Lupine's external alignment, under a deliberate visibility convention:
\nThis page is the public layer.
\nRevealed preference in the field is that funding flows to data-driven, closed-loop\npotential validation and retraining, not to incrementally better DFT. Lupine's bet —\nthe audit layer for the MLIP ecosystem, compounding into a low-rank retraining target —\nis paradigm-aligned with where the money and the open problems actually are.
\nThe MIIT-67 pilot programs were mapped against the Lupine thesis. Tier-1 direct hits —\nproblem classes where the hyper-ribbon error-geometry method applies with little\nadaptation:
\n| Pilot wedge | \nWhy it maps | \n
|---|---|
| #2 — Cu alloys | \nElastic-constant error geometry is exactly the corpus's strongest regime. | \n
| #18 — Mg stiffness | \nStiffness prediction is a direct manifold-projection use case. | \n
| #15 — Powder reuse | \nProperty-transfer validation across reuse cycles is a closed-loop fit. | \n
The corpus is metals-elastic-heavy today; the methodology transfers beyond that, and\nthe Phase-D compute lane extends property coverage (Ecoh, B0) to widen the\naddressable set.
\nTwo pilots have been instrumented and run through the live traced loop:
\ntesting confidence ~0.3 — an honest weak, non-confirming\nsignal, reported as such.Next: Phase-D elastic-constant runs on Cu/Mg to move these from weak signal to grounded\nresult.
\nThe open corpus, the methodology, and the\nconjecture ledger are the front door. Engagement specifics are\nhandled off-page.
\n"}