# License scope

Original Python/JavaScript/HTML/CSS implementation and authored documentation: Apache License2.0, Copyright2026 Renchao Wu. Full text: [LICENSE](LICENSE). Basic MLP/canonicalization patterns reuse the author's own Apache2 AirfoilSplitLab code; this study trains new arrays on a distinct public dataset.

Original Wine Quality CSVs/names and ZIP: Cortez, Cerdeira, Almeida, Matos and Reis (2009), UCI186, DOI10.24432/C56S3T, CC BY4.0. Full text: [data/LICENSE-CC-BY-4.0](data/LICENSE-CC-BY-4.0); exact source/byte provenance [SOURCE.json](SOURCE.json). Original three ZIP members unmodified. The standard CC BY4.0 legal text was reused byte-exact from an already archived standard license, not represented as a fourth ZIP member.

Derived membership/catalog/sourceaudit, fit arrays/normalizers/weights/traces, validation/test predictions/metrics/paired errors, figures, and recorded public-source/custom-demo input reports: CC BY4.0 with source attribution and identification of changes in README. This does not claim original dataset ownership or endorsement. Changes include input-only global duplicate-group splitting, label-blind equal-budget source-domain sampling, six fresh MLP fits, cross-domain evaluation and browser demos; no source row relabeling or deletion. Original authored synthetic gradient/large-bias/malformed-input tests are code/tests under Apache2.

No pretrained model, third-party runtime implementation, paper text/figures/code, employer/private source or competition submission is bundled. NumPy and optional Matplotlib are separately installed dependencies. The original names file is unmodified licensed dataset documentation, not newly copied paper text. This is not a reproduction of the2009paper's model or score.
