Verification Evidence for Certified Prediction Improvement under Matched Contact Occupancy and Switching Flux in a Finite-State Model

This dataset contains certificate records, verification specifications, preserved research code, and provenance records supporting the preprint Certified Prediction Improvement under Matched Contact Occupancy and Switching Flux in a Finite-State Model, by Satoshi Kawasaki, DOI: 10.6084/m9.figshare.33643042.The main CAP4 evidence covers the continuous parameter box 0 ≤ ε ≤ 1/2, 0 ≤ θ ≤ 1, and 63/2048 ≤ g ≤ 65/2048 for the specified 32-state continuous-time Markov model. Its 18,126 accepted closed boxes were verified in two stages: 10,917 in the original verification and 7,209 in the continuation using the frozen verifier. The integrated records identify both accepted sets and their complete covering. They support G ≥ 50425539063434482373081891813729 / 2¹²⁵ > 0. The 27 diagnostic points, the saved P01 point enclosure, and sampled-box checks are supplementary records and do not replace the full-domain proof.Separate CAP2 evidence contains 256 accepted boxes supporting the stronger bound G ≥ 1052/10⁶ on the smaller band 0 ≤ ε ≤ 1/2, 0 ≤ θ ≤ 1/16 at g = 1/32. The post-review selection of this band and its covering records is disclosed. The CAP2 bound is supported by its own records, not inferred from the CAP4 constant.This item includes RWR_PAPER2_EVIDENCE.zip and the standalone EVIDENCE_MAP.md. Begin with the map or README.md inside the archive. The archive separates mathematical acceptance conditions from file-identity and packaging checks. Historical verifier sources are preserved, but their commands are not guaranteed to run unchanged against this archive in a new environment. No fixed verification time is promised. Preparation of this release assembled existing records and did not run a new scientific verification. AI-assisted derivations, implementations, and reviews are disclosed in the preprint; no completed external human or institutional reproduction, proof-assistant formalization, or measured energy saving is claimed.Author-controlled material is licensed under CC BY 4.0, to the extent of the rights held by Satoshi Kawasaki. The 23 research Python files individually identified by archive-relative path and full SHA-256 in CODE_LICENSE_INDEX.json are additionally offered under the MIT License. Recipients may choose CC-BY-4.0 OR MIT for those exact files; compliance with both is not required. LICENSE.md and LICENSE-MIT.txt state the current grant. Earlier dated license-pending records are preserved as historical provenance and do not replace this release designation. Third-party rights and terms are unchanged; external interpreter, library, and typesetting distributions are not bundled.