Reproducibility Artifacts for Sealed Evaluation and Prospective Attribution in TSI Transition Models

This dataset provides the evidence and reproducibility artifacts supporting the manuscripts “Sealed Evaluation of Structural World Models: OOD Generalization, Open-Loop Rollout, and Learned-Routing Criterion Validation” (Paper 03) and “Prospective Attribution of Graph Information and Head Factorization in TSI Transition Models” (Paper 04). The release contains sealed raw results and manifests for the criterion-validity study; a shared prospective cohort of 120 independent synthetic worlds used to evaluate graph recovery, transition-head family recovery, out-of-distribution composition, and open-loop rollout; a second frozen cohort containing unmodeled nonadditive synergy; frozen source snapshots; root-seed and integrity ledgers; portable clean-room inputs; post-review audits; tests; and bilingual reproduction instructions. Endpoints reported by Papers 03 and 04 from the shared 120-world cohort belong to one frozen analysis and must not be counted as independent evidence. The Node.js clean-room implementation was written by the same author as the benchmark and therefore provides a reproducibility check, not an external independent replication. Post-review sensitivity and retrospective power analyses are explicitly distinguished from the frozen confirmatory analysis. This release supports verification and reuse of the reported synthetic experiments. It does not establish real-world validity, unconstrained neural structure discovery, universal model superiority, or external independent replication. Source code is licensed under the MIT License. Experimental data, manifests, reports, and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).