Identifiable Memory-Rank Protocol: Code and Reproducibility Artifacts

Abstract

This release contains the versioned software, frozen experiment records, automated tests, publication figures, and bilingual manuscript sources associated with an identifiability-aware method for selecting shared finite-memory models from sparse grouped observations. The method fits candidate positive-rate realizations with rates shared across independent specimens or material groups and unit-specific amplitudes and offsets. A candidate order is retained only when information gain, held-unit early-to-late prediction, foldwise log-rate stability, and adjacent-rate resolution support the same interpretation. When these criteria disagree, the model order is reported as unresolved. The archive corresponds to GitHub commit 13001874f788f9de9b49632961166d3713561f7b. It includes source code, 151 automated tests, frozen machine-readable results, experiment drivers, vector figures, and English and Chinese AMM manuscript sources. Third-party public datasets are not redistributed. Persistent source identifiers, frozen SHA-256 digests, and a verified downloader are included so that the public inputs can be retrieved from their authoritative repositories. The software is released under the MIT License. Dataset licenses and attribution requirements remain those of the original data providers.