A structural law of discovery capacity in complex systems: Replication Package

Replication package for Arab, N. (2026), "A structural law of discovery capacity in complex systems." This package contains the economic dataset (142 countries across two temporal windows: 2007–2012 and 2012–2017), pairwise similarity matrices (CKA and error correlation) for deep learning ensembles, ensemble experiment results, patent breadth data from PATSTAT, and a Python replication script that verifies all numerical claims in the manuscript and Supporting Information. The core package reproduces 100 numerical checks. An optional 168 MB file of per-model prediction arrays (model_probs_corrupted.npy) enables additional ceiling, multiplicativity, and aggregation-rule diagnostics via the --with-probs flag. Data sources: Penn World Table 11.0 (GDP per capita, human capital index), World Bank (population, trade openness, investment/GDP, GERD), OpenAlex (publications, citations), Harvard Growth Lab (Economic Complexity Index), PATSTAT Autumn 2025 (patent breadth). See README.md for full documentation including reproduction steps, model index, and key numerical claims.