Our shortlist contains 19 candidates below 20 meV/atom on the model hull: 17 rare-earth-free and two rare-earth-containing. Three contain cobalt and 16 are Co-free. The list balances energy, predicted FM-aligned Jₛ, repeat evidence and chemical/framework diversity; it is a screening portfolio, not a hard-magnet performance ranking.

The ledger contains 2,153 proposed designs and 2,024 routine relaxation attempts. There are 1,013 reconciled routine hulls, with 77 initial routine endpoints below 20 meV/atom. Separately, 124 independent confirmation relaxations produced 96 geometry passes. The 306 paired hull/Jₛ comparison endpoints include 145 below 20 meV/atom, across history, controls and repeats; these are not counts of distinct novel frameworks.

Fe46Si8B20 leads on model energy at 0.000 meV/atom and predicted Jₛ 1.2761 T. Fe72Si12B32 follows at 1.674 meV/atom and 1.2807 T with completed selected-library evidence. Fe60Co2B32 intergrowth offers 7.518 meV/atom and 1.6419 T. Fe126W11B40 is a Co-free Fe-rich alternative at 9.264 meV/atom and 1.5645 T. The Nd16Fe112B4 clustered ordering reaches 12.435 meV/atom and 1.6741 T, with known Nd2Fe14B metal-scaffold ancestry.

Ti6Fe23B8 remains a selective magnetic-validation priority. The researcher reports directional MAE +0.888 MJ/m³, Jₛ 0.790 T and κ≈1.34 for the named composition. Its plotted model values are 5.074 meV/atom and 1.0662 T. The external exact CIF and raw DFT files are unavailable, so these external values are kept separate from our exact-CIF model results.

Energy–magnetization Pareto views

Fourteen points are numerically non-dominated across all 306 paired endpoints; this includes binary controls, repeats and held results. The five-point shortlist frontier below contains the selected energy–Jₛ tradeoffs. Novelty is categorical evidence, not an invented numerical score. Differences of a few meV/atom are not established physical advantages.

Open the updated Project 14 Pareto views

All 19 shortlisted candidates

Letters A–S match the charts and link to exact relaxed CIFs. The borosilicide S remains a provisional diversity lead. Confirmation-sensitive structures and known-parent derivatives are explicitly labeled; a new filling or ordering does not automatically count as a new underlying framework.

Compare all 19 shortlisted candidates

Novelty and confirmation evidence

Fe₄₆Si₈B₂₀ remains at 0.000 meV/atom with predicted FM-aligned Jₛ 1.2761 T. The original exact relaxed CIF and both independent Orb/hull repeats now pass their complete declared finite novelty scopes: 2,301 frozen library representatives and the selected 108-entry AFLOW database scope for each exact CIF. The repeats are at 0.001 and 0.005 meV/atom. All 4,602 repeat-library comparisons are resolved as non-matches, including eight recovered timeout directions; none remains pending or unknown. These checks cover occupied prototypes and species-erased occupied frameworks. They do not prove global novelty or a new parent scaffold.

The database scope covers stored atom-count multiples of 74 through the queried maximum and exact B–Fe–Si ternaries; other stored atom counts, continuous prototype parameters, disorder realizations and complete literature are excluded. The representative library supplements these checks without exhausting the exclusions. The AAAB structure joins known Cr₅B₃-type and V₈SiB₄-type building blocks: the hypothesis concerns their occupied junction sequence, not newly discovered building blocks.

Fe72Si12B32 (AAB) and both exact independently relaxed repeats have completed the selected 2,301-representative library audit: 6,903 candidate–reference checks, all non-matches, zero unresolved and zero flagged. The original timeout was resolved on its own exact CIF with unchanged matcher tolerances; repeat verdicts were not substituted. This is a finite library pass, not a complete database/literature audit or proof of a new parent framework. Known Cr5B3-type / V8SiB4-type (M5/M8) building blocks remain explicit. Original and repeat hulls are 1.674, 1.669 and 1.671 meV/atom; original predicted FM-aligned Jₛ is 1.2807 T, with repeat Jₛ unmeasured. No new family credit. The Co/Nb and C16 scopes remain incomplete.

The four AAB/ABB Si/P junction candidates each have two independently relaxed endpoints below 20 meV/atom that pass the primary tetragonal and 3D gates. The Si AAB original and both repeats additionally pass the selected finite library audit. The other three occupied-framework novelty assessments remain unresolved. Their known Cr5B3-type and V8SiB4-type building blocks confer no new parent-family credit. Repeat Jₛ was not calculated.

Routine novelty screening uses Ouro same-formula matching plus local ancestry/history. Exceptional candidates receive explicit finite AFLOW audits. Same-formula non-matching does not establish a new underlying framework. The Co/Nb selected-library run for Nb8Fe81Co2B26 and its two repeats has ended: 6,738 of 6,903 comparisons are resolved non-matches, with 165 unresolved and zero positive flags. This is an incomplete novelty assessment, not a pass. The separate C16 database audit is still in progress; no pass is inferred there either.

The original Fe18CoSi8B4 endpoint is SG111 at the primary 0.01 Å tolerance. Its two repeats pass that primary tetragonal and 3D gate with hulls 16.048 and 16.046 meV/atom; both lose tetragonal symmetry at 0.005 Å. Framework novelty, repeat magnetization and MAE remain unmeasured or unresolved.

Holds and controls

Ti2Fe18B and the other provisional Ti site-filling lead remain on hold after their independent repeats lost primary tetragonal symmetry. The approximately 1 meV/atom Al-rich W–B candidate remains an energy benchmark after user-reported weak MAE; no unsaved numeric MAE value is invented. Binary Fe–B controls remain visible in the full plot, with no assumed permanent-magnet performance.

What changed in this update

All 306 paired endpoints and 2,291 current trial entries have been reconciled. Six newly completed hulls and four new magnetic predictions add no sub-20 candidate. The shortlist retains 19 candidates; the full numerical frontier contains 14 points and the shortlist-only frontier five. Both frontiers are unchanged.

The eight R8Fe88X4B8 fillings/orderings (R = Nd or Sm; X = Co or Ti) are now fully screened. All converged and remained 3D-connected. Six retained primary tetragonal SG122 and received exact-CIF hull calculations; all six are above target. The four Co variants span 31.387–32.723 meV/atom with predicted FM-aligned Jₛ 1.8157–1.8240 T. The two surviving Ti variants are at 55.444 and 57.219 meV/atom; both lose tetragonal symmetry at 0.005 Å, and magnetization was not allocated under the existing 40 meV screening gate. Two other Ti variants became orthorhombic SG43. These are eight orderings on the existing RE cage-vacancy support, with computational Y2Fe23B3 ancestry. No new-framework credit; these fixed recipes are retired.

The four Co variants also fail to improve the recorded parent endpoints on either model objective: the Nd parent is 29.432 meV/atom and 1.8352 T, and the Sm parent is 30.421 meV/atom and 1.8406 T. Historical reference inventories were not synchronized, so small hull differences are screening observations rather than established physical advantages. Rare-earth magnetic estimates do not validate orbital or sublattice contributions.

Searchable results, exact CIFs and complete history

The report contains all 306 paired endpoints, 19 shortlisted candidates, 135 public exact-CIF links, independent repeats and failed-trial evidence. Filter by composition, rare-earth content, cobalt, energy threshold or held results. Older outcomes remain in the evidence archive; the JSON preserves the numeric snapshot and evidence labels.

Open the searchable 306-endpoint report

Download the frozen comparison data (JSON)

Interpretation and next priorities

Energy and FM-aligned Jₛ are Orb/Ouro screening predictions. They do not establish coercivity, anisotropy, magnetic order, finite-temperature stability or synthesizability. Zero model hull is not a synthesis guarantee. Rare-earth orbital and sublattice contributions are not validated by these aligned-moment estimates. Selective MAE work remains with the researcher; REMAG stays paused.

Maintain parallel searches across Si/B junctions, intergrowth, Co-free Fe-rich W/Nb borides, Ti systems and rare-earth-containing candidates. Cobalt remains eligible. Favor plausible anisotropy mechanisms alongside low energy and high Jₛ; the user handles selective MAE validation. Pending hulls and incomplete novelty audits receive no promotion.