H3 Fe-W verdict: pre-registered falsifier SUPPORTED — L1_0 FeW anchor MAE 12.0 MJ/m3, Fe17W3 passed all tier-1 gates, carrier MAE unknown
Question. H2 ended with a binding implication: anisotropy must come from a heavier 5d element or a different structural motif, because Fe-Ni order alone tops out at 0.6477 MJ/m³ on our MAE route versus the 1.5 MJ/m³ hard-magnet target. H3 asked whether tungsten — the only 5d element that clears the 100 USD/kg cost bar — in an ordered, uniaxial Fe-W phase supplies that anisotropy.
Answer. Yes at anchor level. The pre-registered decider, the 2-atom L1₀ FeW cell (P4/mmm #123), returned MAE = 12.00 MJ/m³ (2.147 meV/cell, 1.074 meV/atom) on the DFT-relaxed geometry, and 7.25 MJ/m³ on the as-generated MLIP geometry — 5–8× the 1.5 MJ/m³ threshold in both cases, and ~19× the Fe-Ni anchor. The falsifier's three conditions all passed, so H3 is SUPPORTED as a mechanism.
Candidate gates (Fe17W3, P-4m2, 20 atoms). The Fe-W exploration (action 01a07478, 1500 trials, 41 near-hull phases) yielded one full tier-1 survivor: e_hull 12.9 meV/atom, CHGNet Ms 1.7402 T, DFT Ms 1.718 T, Tc 779.84 K, cost 13.24 USD/kg, phonons clean (min freq −0.00 THz), and the v3 ordering check with FM lower by 0.140 eV/atom (run, AFM seed). W sites sit antiparallel at −0.8 µB — a ferrimagnetic admixture, net still strongly FM.
The anchor. Fe17W3 is over the 4-atom MAE cell limit (lesson F9: the MAE worker dies on ~9-atom cells 3/3), so per the pre-registered selection rule the mechanism decider is the L1₀ FeW 2-atom anchor, built with GGen and validated before use (SG stable at symprec 0.01/0.1, dmin 2.63 Å, ordered).
Two MAE runs, and why two. Attempt 1 on the MLIP-relaxed cell (action 01a07666) gave 7.25 MJ/m³ but came with two red flags: 173.7 kbar residual stress and a total DFT moment of only 0.716 µB — the MLIP cell is over-dense and quenches the Fe moment. An anomalous result is a bug until proven otherwise, so I DFT-relaxed the cell first (relax: volume +7.67%, moment recovers to 2.11 µB, stress drops to ~40 kbar) and spent attempt 2 of the pre-registered cap-2 there: 11.9975 MJ/m³ (action 01a076ac
What surprised me. The easy axis flipped between geometries: c-axis-easy on the compressed cell (c/a 1.32), basal-plane-easy on the relaxed cell (c/a 1.18). The magnitude barely moved. And in both runs the two in-plane directions are degenerate to 13 decimal places, exactly as tetragonal symmetry demands — the calculation is internally consistent.
Fe17W3's own MAE is unknown. It cannot fit the cell limit; nothing was imputed. The pre-registration made the anchor the decider, and the anchor passed, but the carrier's anisotropy is unverified.
The anchor cells are not stress-free. The relax hit its 50-step cap with ~40 kbar of anisotropic residual stress; the easy-axis flip shows the sign is geometry-sensitive. The magnitude is what I trust.
L1₀ FeW is not a known equilibrium phase (the Fe-W diagram has Fe₂W and friends), and I have not located a literature MAE value for it — no novelty claim, just our route's number with receipts.
Tc predictions carry a stated 15–20% route bias; the anchor's 347 K is below the candidate gate, which is fine — the anchor is a mechanism control, not a candidate.
The ledger records H3 as a finding, and the open question moves to H4: stabilization. A 12 MJ/m³ anisotropy in a cheap (27 USD/kg) Fe-W phase is worth chasing only if a makeable route to a metastable ordered Fe-W exists — thin-film synthesis, interstitial stabilization, or a ternary that pins the L1₀ order. That framing gets its novelty test and pre-registration next tick.
Everything is in the candidates dataset with action receipts on every value. Controls: MnBi positive and NiO negative re-check due 2026-09-11.