{"id":"data-provenance","title":"Data & Provenance","subtitle":"Where every number comes from: OpenKIM, NIST IPR, 559 potentials, the integrity gates.","category":"foundations","tags":["provenance","data","trust"],"source":"articles/docs/data-provenance.md","lang":"en","words":338,"readMinutes":2,"toc":[{"depth":2,"text":"Sources","id":"sources"},{"depth":2,"text":"Scope of the corpus","id":"scope-of-the-corpus"},{"depth":2,"text":"Integrity controls","id":"integrity-controls"},{"depth":2,"text":"What we do not claim","id":"what-we-do-not-claim"},{"depth":2,"text":"Reproducibility","id":"reproducibility"}],"html":"

Data & Provenance

Where every number in the corpus comes from. The manifold is only as trustworthy as its\ninputs, so this page is deliberately explicit.

\n

Sources

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
SourceRole
OpenKIMPrimary prediction source — elastic constants from published interatomic potentials, queried via the OpenKIM API.
NIST Interatomic Potentials Repository (IPR)Cross-reference and ground-truth anchor for potential identity and reference properties.
MLIP provenance recordsLattice constants (a0a_0a0​) recovered from the originating MLIP runs — the basis for the de-myopization that proved the ribbon is not a CijC_{ij}Cij​ artifact.
Phase-D compute laneReal LAMMPS runs (not cache) that produce EcohE_\\text{coh}Ecoh​/B0B_0B0​ and surface recipe bugs the cached pipeline masked.
Kimi 2026-06-07 MLIP evidence importCloud Run MACE/CHGNet/SevenNet elastic constants, Vandermonde checks, real early-exit timing, and MLIP/MD refusal evidence curated under data/mlip_benchmarks/kimi_2026_06_07/.

Scope of the corpus

\n

Integrity controls

These exist because we got burned — see\nthe BCC/FCC self-correction:

\n\n

What we do not claim

The corpus is metals-elastic-heavy. Statements outside that regime are methodological\n("the method transfers") not empirical ("we measured it") until Phase-D widens the\nproperty basis. The ledger marks every such boundary explicitly.

\n

Reproducibility

Every figure is regenerable from these sources via the open atlas-distill engine —\nsee Reproduce Our Results.

\n"}