Stability of ARGIRA's Fractal-D Feature and RMA-0 Audit: Experimental Validation and Reproducibility Package

Summary This repository contains the experimental validation and reproducibility materials for the stability of ARGIRA's fractal_D feature, together with the associated RMA-0 residual audit and control experiments. The study investigates whether the observed behavior of fractal_D remains stable across image scale, interpolation method, intermediate resolutions, and controlled transformations, and examines the relationship between the feature and image characteristics under the tested conditions. The repository contains the experimental artifacts required to inspect and reproduce the analyses, including source code, numerical results, figures, reports, and the image corpus used in the experiments. Scientific Scope The purpose of this deposit is to characterize the behavior, stability, and limitations of the investigated ARGIRA feature under controlled experimental conditions. It does not claim that fractal_D alone can determine the origin of an individual image, nor does it constitute a complete AI-image detection system. The results should be interpreted within the corpus, implementations, transformations, and experimental conditions documented in the included materials. Experimental Chain The deposit brings together seven documented experimental phases: Stability analysis of fractal_D across the 52-image corpus. Analysis of individual D curves and successive differences. Control of Lanczos interpolation at different scales. Comparison between bilinear and Lanczos interpolation. Analysis of intermediate resolutions. Residual transformation analysis. RMA-0 residual audit and control of image format/origin effects. The numerical results and conclusions are documented in the corresponding reports and machine-readable data files included in the deposit. Main Research Question The central question is whether the observed fractal_D signal represents a stable property of the analyzed images or whether its measured behavior can be substantially affected by scale, interpolation, resolution, or controlled image transformations. The experiments therefore examine the feature itself and the conditions under which its measured value changes, rather than treating a single numerical value as independent evidence of image origin. RMA-0 Audit The RMA-0 material included in this deposit concerns an audit of the predictability of the residual associated with the investigated feature. This should not be confused with the separate RMA calibration repository. The calibration work is documented independently in the RMA repository: https://doi.org/10.5281/zenodo.21935413 Reproducibility Package The complete reproducibility material is distributed inside a single ZIP archive. Because Zenodo displays the ZIP as a single file rather than exposing its internal directory structure as individually browsable files, the archive contains 100 files organized into functional directories. The archive contains: scripts/ — analysis and experimental scripts; datos/ — CSV and JSON numerical results; figuras/ — figures generated during the analyses; informes/ — formal experimental reports; corpus/ — the 52-image experimental corpus; fase_control_formato_origen/ — format/origin control experiment; fase_rma0/ — RMA-0 audit materials; pipeline_externo_sonificacion/ — original source used to verify the implementation of box_counting_dimension; README.md — scientific documentation and repository description; CITATION.cff — machine-readable citation metadata; MANIFIESTO.md — provenance and corpus documentation; MANIFEST.md5 — integrity manifest for the distributed files. The ZIP therefore constitutes the complete reproducibility package rather than merely an auxiliary attachment. Source Verification The repository includes sonify_painting_v3_5_7_FINAL.py because it contains the original implementation of box_counting_dimension used as the reference for verifying the fidelity of the implementation analyzed in this study. Its inclusion is therefore methodological and reproducibility-related, rather than incidental. Corpus The experimental corpus contains 52 images: 18 AI-generated images; 14 photographs; 20 paintings. The corpus and its provenance documentation are included in the reproducibility package. Relationship to ARGIRA v1.4.4 This deposit is a scientific companion study to ARGIRA v1.4.4 and provides experimental evidence concerning the stability and limitations of the fractal_D feature. ARGIRA v1.4.4 DOI: https://doi.org/10.5281/zenodo.21924218 The published ARGIRA v1.4.4 software itself is not duplicated in this repository. Relationship to the Origin-Discrimination Study This deposit is complementary to the independently published study on the vulnerability and controllability of ARGIRA's origin-discrimination features: https://doi.org/10.5281/zenodo.21945633 That study focuses primarily on origin-discrimination signals based on luminance and saturation, whereas the present deposit focuses on the stability of fractal_D and the associated residual audit. Intended Application in ARGIRA v1.4.5 The results of this deposit are intended to inform the continued development and evaluation of ARGIRA, including the planned evolution toward version 1.4.5. Publication of this research and implementation in the ARGIRA software are separate steps. The experimental evidence is first preserved in a stable, citable repository; subsequently, specific conclusions may be considered for incorporation into the interface, interpretation, or analysis logic of a future release. This deposit therefore does not itself constitute the implementation of ARGIRA v1.4.5. Scope and Limitations The experiments document the behavior of the investigated feature under the specific corpus, implementations, resolutions, interpolation methods, transformations, and experimental conditions described in the repository. They should not be interpreted as establishing universal behavior across all possible images, image-generation systems, datasets, resolutions, or processing pipelines. The findings characterize the tested behavior and its experimentally demonstrated limitations; they do not, by themselves, establish the origin of any individual image. Integrity and Provenance The repository contains an MD5 integrity manifest covering the 99 files other than the manifest itself. Together with MANIFEST.md5, the deposit contains 100 physical files in the ZIP archive. Repository Identification Title: Stability of ARGIRA's Fractal-D Feature and RMA-0 Audit: Experimental Validation and Reproducibility Package Author: Jose Ranero García Year: 2026 DOI: 10.5281/zenodo.21946989 License: CC BY-NC-SA 4.0 Related repositories: ARGIRA v1.4.4 — 10.5281/zenodo.21924218 Origin-Discrimination Feature Study — 10.5281/zenodo.21945633 RMA Calibration — 10.5281/zenodo.21935413