A Bounded Reproducibility Package for the HSIT Program: Corrected Claims, Finite Regressions, and Open Controls

This revision corrects and narrows the reproducibility claims of an earlier methodology paper concerning a finite weighted Farey-Laplacian component of the H SIT program. The historical public record contains a PDF but not the complete executable object described by that paper. We therefore stage a private, self-contained executable replay for the stated finite checks, with source, tests, pinned package versions, generated JSON, provenance, license disposition, review records, and an exact-scope SHA-256 manifest. The external mathematical proof sources are identified but are not contained or verified by the executable replay. The bundle evaluates a one-dimensional kernel, a finite Mertens trace, weighted-degree and operator-norm bounds, and a finite Commutator-Pythagoras identity at Farey levels 8, 16, 32, and 64. Across pseudoinverse cutoffs 1e-8, 1e-10, 1e-12, and 1e-14, the largest absolute identity residual is 2.0641e-12. These rows are numerical regressions against proof-owned statements, not proofs. A fixed-seed, unfitted magnetic-operator baseline at level 64 retains 820 bulk modes and 819 nearest-neighbor spacings, has Kolmogorov-Smirnov distance 0.727395592237 from the GUE Wigner surmise, and produces rounded-eigenvalue hash prefix 308fe8684567ce7b. This is one descriptive finite configuration, not an ensemble law or zeta-zero result. The primary logical correction concerns an infinitely-often deficit floor. Finite all-below-floor observations cannot contradict a subsequence theorem unless a separate proof establishes floor-attainment coverage for the exact evaluated set, epsilon, and constant. Because a JSON object cannot verify such a proof, the code never emits or authorizes THEOREM-CONTRADICTION. Schema-complete coverage metadata can produce only CONDITIONAL-INCONSISTENCY, pending independent proof and exact-numerical review. The staged artifact includes a fresh disposable-environment reproduction. CPython 3.12.10, NumPy 2.4.6, threadpoolctl 3.6.0, and OpenBLAS 0.3.31.188.0 produced two byte-identical result files with SHA-256 9f48dadbe876dc29ae8cc64c646b49641d3ce81024f743098adf6bb4ceeebfd1; all 22 tests passed, including three source-identity and claim-anchor tests. This establishes a second local environment build on the same Windows host, not an independent laboratory reproduction or a cross-platform guarantee. The zero-data analysis is permanently omitted from this revision. Holdout, fake-target, budget-equivalence, and all-runs controls remain unimplemented, and no trained result is reported. Nothing in this artifact proves the Riemann Hypothesis or physical Super Information Theory.