Evaluations
How We Test the Machinery
A framework earns trust by surviving scrutiny of its own. These evaluations test the machinery behind the analysis from three directions: whether the synthetic networks behave like real human organizations, whether the indexes measure distinct things rather than one signal in several hats, and how much an incomplete collection could move the read. Each is a standalone, reproducible report. The statistics are computed by recognized R packages, and the full code that produced every figure is printed in each report's appendix, so any result can be re-run and checked.
Network Realism
Do the generated networks behave like real human organizations? Four structural signatures -- cliquishness, small-world clustering, community structure, and degree mixing -- measured against documented real-network ranges, alongside a conditional uniform graph (CUG) suite, across 50 generated networks.
View report →Index Validation
Are an index's inputs distinct, or the same signal repeated? Input redundancy by variance inflation factor, correlation and dimensionality screening (OECD/JRC Step 4), and single-input sensitivity (Step 7), across 50 regenerations of one organizational construct.
View report →Collection Margin of Error
What an incomplete collection might have missed. The empirically-known core re-scored across 25 plausible unseen peripheries, with the margin shown for each officer and the test's own diversity checked against the model running unconstrained.
View report →