Benchmark data and evaluation records for MetaboSquad

Version 1.2 (2026-11): Relabeled the five nominal 300 dpi figures in benchmark_images/tool_50 (benchmark_*_03.png) from PASS to FAIL in benchmark_gt.json, because Matplotlib renders them at 299.72 dpi; the images themselves are unchanged. Added boundary_measurements.csv (re-measured resolution, font size and line width of the 50 tool_50 figures), model_only_font_line_estimates.jsonl and model_only_prompt.txt (the model-only font-size and line-width comparison). Updated README.md.

Version 1.1 (2026-11): Updated README.md and renamed pubmed_reverse_validation_logs.json to case_study_retrieval_logs.json.

This dataset contains the benchmark data and evaluation records for the manuscript "MetaboSquad: A Graph-Orchestrated LLM Assistant for Metabolomics Workflows". It includes (1) benchmark_100_ground_truth.csv, the 100-query task benchmark with the expected outputs; (2) benchmark_images, the figures of the figure-review benchmarks (benchmark_20, semantic_50 and tool_50) with their ground-truth labels; (3) boundary_measurements.csv and model_only_font_line_estimates.jsonl, the tool measurements and the model-only estimates for the tool_50 figures; (4) raw_record, the LangSmith exports and the literature and plot-review reports; and (5) case_study_retrieval_logs.json, the retrieval logs of the three-stage case study. The source code is available at https://github.com/CMDM-Lab/MetaboSquad.