Curated claim facets

Lupine research claims

The 146 claim/S* nodes in the Lupine research ontology. Internal scaffolding (relation:* predicates, superclass:* definitions) stays in the wiki DB. Domain claims are surfaced here in 8 named facets, ordered so the story reads from the failure modes to the corrective framing to the published artifacts.

Systematic biases 10 claims The ten failure modes the correction layer defends against. Each is a recurring pathology in machine-learned interatomic potentials or the workflows that depend on them — the reasons universal MLIPs mispredict and why we measure and correct at runtime.
Historical markers 7 claims Published structural-failure cases from 2020–2023: A-Lab's 41 "novel" compounds, GNoME's 2.2M crystals, LK-99, CSH/Lu-hydride room-Tc claims, the "AI-assisted 44%" study, DFT-guided N₂-reduction literature, and ML water-stability classifiers. The empirical record on which the systematic-bias catalogue rests.
Frame & ladders 9 claims The wedges, levels, and correction framing. F1–F3 are the inherited floor, the level playing field, and the validation-first wedge. K1–K3 name the specific wedges (90:10 fidelity, barrier, fusion long game). L1–L3 name the ladders (non-equilibrium pretraining, multi-fidelity Δ-learning, systematic-error surgery).
Risks 7 claims Risks the program acknowledges, with the survival condition that closes each one. R1 is the fidelity ceiling; R2 the many-species data wall; R3 solo-execution credibility; R4 export-control shocks; R5 compute-ladder gating at top; R6 instrument disappearance; R7 foundation-model commoditization closing the wedge early.
Climate targets 9 claims The five priority material-class bets and the cold-state numbers that frame them. CT1–CT5 name the targets (cobalt-free cathodes, halide solid electrolytes, MOF sorbents, ammonia catalysts, lead-free perovskites). CZ-1..4 are the preregistered cold-state figures — NEB barrier headline, hydride compute share, China storage projection, tritium balance.
Excluded territory 3 claims Where the correction layer is not appropriate. X1: conventional steels, simple metals, commodity ceramics — empirically converged, no room for the wedge. X2: bio/pharma, small-molecule discovery — different workflow. X3: quantum-computing hype — adjacent industry. The correction layer is a tool, not a religion; it has a perimeter.
Epistemic grammar 17 claims The grammar of confidence: epistemic markers (EMK-OBS/INF/TRN/PRP/FRC), confidence grades (CG-High/CG-Medium), readiness grades (RG-H/M/L), and success gates (SG1–SG7 across simulation accuracy, hit-rate uplift, synthesis, stability, component performance, manufacturability, technoeconomics). How a claim earns its place on the page.
Program milestones 7 claims Published artifacts and program-level milestones. PUB-1 is the shared-DFT-anchor proof pack — published and verified (the public-economics guardrail "$14.65 per 129 anchors" derives from it). FP1 is the formal proof library. LM1 is the Lupine Method. RP1–RP3 are program milestones; CA1 is the climate aggregate that ties them together.