (2024QL007) funded by the Ningbo Science and Technology Bureau, China, Research Fund for International Excellent Young Scientists (W2532017) from the Natural Science Foundation of China (NSFC), Zhejiang Provincial Key Laboratory for Carbonaceous Wastes Processing and Process Intensification Research funded by the Zhejiang Provincial Department of Science and Technology (2020E10018), and Ningbo Municipal Key Laboratory on Clean Energy Conversion Technologies (2014A22010)