A Validation Protocol for Structural Similarity Metrics Using Null Models and Block-Shuffle Analysis Application to 1D Spectral Signals

This repository contains a validation note and experimental results for a general protocol aimed at assessing whether a structural similarity metric applied to 1-D signals captures genuine multi-scale organization or artifacts driven by global statistical properties. The protocol is based on a systematic comparison between real signals and multiple null models, including: phase randomization (PSD-preserving), IAAFT surrogates (preserving both PSD and amplitude distribution), block-shuffle perturbations with variable block size. A block-size sweep is used to probe scale-dependent degradation of structural similarity and to identify characteristic coherence lengths in real data. The validation framework is metric-agnostic and can be applied to any structural similarity measure.In this repository, the protocol is demonstrated using SISSI-H3 as a reference implementation, without releasing its internal algorithmic details. Contents The repository includes: a PDF validation note describing the protocol and methodology, detailed experimental results (tables and summaries), documentation of null models, parameters, and reproducibility settings, integrity hashes for all released artifacts. Scope and limitations This work does not propose a new similarity metric.Its purpose is to provide a reproducible and audit-safe validation methodology to distinguish genuine structural organization from statistical artifacts in 1-D signals. The reference metric (SISSI-H3) is proprietary and is not distributed in this repository.Only validation procedures and results are released. Relevance The protocol is relevant for researchers and practitioners working in: spectroscopy and chemometrics, signal processing and quality control, anomaly detection and structural analysis of 1-D data. Citation If you use or adapt this validation protocol, please cite this repository.