Reproducibility Code for Memory Kinetics and State-Dependent Contact in Finite Stochastic Predictors

Abstract

This MIT-licensed software archive supports the preprint “Memory Kinetics and State-Dependent Contact in Finite Stochastic Predictors” by Satoshi Kawasaki, Independent Researcher. It contains the Python source required to replay the finite Markov mechanism comparisons, preserved historical helper code, the portable replay entry point, dependency requirements, code-specific instructions and a SHA-256 file manifest. The related CC BY 4.0 data record supplies the specifications, stored results, historical result dependencies, ledger and full validation instructions. Both archives are required. Extract both into the same new working folder; their files do not overlap. Follow README.md and CODE_README.md. The data and paper are not relicensed under MIT by this code record, and the code is not relicensed under CC BY 4.0 by the data record. The preserved original runners retain their historical execution guards. A separate replay entry calls their calculation functions and writes to a fresh destination. Exact rational outputs are compared exactly; approximate stationary outputs use absolute tolerance 1e-10 plus relative tolerance 1e-9, and accounting outputs use absolute tolerance 1e-8 plus relative tolerance 1e-9. Same-runtime byte agreement is a supplemental check rather than a cross-platform guarantee. This release changes distribution and licensing, not the scientific models or registered numerical results. The code and checks were developed with substantial generative-AI assistance. The related manuscript describes the roles of ChatGPT/Codex, Claude and Gemini, author-reported model labels, incomplete historical OpenAI model-version records, and recorded author-verification scope. Computational checks do not certify human or external independent reproduction, journal peer review or benchmark-level AI performance. Private correspondence, full AI-review conversations and future research plans are not included. Related records: preprint, https://doi.org/10.6084/m9.figshare.33472102 ; required data and specifications, https://doi.org/10.6084/m9.figshare.33472243 . These DOI identifiers were reserved before release and become active when the corresponding records are published.