Reversible moment bounds — 0.1.0-candidate

Unrefereed, AI-assisted mathematical candidate, version 0.1.0-candidate.

Sharp positive/negative moment bounds for exchangeable pairs with negative linear regression, with an exact universal exponent classification for that probabilistic setting. Positive-map reductions yield strict-vector and graph-homomorphism consequences for real exponents above two; the square-energy endpoint and target-projector construction are explicitly credited to prior work. Elementary Hermitian and residual-error corollaries are included.

The ordinary-vector conjecture and the complete graph exponent range below two remain unresolved. Written proofs, an exact five-state certificate and a separate floating-point grid have different assurance roles. Internal AI review and producer replay are not external peer review, formal verification or unaffiliated reproduction. The source comparison is bounded, not an absolute priority guarantee.

The three attached files are the full evidence/source package, the standalone paper and SHA256SUMS. Version DOI: https://doi.org/10.5281/zenodo.23996890