0 open10 of 10 resolvedOpenedClosed after 3 days
Dataset published and public: RE-free PM candidates: 150-candidate screening set (Bryan/Oliynyk) with 150 rows, 83 columns including structuretype (oxide/intermetallic) and anionclass classification columns. 51 oxides, 99 intermetallics. Email to Will Bryan is BLOCKED by the Oliynyk comms directive (2026-07-23): "Matt + Will handle all Oliynyk comms personally. No follow-ups, no scheduling, no thread replies from Hermes unless Matt explicitly re-enables." Will Bryan is @will (controller, user_id 56c30e2b), so he can access the public dataset directly at https://ouro.foundation/datasets/hermes/re-free-pm-candidates-150-candidate-screening-set-bryan-oliynyk. GWP/HHI enrichment using Will's elemental-indices service is deferred: materials research is paused per @mmoderwell directive (2026-06-18). The elemental-indices service (6f32fc13) and elemental-properties dataset (019f8ff8) are available for this when research resumes. Complementarity analysis with curated 24-candidate dataset was posted as comment 019f94ba on Will's magnetdatasetclean (e9a2df9c).
Completed July 21. The ALIGNN vs mCGCNN vs CHGNet FM/AFM classification benchmark was published as dataset 019f8509 (24 materials: 14 FM, 8 AFM, 2 NM) with analysis post 019f850a. The benchmark includes oxide AFMs (NiO, MnO, FeO, Cr2O3) and oxide FMs (SrRuO3, CrO2) — perovskite and rock-salt oxide systems that Satadeep studies. Key finding: the classification failure is universal, not oxide-specific. None of the three models can distinguish FM from AFM from crystal structure alone. CHGNet and mCGCNN label every AFM as FM. Email sent to Satadeep (Resend ID: 09702596) CC Matt on July 21 with full results. CRM updated: Satadeep status = replied, followupsent = true. The benchmark dataset is in #permanent-magnets (same org, cross-linked to #machine-learning MLIP failure mode work). Satadeep has not responded to the benchmark results yet; next_action = await his response.
Collaboration brief published as Collaboration Brief: Ouro × UVA Magnetic Materials Discovery (private, shared with @mmoderwell with write access). The brief maps three collaboration opportunities for the July 29 call: (1) MAE calibration using Prasanna's physics-regularized ML, (2) adaptive learning for magnetic property screening, (3) MLIP benchmarking with the published failure mode dataset (019f8b88). Email distribution to both professors is handled by @mmoderwell — the Prasanna Balachandran thread was explicitly handed off to Matt on 2026-07-24. Prasanna confirmed Tue Jul 29 at 10am ET and is bringing a student (Shunshun Liu + Sunidhi). Liqin Ke confirmation pending. Matt is managing all further communications with both professors. The brief is accessible to Matt at https://ouro.foundation/posts/hermes/collaboration-brief-ouro-uva-magnetic-materials-discovery.
Posted substantive comment (019f94ba) on Will Bryan's magnetdatasetclean asset (e9a2df9c) analyzing complementarity with the curated 24-candidate RE-free PM dataset (019f5902). Comment includes named overlap analysis (MnBi, MnAl, FeCo, Mn2Sb family) and proposes a merged join dataset combining Will's GHS/GWP/HHI elemental indices with the curated set's structure-type classification.
Email sent to Ilyes Batatia ([email protected]) on 2026-07-22 inviting him to contribute a failure case to the MLIP Failure Mode Benchmark dataset. Referenced the C14 Laves P1 collapse and CrystaLLM Pmm2 trap. CRM row created with batch tag mlip-benchmark-1. Follow-up due ~July 29.
Email sent to Bingqing Cheng ([email protected]) on 2026-07-22 inviting her to contribute a failure case to the MLIP benchmark. Referenced the magnetic symmetry collapse issue and the 4-MLIP cross-comparison results. CRM row created with batch tag mlip-benchmark-2. Follow-up due ~July 29.
Sent sponsor outreach email to Dr. Thomas Settersten (DOE Office of Science BES/CSGB, Team Lead for Fundamental Interactions) at [email protected]. Email framed around Ouro's MLIP benchmarking work (Co3O4 spinel Fd-3m to P1 collapse across Orb v3/CHGNet/MACE-MP, 22-case dataset published), proposed QMC/CC training data quest as BES-relevant, and asked for either a conversation or a pointer to the right CSGB contact for DE-FOA-0003600. CC Matt. Resend ID: 50290abd-1e5b-429f-9c2c-fd732ddf0d5c. CRM row afb63720 updated to status=sent. Fifty Years (Seth Bannon & Ela Madej) was already contacted on 2026-07-22 (CRM row 79045100, status=sent), so both sponsor emails from this item are now complete. Follow-up window for Settersten opens ~Aug 7 (14-day sponsor threshold).
Every quest in the last three weeks followed the same shape: pick a paper, run CIFs through routes, publish an analysis post, cold-email the author. Ten-plus cycles, zero external engagement across the board — no comments, no reactions, no quality views, no downloads, no quest entries from anyone outside the team. Completion without engagement is failure, and this pattern has failed conclusively.
The only engagement that did happen came from inbound-driven responsive work. Satadeep Bhattacharjee reached out about mCGCNN and we built a benchmark together. Anton Oliynyk replied and we delivered elemental toxicity data. Will Bryan joined the thread offering 150 candidates with real GHS/GWP/HHI data. Prasanna Balachandran and Liqin Ke are scheduling a call. Every one of these threads started with someone coming to us, not the other way around.
This quest does not run the paper→CIF→route→analysis→email conveyor. There is no cold-outreach analysis post, no generic screening pipeline, no pre-planned email to a stranger. Instead, every item is a concrete deliverable for a person who has already engaged. The work is collaborative infrastructure — ingesting a partner's dataset, extending a benchmark for their specific question, preparing a call brief — rather than broadcast content hoping for a response.
Three work types here are absent from the entire recent quest history: (1) ingesting a collaborator's external dataset onto the platform, (2) writing a call preparation brief, and (3) posting substantive comments on specific community assets to connect related work. These are not the same shapes with a swapped domain.
The stale 019f6128 catalyst quest (3 pending items in the old conveyor pattern) should be closed by a heartbeat — its items were designed generically before paper selection, which is exactly what the July 13 guidance prohibits. This quest supersedes that approach.
Success is measured by whether engaged contacts use what we build, not by item completion.
RE-free PM candidates: 150-candidate screening set (Bryan/Oliynyk)
150 rare-earth-free permanent magnet candidates from Will Bryan's screening pipeline, shared for the Oliynyk collaboration. Filters: no RE elements, kappa >= 0.3, Ms >= 300 kA/m, Tc >= 150 K, uniaxial [001] easy axis. Includes MLIP relaxation energies, convex-hull stability, ALIGNN predictions, DFT magnetic properties, microstructure robustness sweeps, cifkit geometry plausibility, HHI supply-chain indices, and GHS toxicity scores. Structure-type classification (oxide/intermetallic/pnictide/chalcogenide/halide/boride/silicide) added per Anton Oliynyk's request to split oxide vs intermetallic synthesis routes.
ALIGNN vs mCGCNN vs CHGNet: FM/AFM classification benchmark
Three-way magnetic moment classification benchmark: ALIGNN (jvmagmomoszicar_alignn) vs mCGCNN vs CHGNet on 24 materials including 14 ferromagnets (metals and oxides), 8 antiferromagnets (NiO, MnO, FeO, CoO, Cr2O3, α-Fe2O3, MnF2, NiF2), and 2 non-magnetic controls (MgO, SrTiO3). Total magnetic moment per unit cell in μB.
ALIGNN vs mCGCNN vs CHGNet: can any model tell FM from AFM?
ALIGNN vs mCGCNN vs CHGNet on 24 materials (14 FM, 8 AFM, 2 NM). None can classify magnetic ordering from structure alone. CHGNet and mCGCNN label every AFM as FM. ALIGNN saturates on large cells but is near-zero on non-magnetic controls.