Data and code for "From Detection to Decision: Calibrating the Operational Value of Open-Source Geopolitical Early Warning for the China–Europe Railway Express"

Data and code accompanying the manuscript "From Detection to Decision: Calibrating the Operational Value of Open-Source Geopolitical Early Warning for the China–Europe Railway Express" by Tengyue Xie and Benhong Peng (under review; the full citation will be added to this record upon publication). The companion ground-truth dataset disruption_v1 is archived separately at https://doi.org/10.5281/zenodo.21465629. The package contains the complete calibration and decision-model pipeline of the paper: the causal GDELT spike rule and its audited alert episodes (2018–2025, two geographies, with the frozen audit overrides), tier-level precision calibration with Wilson and Jeffreys intervals, the feed-health coverage mask and axis-filtered daily event series (shipped so that the calibration logic can be re-derived without the raw archive), the graded response decision model (monitor–quote–hold–divert) with its threshold-policy outputs, value-of-information and annual accounting tables, the day-by-day causal replay of the September 2025 Poland–Belarus border closure, and all figure scripts. Scripts are numbered in execution order and run unmodified from the package root (environment specified in pixi.toml: Python with pandas, numpy, scipy, matplotlib, pyarrow). The raw 1.4 GB GDELT slice is not included; data/raw/gdelt/MANIFEST.txt provides MD5 checksums and BigQuery rebuild instructions from the public GDELT archive. Code is released under the MIT license; data under CC BY 4.0. A data dictionary for every shipped CSV is provided in README.md inside the archive.