Data and code for: A composite risk framework reveals distant-water transshipment as a governance blind spot for migratory species in marine protected areas

Reproducibility deposit for the manuscript "A composite risk framework reveals distant-water transshipment as a governance blind spot for migratory species in marine protected areas" (under review). Contains everything needed to reproduce every figure, table and statistic in the paper without downloading the multi-gigabyte raw geospatial sources, which are linked rather than redistributed. Contents: notebooks/ - five Jupyter notebooks: the preprocessing pipeline (included for provenance), the composite risk model, the MPA analysis and all paper figures, the network analysis, and a regression-check notebook asserting 37 frozen invariants. src/ - four shared modules: path resolution, data loading, figure style, and analysis helpers. data/processed/ - df_model.pkl, the frozen 274,375-event x 148-column dataset from which every published number was computed, plus model parameters and an Arrow-format portable copy. data/spatial/, data/iuu/ - EEZ v12 boundaries and tuna-RFMO IUU vessel names. results/ - the 52 tables behind the reported results, and the paper's six figures plus one supplementary figure (PDF and 600 dpi PNG). Method: carrier-vessel encounter and loitering events from Global Fishing Watch (January 2017 - November 2024) intersected with the June 2024 World Database on Protected Areas; a ten-indicator composite risk index built with Entropy-CRITIC weighting over a Bayesian hierarchical Poisson-Gamma model, classified by Jenks natural breaks. To reproduce: install requirements.txt, set ANALYSIS_ALLOW_STALE_MANIFEST=1, and run 99_regression_check.ipynb. It recomputes the headline statistics independently and prints "ALL INVARIANTS PASS". Full instructions in README.md. Built and verified with Python 3.12.3. Code is MIT-licensed; data and derived outputs are CC BY 4.0. Raw source data remain under their providers' terms.