Residual Mapping Auditor (RMA): A Reproducible Confirmatory Feature-Validation Pipeline for Symbolic MIDI Features (MAESTRO)
Abstract
Repository maintenance Prior to public release, the repository was reorganized to improve clarity and reproducibility. Temporary files were removed, redundant artifacts were consolidated where appropriate, and the project structure was simplified. These maintenance changes did not modify the datasets, experimental procedures, or reported results. About this project This repository documents the Residual Mapping Auditor (RMA), a reproducible audit pipeline for the analysis and validation of features extracted from symbolic MIDI representations of music (MAESTRO dataset). It includes feature extraction code, multicollinearity screening (VIF and hierarchical clustering), a confirmatory RMA experiment (EXP-006/EXP-006C), and a practical framework for confirmatory feature validation, together with full traceability of datasets, code versions, and results. The primary confirmatory result concerning `duration_std`—the only feature in `feature_set_v1` that satisfied all three confirmation criteria in the EXP-006C execution included in this repository—was obtained under a reduced ("fast profile") parameter regime rather than the canonical full-LOO protocol originally specified. This deviation, and its implications for interpreting the result, are documented in `KNOWN_ISSUES.md` (Issue 002) and preserved as explicit metadata in each run's generated configuration file. See `README.md` and `RMA_PROJECT_LOG.md` for methodological details, and `KNOWN_ISSUES.md` for open limitations and reproducibility notes. Package contents rma_project_FINAL/├── README.md Overview and project documentation├── README_EXP004.md EXP-004 (feature comparison)├── README_SESSION_HANDOFF.md Historical session handoff index├── RMA_PROJECT_LOG.md Complete project log (source of truth)├── KNOWN_ISSUES.md Open limitations and reproducibility notes├── CHANGELOG.md├── CITATION.cff├── LICENSE├── requirements.txt├── rma_core_v3.py Core RMA methods (including compute_vif)├── features_16.csv / correlation_matrix_16.csv / dendrogram.png├── exp004_*.csv / exp004_figure01_classification.png│└── rma_audit_package/ ├── EXP-006_PREP.md ├── exp006_basic.py ├── exp006_confirmatory.py Public confirmatory script (preserves │ variant_note / │ reduced_params_amendment in │ config_used.yaml) ├── exp006_run_single_feature.py ├── exp006_traceability.py ├── features.py ├── midi_parser.py ├── rma_audit_harness.py ├── run_pipeline.py ├── maestro-v3_0_0.csv ├── rma_audit_results_full.csv ├── rma_audit_results_nontrivial.csv │ ├── notebooks/ │ └── exp006_colab.ipynb EXP-006C Colab notebook │ ├── exp005_validated_dataset_n140/ │ ├── original_features.csv │ └── quantized_features.csv │ ├── audios/, audios1/ MIDI corpora (n=140, n=50) ├── results/, results1/ Results associated with audios/audios1 ├── results_audit/ Sanity checks, VIF and PCA dominance │ ├── results_exp006/ │ ├── exp006_basic_vv0_1_20260703T200507Z/ │ │ EXP-006 basic execution │ ├── exp006_confirmatory_EXP-006C_fast_colab/ │ │ ├── exp006_summary.csv Primary confirmatory result │ │ ├── exp006_rma_results.csv │ │ ├── exp006_permutation.csv │ │ ├── exp006_bootstrap_ci.csv │ │ ├── exp006_order_stability.csv │ │ ├── config_used.yaml │ │ ├── dataset_hash.txt │ │ ├── code_version.txt │ │ ├── timestamp.txt │ │ ├── feature_list.txt │ │ └── run_stdout_stderr.log │ └── exp006_confirmatory_v*/ │ Six truncated confirmatory attempts │ (see KNOWN_ISSUES.md, Issue 003) │ └── resultados2paper/ ├── EXP-005_REPRO/ Independent forensic validation ├── EXP005_Cluster_Stability_Report.pdf ├── EXP005_Final_Interpretation_and_Project_Organization.pdf ├── RMA_Analysis_Report.pdf ├── RMA_Project_Context_and_Next_Steps.pdf ├── vif_results.csv ├── pca_variance.csv ├── cluster_metrics.csv └── RMA_FULL_PROJECT.zip RMA_REPORTS.zip RMA_DATA_RESULTS.zip (historical backup packages; not required to reproduce EXP-006/EXP-006C) Minimal reproducibility path rma_audit_package/├── exp006_confirmatory.py├── exp005_validated_dataset_n140/└── results_exp006/ └── exp006_confirmatory_EXP-006C_fast_colab/ License This repository is released under the Creative Commons Attribution–NonCommercial–ShareAlike 4.0 International (CC BY-NC-SA 4.0) license. This license applies to the original source code, documentation, and original materials authored for this repository. This project is released under CC BY-NC-SA 4.0 to encourage open scientific research, reproducibility, and collaborative improvement while preventing unauthorized commercial exploitation. Commercial use requires prior written permission from the copyright holder. Third-party datasets and external resources referenced or included in this repository remain subject to their respective original licenses and are not relicensed by this repository. In particular, the MAESTRO dataset (Google Magenta) is distributed under its own CC BY-NC-SA 4.0 license and is not relicensed by this repository. See `LICENSE` for full details. How to cite This record: https://doi.org/10.5281/zenodo.21191746 Please cite both this Zenodo record and any third-party datasets (e.g., MAESTRO) according to their respective citation guidelines when applicable. See `CITATION.cff` in the repository for complete citation metadata.
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 4 July 2026
- Language
- EN