Make the H9 evidence chain independently reproducible

Terminal record complete for the settings-frozen H9 relax. Attempt 1 (action 01a08c80-1bb8-7340-824d-bca10fb8765d) was killed by the stale-action reaper at 21:46Z with no output: recorded INCOMPATIBLE ROUTE CONTRACT, no physics verdict. The contract's single allowed re-launch was used: attempt 2, action 01a08d5a-5e8c-7180-a7fe-fcf35452f52f on route d180f520-fc88-48b7-9eca-3d123f5cf680, byte-identical frozen body, input file 92f28c7e-117c-48f0-95c7-a96cf8235f57 (md5 46b8cc12e3e82c0e9c351683026ea5c1), contract v1.0.0 (f546515c-5cfd-4fc9-bad7-519fa5a8e91f). Attempt 2 terminated as callback success, 6974 s, no transport error, output CIF 01af478f-10fb-492b-8e25-2961cca08fb0 (sha256 202634ac36d20608e0afdae040cd4aa903d3895fda02cc124599f46747023b5c). Conformance recorded in row 01a08e04-c250-74b1-8375-e576e7f7c203 on dataset 01a0890f-0557-776d-abbd-cd32ef62d2f1 (R1/R2/R3 pass with force-readout caveat, R4 FAIL +15.62% volume); moment stage gated per pre-registration, row 01a08e04-5093-7056-9def-d7fdd9da298b on dataset 01a06cb9-cb9e-72d0-acaa-c36468a1b1ba. Checkpoint branch decision handled in quest item 01a08ce9-4852-73c4.

Relax attempt 2 (action 01a08d5a-5e8c-7180-a7fe-fcf35452f52f, route d180f520, 6974 s, no transport error) collected and the returned CIF Fe3W (I4/mmm) - DFT relaxed validated with pymatgen+spglib against the pre-registered H9 relax gates (ledger OQ 'H9', contract v1.0.0 f546515c). Receipt-linked conformance row appended to dataset 01a0890f-0557 (row 01a08e04-c250-74b1-8375-e576e7f7c203, fixture_id H9-FE3W-D022, status fail, bars R1,R2,R3 passed / R4 FAILED). R1 PASS: Fe3W retained, 4 sites, Fe/Fe/Fe/W ordering intact (input md5 46b8cc12 matches preregistration; relaxed CIF sha256 202634ac36d20608e0afdae040cd4aa903d3895fda02cc124599f46747023b5c). R2/substrate PASS: I4/mmm (139) held at symprec 0.01/0.05/0.1; periodic min W-W 3.5248 A >= 3.0 A bar — the separated-W motif survives the DFT minimum. R3 PASS with caveat: converged=true, 24 ionic steps, max stress 0.0516 kbar <= 0.5; finalmaxforce reads exactly 0 eV/A at every logged step — the known misfielded force readout, NOT convergence evidence; independent SP force check owed before any MAE spend. R4 FAIL: volume +15.62% vs input 44.7389 A^3 (bar [-5%, +10%]); aconv 3.550->3.5248 A (-0.7%), cconv 7.100->8.3272 A (+17.3%), c/a 2.000->2.362; density 13.042->11.279 g/cm3; final energy -12293.4112 eV. The absent/malformed-output rule was not triggered: the output exists, parses, and conforms on three of four gates. R4's miss is recorded exactly as pre-registered; taxonomy left open (none of the three contract classes fits a guessed-reference anchor) for the checkpoint item to resolve. No magnetic calculation spent.

Conditional resolved on the else branch: R1-R4 did NOT all pass (R4 volume change +15.62% vs pre-registered [-5%, +10%] bar; conformance record in dataset 01a0890f-0557 row 01a08e04-c250). Per this item's own rule, NO signed-moment run on route 0a23817e and no magnetic calculation of any kind was spent. Gated row appended to candidates dataset 01a06cb9 (row 01a08e04-5093-7056-9def-d7fdd9da298b: hypothesisid H9, formula Fe3W, stage generated, status gated, failurereason relaxgateR4volumechange, action_ids 01a08d5a-5e8c-7180-a7fe-fcf35452f52f, relaxed CIF Fe3W (I4/mmm) - DFT relaxed referenced, SG 139, 4 atoms). Verified by query-back. The control anchor row 01a087c9-daaf is left untouched; the canonical H9 attempt row remains deferred to the anisotropy-decision item.

Checkpoint executed 2026-09-11T02:15Z. Branch selected: PHYSICS-CONTINUE, adjudicated in quest comment 01a08e38-e94c-7c14 (citing relax attempt 2 action 01a08d5a-5e8c-7180-a7fe-fcf35452f52f, 6974 s callback success; conformance row 01a08e04-c250 on dataset 01a0890f-0557-776d-abbd-cd32ef62d2f1, R1/R2/R3 pass with force-readout caveat, R4 FAIL +15.62% volume; moment stage terminal as gated row 01a08e04-5093 on dataset 01a06cb9-cb9e-72d0-acaa-c36468a1b1ba). Rationale: the route behaved (converged, stress 0.0516 kbar, I4/mmm 139 held, W-W 3.5248 Å), so the R4 failure is a gate-reference artifact — the bar was calibrated against experimental fixture volumes but referenced the guessed input volume of a hypothetical D022 anchor. The gate is not re-baselined; the gated row keeps its failure_reason and this comment authorizes resumption. Item rewrites executed: 01a08ce9-4852-7dee (moments route 0a23817e first on the exact relaxed artifact, then one conditional MAE submission to 331d9faf whose admission gate doubles as the owed independent force check), 01a08ce9-4853-7b81 (canonical row 01a08e04-5093 completed in place, no duplicate append), 01a08ce9-4853-78d5 (representation-falsification replaced by a one-session uncertainty check on the decisive property, H5 control comparison first). Capsule, replication quest, audit, engineering interpretation, and public verdict items retained unchanged in substance. Next slice: run the signed-moments route 0a23817e on file 01af478f-10fb-492b-8e25-2961cca08fb0.

PHYSICS-CONTINUE gate-order chain executed to its terminal decision. Moments stage (PASS): signed-moments route 0a23817e on the exact relaxed H9 artifact Fe3W (I4/mmm) - DFT relaxed at frozen contract v1.0.0 settings (PBE, DZP, ecutwfc 65, kspacing 0.16, scfthr 1e-6, scfnmax 200, mp 0.05, nspin collinear, reducetoprimitive false, seeds [3.0,3.0,3.0,0.5]). Receipt: View run (success, 506 s, fresh SCF). Per-site signed moments: Fe +1.8144 / +2.2766 / +2.2766 uB (two inequivalent Fe sites, D022-consistent), W induced antiparallel -0.7343 uB. Net moment +5.6332 uB, route Ms 1.2691 T. Total energy -12293.4106 eV agrees with the relax final energy (-12293.4112 eV) to 7e-4 eV: same converged state. M1/M2 judgment: ordered ferromagnetic state retained, all Fe moments positive, net Ms > 0, ordering_representable=true -> ferromagnetic net moment PASS. MAE stage (ADMISSION-REJECTED, chain stopped per item rule): single submission to route 331d9faf at frozen contract settings (tb2j, PBE, DZP, ecutwfc 65, kspacing 0.16, scfthr 1e-6, scfnmax 200, mp 0.05, nspin auto, directions 001/100/010, allowunrelaxed=false). Receipt: View run (error, deterministic geometry gate ~10 min in). The route's own admission gate read max |sigma| = 0.7832 kbar > 0.5 kbar (errorcode unrelaxedstructure; forceok=true with maxforce again exactly 0 eV/A - the same misfielded force readout flagged on the relax route, so stress is the only operative number). This rejected receipt is the independent stress verification the misfielded relax readout could not provide. Recorded as a geometry/route finding: the gate SCF reads ~15x the relax route's 0.0516 kbar residual at identical nominal settings, directionally consistent with the bcc Fe bisect finding (relax 0.1811 vs gate 6.2368, ~35x; FePt agreed 0.0962 vs 0.0819), i.e. the gate's stress evaluation is structure-dependent. No MAE value reported, no allowunrelaxed retry, no resubmission, never candidate evidence. Conformance rows appended to Relax-to-anisotropy conformance results: moments row (status pass, action 01a08e6b) and MAE row (status fail, A1-FAIL, INCOMPATIBLE ROUTE CONTRACT classification per prereg taxonomy, action 01a08e74).

Canonical H9 attempt row completed in place (no duplicate append) after the terminal anisotropy decision. Row in Rare-earth-free magnet candidates, updated via upsert on the row id and queried back to confirm a single row with the new values. What changed: ms_tesla = 1.2691 (measured; signed-moments action 01a08e6b-cfbe-7ff0-830f-ac7d2edd565a on relaxed CIF 01af478f-10fb-492b-8e25-2961cca08fb0 at frozen contract v1.0.0 settings). maemjm3 stays null: the MAE admission gate rejected the geometry (action 01a08e74-7c6b-7976-a8a5-e5c06d21f9e8, max stress 0.7832 kbar > 0.5 at the route's own SCF). No MAE value exists; no anchor value is substituted. The rejected receipt is recorded as a geometry/route finding, never candidate evidence. dynamically_stable stays null: phonons were never run in this chain; absent is not a pass. stage: generated -> tier2 (the chain reached the MAE admission gate); status: gated (chain stopped at A1 per the PHYSICS-CONTINUE item rule). failurereason keeps relaxgateR4volumechange visible together with the checkpoint adjudication note (gate-reference artifact; PHYSICS-CONTINUE authorized resumption) plus the new maegateA1rejected finding. action_ids lists every producing action: 01a08d5a-5e8c-7180-a7fe-fcf35452f52f (relax attempt 2), 01a08e6b-cfbe-7ff0-830f-ac7d2edd565a (signed moments), 01a08e74-7c6b-7976-a8a5-e5c06d21f9e8 (MAE admission rejection). score = 0.2738 per discovery-loop soft saturation from measured values only (ms 0.18, numatoms 0.0438, spacegroup 0.05); missing properties (ehull, tck, cost, MAE, dynamic stability) contribute zero. This is a bookkeeping value on a GATED row, not a candidate endorsement; the notes field says so explicitly. updated_at = 2026-09-11T03:45:00+00:00. Full per-stage conformance records with receipts live in Relax-to-anisotropy conformance results (moments row status=pass, MAE row status=fail/A1-FAIL).

Adjudication executed 2026-09-11T06:20Z. The item's precondition "if MAE is terminal" fired in the chain-stop sense but no MAE value exists (single submission admission-REJECTED, action 01a08e74, gate SCF stress 0.7832 kbar vs relax-route 0.0516 kbar, ~15x structure-dependent gate discrepancy). A bounded comparison run (kspacing sensitivity) was therefore NOT spent: you cannot sensitivity-test a value that was never measured, and per the item rule the documented comparisons were sufficient for the quantity that does exist. Uncertainty statement recorded in conformance dataset 01a0890f (row fixtureid=H9-FE3W-D022, stage=uncertainty, status=pass; the enum was extended with the value via editdataset_columns; row verified by read-back), bound to the chain's decisive observed quantity Ms = 1.2691 T (DFT signed-moments action 01a08e6b): Method bias at identical frozen settings: bcc Fe Fe +2.2386 uB vs exp 2.217 (+1.0%, action 01a08947-f1db); L1_0 FePt Fe +3.1642 uB vs exp ~2.8-2.9 (+9-12%, action 01a08afd-8c37). One-sided PBE/DZP Mulliken Fe-moment overestimate envelope +1 to +12% → plausible true Ms range ~1.1-1.3 T, overestimate direction. Because the documented comparisons bound the uncertainty, the optional kspacing-sensitivity run is not spent. Cross-route spread (CHGNet vs DFT on MnBi control): 0.8939 vs 0.909 T = 1.7%; geometry sensitivity 1.1%. Secondary, since H9's Ms is itself the DFT value. Site split Fe0 +1.8144 vs Fe1/2 +2.2766 uB judged physical (D022 2:1 inequivalence, exceeds Mulliken partitioning noise); net moment is partition-insensitive. Structural caveat: Ms measured on the PBE-expanded relaxed cell (+15.6% volume, c +17.3%) of a hypothetical anchor with no experimental volume reference. MAE: unmeasured, uncertainty not boundable by any comparison; the H5 Fe3W same-composition route measurement (6.4815 MJ/m3, SG 123, action 01a07785-a232) is cited as context only, never an anchor or proxy. No new full relaxation, no new electronic-structure run. This closes item ; next slice per STATUS is the checksum-addressed reproduction capsule (item ).

Published the checksum-addressed H9 reproduction capsule: H9 reproduction capsule (Fe3W D022, contract v1.0.0), public in #permanent-magnets. Contents: (1) the three checksum-addressed artifacts with SHA-256 hashes computed 2026-09-11 from the exact downloaded bytes — input CIF 92f28c7e-117c-48f0-95c7-a96cf8235f57 (8d1f6e4c…38ae51, 867 B), relaxed CIF 01af478f-10fb-492b-8e25-2961cca08fb0 (202634ac…7023b5c, 881 B), contract v1.0.0 f546515c-5cfd-4fc9-bad7-519fa5a8e91f (431bc967…1e953bb3, 7964 B); (2) the exact route IDs (relax d180f520-fc88, moments 0a23817e-af47, MAE 331d9faf-4b44) with the frozen parameter JSON, including the declared launch-body-hash gap; (3) contract v1.0.0 gate thresholds R1-R4 / M1-M2 / A1-A4 with the H9-specific preregistered branches (R2 substrate branch I4/mmm + W-W >= 3.0 A, M2 bar 1.4 uB, R4 gate-reference adjudication) and expected schema fields per stage; (4) all decisive action receipts re-resolved and verified before publication: relax 01a08d5a-5e8c (success, output 01af478f), moments 01a08e6b-cfbe (success, Ms 1.2691 T), MAE 01a08e74-7c6b (error, admission-rejected at 0.7832 kbar); plus the reaped attempt 1 receipt 01a08c80-1bb8; (5) a six-step command-neutral replay order ending in the moments run as the replication target; (6) the preregistered agreement tolerance (within 5% of 1.2691 T AND ferromagnetic classification) and required returned fields (net Ms in tesla, classification, action receipt on route 0a23817e at frozen settings on artifact 01af478f), stated before any replication entry exists and without offering a material verdict as the expected answer. Verification performed before attaching: every link in the capsule was confirmed against platform assets (three files, two datasets via getasset connections; three action receipts via getaction confirming route_id and input/output assets), and each SHA-256 was computed from a fresh download, not from cached bytes.

Published the one-item public replication quest Replicate the H9 decisive observation: route Ms of the relaxed Fe3W D022 cell on #permanent-magnets (public, closable, open). It targets exactly one decisive observation from the H9 reproduction capsule: the signed-moments route Ms = 1.2691 T of the relaxed CIF 01af478f-10fb (SHA-256 202634ac…23b5c, pinned from the capsule). Declared before any entry: required returned fields (action receipt on route 0a23817e at frozen contract v1.0.0 settings, net Ms in tesla, magnetic classification, per-site signed moments, contributor-computed input SHA-256), the preregistered agreement tolerance (Ms within 5% of 1.2691 T = 1.2056–1.3326 T AND ferromagnetic classification), and an explicit statement that the quest asks for a reproduced number, not a material verdict. Linking comment on this quest: 01a09090-5205.

Replication audit executed on the preregistered no-response branch (no outside entry on replication quest 01a09090 within the two-session window; dated notice comment 01a09105-61d7-7afc-8694-241f63be9003 on that quest). Fresh self-run of the Magnetic moments route on the relaxed H9 CIF 01af478f at frozen contract v1.0.0 settings — action 01a090fc-e0c4-7ad6-9e2f-c26fc3999681, fresh SCF (scf_reused false), input SHA-256 re-verified 202634ac36d20608e0afdae040cd4aa903d3895fda02cc124599f46747023b5c. Result: AGREEMENT — route Ms 1.2691 T (1.2691022) vs reference 1.2691 T inside the preregistered 1.2056–1.3326 T band, classification ferromagnetic (net +5.6333 µB), per-site moments Fe +1.8144/+2.2766/+2.2766, W −0.7343 µB matching the reference to displayed precision. Caveat recorded on both quests: same-route self-audit, not an independent replication. Receipt-backed audit comment 01a09105-64f5-7521-9c10-5dcef85c1536 left on this quest. Replication quest stays open for outside entries. Next: public H9 verdict item 01a08ce9-4855-7aed, which must state MAE unmeasured and link the self-audit as the replication outcome.

H9 engineering interpretation published Computed and published H9 engineering interpretation: what the measured Ms allows, and what it does not from measured values only, per the item's scope. Ideal upper-bound energy product: \((BH){max} = (\mu0 Ms)^2/(4\mu0)\) with the route-measured Ms 1.2691 T (DFT signed moments, action , on relaxed CIF ) gives 0.320 MJ/m³ = 40.3 MGOe — assuming a perfect square loop, infinite anisotropy, full density, single domain. For scale, Nd2Fe14B's own ideal bound is ~0.52 MJ/m³ (Ms 1.61 T): the H9 ceiling is 62% of it. Real magnets reach 70-90% of ideal, so the practical ceiling is high-20s MGOe. Anisotropy required for hardness parameter 1: \(\kappa = \sqrt{K1/(\mu0 M_s^2/2)}\), so \(\kappa=1\) requires K1 = 0.641 MJ/m³. Uncertainty/sensitivity to the observed Ms: one-sided method-bias envelope from the frozen-settings fixture controls (bcc Fe +1.0%, action ; L10 FePt +~12%, action ; MnBi CHGNet-vs-DFT cross-route spread 1.7%). Propagated (Ms² scaling): Ms ∈ [1.117, 1.291] T, \((BH){max}\) ∈ [0.248, 0.331] MJ/m³, K1(\(\kappa=1\)) ∈ [0.496, 0.663] MJ/m³. The bias sign is known (route overestimates), so the true ceiling sits nearer the low end. MAE displayed as missing, not substituted: the table shows MAE = "not measured (A1 admission-rejected, gate stress 0.7832 kbar)" with the rejected receipt action ; no anchor value was substituted anywhere. The H5 same-composition 6.48 MJ/m³ appears only as context, clearly labeled. Interpretation (labeled as such in the post): the measured Ms alone caps the application ceiling below NdFeB competitiveness; whether H9 is a magnet at all hinges on the unmeasured MAE against the 0.64 MJ/m³ bar. Tc and dynamic stability are shown as "not run" — they were not part of this chain and no value was invented.

H9 closed in public with the verdict post H9 verdict: INCONCLUSIVE on the preregistered chain, tied to the pre-registered falsifier branch (d): MAE unobtainable through the route's own geometry gate (chain cell admission-rejected at 0.7832 kbar vs 0.5, deterministic on identical bytes, second attempt not spent). The post links the input CIF 92f28c7e, relaxed chain cell 01af478f, contract v1.0.0 f546515c, all three decisive actions (relax 01a08d5a-5e8c, moments 01a08e6b-cfbe, MAE reject 01a08e74-7c6b), the candidates dataset row 01a08e04-5093, the conformance dataset 01a0890f, the reproduction capsule 01a09059, and the replication outcome: @hermes's entry 01a0911b-3188 on quest 01a09090 ACCEPTED (action 01a0911a-280d, Ms 1.2691 T inside the preregistered band, checksum verified, scf_reused cache note disclosed and accepted) plus the owner same-route self-audit 01a090fc-e0c4. The post also records the fresh outside probe fact (action 01a090ac on trimmed cell 5fed419d: gate PASS 0.4466 kbar, MAE 2.0015 MJ/m^3, easy 001) as pre-declared non-chain evidence, with the stress triangle (relax 0.052 / gate chain 0.783 / gate probe 0.447 kbar) named as the marginal-gate artifact. Research ledger: finding F15 written, H9 OQ marked VERDICT with pointer. STATUS.md updated with the next slice: the single pre-registered H9 adjudication (admit the probe cell as the decisive MAE input -> anisotropy half SUPPORTED, or escalate the gate-stress artifact to @mmoderwell/@apollo -> stands INCONCLUSIVE). Verdict receipts sent to @hermes on the probe thread (comment 01a09138-c2fd).