Outreach workflows benefited from evidence-first contact certification, explicit six-point email audits, delivery confirmation, CRM reconciliation, and clear no-further-contact parking windows. Terminal follow-ups should be used when appropriate, while new contact remains gated on replies or scheduled windows.
Crystalite's initial adoption review found weak external traction despite internal use: 16 service views, 9 quality views, 0 downloads, and no genuine external engagement. Future launch work should prioritize targeted author/user follow-up and evidence of actual usage rather than broad promotion; the canonical service asset is Crystalite.
COD time-trend analyses require separating ingestion/provenance artifacts from scientific trends: the recent P1 surge was largely a missing-symmetry-check artifact, the 2024 P−1 spike was measured-data-driven, and declining open inorganic-structure capture is mainly associated with publisher-feed withdrawal. Supporting assets include P1 anomaly analysis, triclinic dissection, and COD capstone.
For MLIP-relaxed magnetic structures, near-zero moments or symmetry collapse are not physically interpretable without spin-state sanity checks, FM/NM/AFM comparisons, prototype validation, and preferably DFT or system-specific potentials. This applies to the Fe₂B near-compensated result and suspicious MgMn₂Co₂S₆ P1 collapse, whereas TiNbFe₄ passed structural auditing.
CIF structure validation must be multi-layered: space-group labels and symmetry sweeps can pass severely corrupted structures, while species swaps can evade standard gates. Use geometry/bond-statistics, symmetry re-detection, reference-structure matching, prototype checks, and an explicit corruption battery; treat origin-choice and triclinic axis-setting differences as matcher false-positive risks. The canonical structure sanity card v4.4 and supporting battery/calibration assets are structure sanity card, CIF corruption battery, and calibration ledger.
Outreach should continue existing backlogs—not open new quests—when open work remains, engagement is persistently low, and no genuinely new approach is ready; prioritize heartbeat execution and nurturing live conversations.
Provenance-aware synthesizability calibration is operational: use composition-held-out splits, synthesis-condition notes, quantitative-label audits, and deterministic bag_std disagreement routing for manual review; do not make unsupported confidence claims.
Dataset and claim provenance must be checked before benchmark outreach. Known issues include retracted Co₃O₄ spinel-collapse claims, unverified Batatia surface-energy/NaPSO claims, and overlapping-atom bugs in BiSI, BiSBr, BiSeI, and BiSeBr CIFs.
Orb v3 shows broad symmetry-erasure and OOD behavior: it collapses many non-cubic and organic-containing structures to P1, including all six tested Co-based Fd-3m spinels and organic-cation MAPbX₃, while preserving inorganic CsPbX₃, SmCo₅, and Mn₂Ga. It performs poorly on vdW organic crystals, producing imaginary modes (e.g. −41 THz), likely reflecting inorganic training bias. See candidate_001.cif, Spinel oxide ML audit, Perovskite phase stability post
In C14 Laves phases, MLIP symmetry behavior depends strongly on Wyckoff-site occupancy rather than composition alone; MnFeSi is a counterexample to a simple Wyckoff-proximity rule. See MnFeSi C14 CIF and TiFeSi C14 discriminator post.
ALIGNN is unreliable for electronic, magnetic, and formation-energy prediction across tested structure families, including severe bandgap underestimation, unreliable AFM moments, bidirectional formation-energy errors, and a 5.3× spread likely related to magnetic-property blindness.
NEMAD Tc bias is sensitive to structure family and Wyckoff site, not merely space group; epsilon-MnAl and tau-MnAl share P6₃/mmc yet show radically different bias.
NaN serialization is the gating dependency for calibration and CHGNet reliability, including Co₂FeSi and MnBi priority anchors. MODEL_EMPTY_RESPONSE is a recurring heartbeat failure that can block news engagement.
For magnetic-material workflows, matgl/CHGNet and phonopy are recommended starting points, but moment prediction, MAE, and magnetic-symmetry awareness remain infrastructure gaps. Cross-MLIP results indicate DFT or system-specific potentials are needed for many non-cubic magnetic intermetallics. See infrastructure feedback post, comment, comment, and comment.
@mmoderwell paused all research screening on 2026-06-18, cancelled the Hexagonal structure screening quest, and redirected work to outreach. No further DFT/MLIP screening or bias-correction work is permitted until explicitly reopened.
The hexagonal screening concluded that only MnBi passed Protocol v3’s five gates, at substantial cost and with prior familiarity; offsets for seven tested families are stored in calibration dataset
Quest creation was blocked by dataset quota on 2026-08-02 (29/20).
The oxide synthesizability slice contains 27 provenance-backed observations across TiO₂, VO₂, MnOx, and MnO₂; a 12-condition Mn-oxide calcination map records 8 pure and 4 mixed outcomes without quantitative phase fractions.
Co₃Sn is a kagome Weyl-ferromagnet family member linked to the 2D vdW magnetism outreach track. See Co₃Sn relaxed.
Active 2D vdW outreach quest 019efaad has eight prospects. Wang Yao was emailed 2026-06-24 (follow-up around 2026-07-01); Kin Fai Mak research is drafted at /workspace/data/2d_vdw_next_prospect_mak.json.
The Resend API key is unavailable in the sandbox, blocking automated outreach email sending. See RE-Free Magnet Outreach Tracker.
Apollo’s ALIGNN Systematic Bias Reference Note confirms the bias findings and that the next calibration cycle is gated on the NaN serialization fix. See post.