The additive theory suite develops ten falsifiable operational hypotheses around Jev as a bounded decision component. Start with the executed analysis, then inspect the protocol, machine-readable evidence, and source/tests.
These experiments distinguish exact archived Jev-response reanalysis, conditional score simulations, and controlled GraphStore invariants. Zero fresh Jev service calls. Passing a software invariant does not establish better model accuracy. The original manuscript, frozen evidence, inventory and default graph compiler are unchanged.
python -B -m unittest discover -s graph_synthesis/theory_suite/tests -v
python -B -m graph_synthesis.theory_suite.run --checkThe suite adds conformal abstention, residual-error dependence, rare-edge prevalence sensitivity, component risk budgets, topology-aware review, calibration-selected cascades, joint identity decoding, dependency-aware cache invalidation, alternative-proof lineage, and query-contract migration previews. Negative and mixed results are retained, including a nominal-coverage shortfall and a graph-precision cost hidden by near-equal classification macro-F1.