AIS-2026-53: computational supplement

Requirements used: Python 3.12.13, NumPy 2.3.5, SciPy 1.17.0, Matplotlib 3.10.8.
No patient data, trained model, NetworkX installation, network access or external source dataset is required to execute the analyses.

From this extracted directory:
python code/revise_analysis.py
python code/budget_sensitivity.py
python code/verify_analysis.py
python code/make_figures.py

The first script recomputes the original graphs, all 48 compartmental formulations and their exact Shapley allocations, serial-subdivision benchmark, exact capacity-degradation curves, 117 reference single-segment contingencies, continuous venous response surfaces and weight perturbations. The budget script recomputes seven budget contrasts over all 48 formulations, giving 10,752 exact coalition solutions, and checks reference replication and equal increments. The verification script independently checks cut-based maximum flow against the SciPy maximum-flow solver on 100 deterministic seeded edge-subset checks. The figure script creates the computational figure components; subsequent author assembly and presentation edits in Adobe Illustrator are described in FIGURE_PROVENANCE.txt and Supplementary Section S6.

results/ contains machine-readable full-precision numerical outputs.
legacy_edges.json specifies the original graphs. compartment_A.json and compartment_B.json specify the complete enhanced/baseline reference graphs, solved node pressures, tissue-unit flows and directed segment flows.
structural_scenarios.csv and hydraulic_shapley.csv contain all 48 formulations.
continuous_venous_surface.csv contains 882 joint MCLV/SDC states.
legacy_results.json contains original exact coalition values, Shapley values, all degradation steps and subdivision results.
validation.json and independent_flow_checks.json contain numerical verification results.

Original-graph Shapley calculations use no random seed. Original spectral weight noise uses seeds 4200–4203, with 2000 draws per scale. Paired compartmental noise uses seed 8400, with 1000 draws at sigma 0.20. Independent flow checks use seed 53. Exact subset enumeration replaces random-order failure sampling.

All dimensions, weights, enhancement levels and outlet assumptions are synthetic model definitions. The tissue layouts are not patient reconstructions or a mesh-convergence validation. Model verification is not clinical validation.

Copyright: original study code and figures, the authors. Dependency software and fonts retain their respective licenses. See FIGURE_PROVENANCE.txt.

budget_sensitivity.csv and budget_summary.csv contain full budget-contrast outputs. budget_validation.json records conservation, positive tissue exchange, exact reference-budget replication, Shapley efficiency and equal-increment verification. Budget multipliers are numerical stress tests, not an empirically calibrated range.
