job-run/source_claim_audit.csv

claim,primary_anchor,verdict,scope_or_qualifier,arxiv_revision
1,Theorem 3.7,source-supported + finite-population inequality audit,"Requires a depth-D path, M-coverage, bounded coordinate second moments, and a bounded l1 norm for the global-optimum coefficients.",2605.01082v1
2,Theorem 4.5,source-supported + Gaussian hard-instance quadrature,Construction-specific: D=kp and p<=k-1; numerical scaling is not a proof.,2605.01082v1
3,Lemma 3.1,reproduced to numerical precision,First-order identity at the population logistic optimum on its feature space.,2605.01082v1
4,Lemma 3.3 and Pinsker inequality,reproduced to numerical precision,The reference p* must be optimal and q must use the same feature space.,2605.01082v1
5,"Section 2, Equation (3)",reproduced on exact finite populations,Sequential path audit; each agent exactly minimizes population BCE.,2605.01082v1