Latent Autonomic Regulation States: Dynamic Fingerprints and Cross-Dataset Generalization in Three Independent Datasets

This repository accompanies a preregistered secondary-data study on latent autonomic regulation states in physiological time series. The project tests whether multichannel physiological recordings are better described by switching state-space models with recurrent latent states than by a single non-switching baseline, and whether the retained states show measurable physiological fingerprints, partial cross-dataset correspondence, and partial frozen-definition transfer across independent datasets. Three datasets are analyzed: WESAD, CASE, and VitaStress. All datasets are processed using a common analysis logic based on protocol-grounded baseline–task–recovery segmentation, fixed one-second windows, within-dataset standardization, and candidate state solutions from K = 2 to K = 10 compared with a K = 1 baseline. The repository contains analysis scripts, derived one-second window files, model-fit summaries, inferred state sequences, state-level fingerprint tables, cross-dataset matching outputs, frozen-transfer outputs, figures, and implementation notes documenting dataset selection and deviations from the original preregistered dataset plan. Version 1.2: This version updates the manuscript and repository description to match the revised manuscript language and final reporting structure. The updated version clarifies the H1–H4 analysis sequence, replaces stronger invariance wording with cross-dataset correspondence where appropriate, reports the final WESAD, CASE, and VitaStress implementation, and documents the final state-fingerprint and frozen-transfer reporting used in the manuscript. Preregistration and full project materials are also available on OSF:Full projectPreregistration