v1.0.0 — Reproducibility Package

CEMA causal sEMG reproducibility package v1.0.0

This is the first stable public release accompanying the manuscript:

Single-Scale EMA Latent Memory for Strictly Causal Dense Arm-Motion Recognition from Multi-Channel sEMG

The package includes:

• training and evaluation code for CNN, CT, and CEMA;
• fixed subject-independent splits;
• deterministic five-seed configurations and fold-level results;
• causal smoothing analyses;
• stream-level boundary sensitivity analyses;
• frozen-checkpoint EMA-memory diagnostics;
• statistical analysis scripts and environment records;
• source code, data tables, and reproduction files for Figures 1–4.

The raw MyoelectricData dataset is not redistributed. Please obtain it from the official MathWorks source described in DATA_SOURCE.md.

License: MIT.

CEMA reproducibility package v1.0.0-rc1

Audited reproducibility release candidate accompanying the revised manuscript. Includes deterministic five-seed configurations and results, stream-level boundary analysis, frozen-checkpoint memory diagnostics, smoothing controls, fixed subject splits, and statistical scripts. Raw MathWorks MyoelectricData files and trained checkpoints are not redistributed.