MAGNET-GEN: A Lakatosian, Pre-registered Research Programme for the Accelerated Design of Rare-Earth-Free Permanent Magnets — Phases, Articulations, Falsification Results and VV&UQ Assessment

Europe imports about 98% of its permanent magnets from a single supplier country, at the very moment when the Critical Raw Materials Act (Regulation (EU) 2024/1252) imposes benchmarks of 10% extraction, 40% processing and 25% recycling by 2030, and when successive Chinese export-control announcements (No. 18 of 4 April 2025; Nos. 55–58 and 61–62 of 9 October 2025; suspension No. 70 of 7 November 2025, expiring on 10 November 2026) have turned rare-earth supply into an instrument of industrial policy. This preprint reports a complete, auditable research campaign — MAGNET-GEN — that subjects a three-level, physics-informed quantum modelling paradigm for the accelerated design of permanent magnets to the discipline of a Lakatosian research programme: a sealed antecedence gate (P0) with a mechanical scoring function, pre-registered falsification tests executed exclusively on real, publicly identified datasets (Zenodo 23361847; Mendeley Data 10.17632/rvxsxsy9yw.2), verification–validation–uncertainty-quantification (VV&UQ) after Thacker et al. (2004) and Roy and Oberkampf (2011), and verdicts that were never adjusted a posteriori. Of the five sealed hypotheses, two were stopped at the gate by documentary antecedence, one was reformulated, and two progressed. Of the seven computational tests, three were refuted (mean-field Curie temperatures, Stoner–Wohlfarth hysteresis, out-of-distribution transfer) and four were not refuted (a 76-feature ridge surrogate, R² = 0.824; finite-size scaling of the 3D Ising model, Tc(∞) = 4.5071 ± 0.0412 against the reference 4.5115, a 0.10% deviation; residual-bootstrap uncertainty quantification, 90.5% coverage; bibliographic anchoring on 70 sourced values). Sealed re-tests confirmed the out-of-distribution failure (H-E2, H-F2) and validated a ferrite extension (H-G2). The campaign yields two patent technical notes (twenty claims each), a costed TRL-5 acceleration plan (T1–T4, €58,340, six weeks), and an application form for the Horizon Europe topic HORIZON-CL4-2027-01-MAT-PROD-06. The scientific contribution is twofold: a demonstrated methodology that makes a materials-modelling programme falsifiable end-to-end, and a quantified map of what survives falsification — and what does not — in data-driven coercivity prediction. Keywords: permanent magnets; rare-earth-free magnets; coercivity; Lakatosian research programme; falsification; pre-registration; verification, validation and uncertainty quantification (VV&UQ); machine learning; out-of-distribution generalisation; micromagnetism; Ising model; Critical Raw Materials Act; technological sovereignty. Version 2 (extended), November 2026 — 92 pages, 11 figures, 6 tables. Preprint prepared for arXiv submission.