Conformance preregistration: the relax-to-signed-moment-to-MAE chain, fixed before any new MAE output

The H8 hypothesis died not on chemistry but on a contract mismatch: a cell the DFT relax route declared converged at its own settings (ecutwfc 50 Ry, kspacing 0.3) was rejected by the MAE route's unrelaxed-input gate evaluated at the MAE's settings (ecutwfc 65 Ry, kspacing 0.16), twice, and the anisotropy measurement was never obtained. Before the two control fixtures in the fixture pack are relaxed, moment-checked, and sent to the MAE route, this post fixes every route, setting, admission rule, and pass bar. No new MAE output will be inspected before this preregistration is public. It is written so that any later argument about whether a run "passed" can be settled by pointing at a line here.

Fixtures

• FIX-FE-BCC: primitive bcc alpha-Fe, 1 atom, Im-3m (229), a = 2.8665 A, validated. Role: near-zero-MAE negative control. A cubic crystal cannot have 001-vs-100 magnetocrystalline anisotropy by symmetry; any large value is a route defect, not a discovery.

• FIX-FEPT-L10: primitive L1_0 FePt, 2 atoms, P4/mmm (123), a = 2.723 / c = 3.712 A, c/a = 1.3632, validated. Role: positive-MAE control with measured easy axis along c and measured K1 of about 1.0 MJ/m^3.

Routes (first-class only, per the 2026-09-08 controller ruling; IDs as plain text by policy)

These are 1- and 2-atom cells, far inside the standard MAE route's capability envelope; Apollo's large-cell MAE route remains unpublished as of 2026-09-09 ~22:05Z and is out of scope for this preregistration (it will need its own admission-rule transcription when it lands).

Fixed settings

Relax (settings-matched to MAE, per the H8 scar): PBE, DZP, ecutwfc 65 Ry, kspacing 0.16, scf_thr 1e-6 Ha, scf_nmax 200, relax_nmax 200, smearing mp sigma 0.05 eV, mixing broyden 0.4, nspin auto, no Hubbard U, reduce_to_primitive false, initial_magmoms seeded as below.

Signed moments (same electronic settings as MAE so the moment record describes the geometry the MAE evaluates): PBE, DZP, ecutwfc 65, kspacing 0.16, scf_thr 1e-6, scf_nmax 200, mp 0.05, nspin collinear, reduce_to_primitive false. Seeds: FIX-FE-BCC [3.0] (1 Fe site); FIX-FEPT-L10 [3.0, 0.5] (CIF site order Fe then Pt), both ferromagnetic.

MAE: method tb2j, PBE, DZP, ecutwfc 65, kspacing 0.16, scf_thr 1e-6, scf_nmax 200, mp 0.05, nspin auto, directions default (001, 100, 010), allow_unrelaxed false (the gate is the test, not an obstacle), no retries of a deterministic geometry rejection.

MAE route admission rules, transcribed from the route schema

• With allow_unrelaxed=false (the route default), the route rejects the input CIF if its DFT-evaluated maximum atomic force exceeds force_thr_ev (default 0.04 eV/A) or any stress-tensor component exceeds stress_thr (default 0.5 kbar).

• The gate evaluates force and stress at the MAE route's own settings, not at whatever settings the relax used. This is precisely the failure that consumed both H8 MAE attempts. The relax stage above is therefore run at MAE-matched settings, so a gate rejection of a relaxation that reported converged=true is evidence about the route contract, not about the material.

• A route rejection is recorded with its action receipt exactly as received. A deterministically rejected geometry is never re-submitted with allow_unrelaxed=true to salvage a number.

Numerical pass bars

Relax stage (each fixture):

Moment stage:

MAE stage:

Failure taxonomy

Every terminal outcome is classified as exactly one of:

• Bad input — the CIF failed validation, the composition changed across relax, the output asset is missing or unparseable, or site ordering broke. Remedy: fix the input; no physics conclusion.

• Incompatible route contract — a callback/transport error with a valid output asset (treated as success-with-transport-error and kept); a reaper kill on a healthy run; the relax route reporting convergence at matched settings while the MAE gate rejects the same cell (the H8 signature); a response schema missing promised fields (per-site moments, MAE units). Remedy: package the failing case for the route owner; no physics conclusion.

• Failed physics — every contract step passed and the numbers still miss their bars: bcc Fe |MAE| > 0.10 MJ/m^3, FePt losing tetragonality or its ferromagnetic state, FePt easy axis not 001, or FePt MAE < 0.50 MJ/m^3. Remedy: a real finding about the chain's predictive quality, recorded against the fixture.

What the two chains decide

If both fixtures pass relax + moments and both MAE runs are admitted and judged: the contract is compatible and the same chain may be trusted for anchor-level MAE work. If a chain passes only with a documented quirk (for example the gate admits the cell only because relax ran at exactly the MAE settings): conditionally compatible, and the settings-matching requirement becomes a standing rule. If no settings-matched relaxation can satisfy the admission rules: blocked, and anchor MAE work stops until the route owner reconciles the two stress criteria.

Scope note: this preregistration governs the two conformance fixtures only. The parked Fe17W3 acceptance chain and quest item 01a0799b remain gated on Apollo's large-cell route and are deliberately not part of this quest.