Answers are grounded in this paper and cite their sources.
Computational verification study demonstrates complete benchmark reproducibility across forty-seven fusion reactor solvers, highlighting rigorous mathematical provenance for nuclear plant design.
A design frozen for construction is only as credible as the solvers behind it, and a solver is only as credible as its provenance. We present the verification backbone of the Kronos physics de-risking campaign, which de-risks a compact spherical-tokamak breeder (design point Q=3.076, P_ fus=85.04MW, I_p=9.66MA, negative triangularity δ=-0.30, B_0=8T, elongation κ=2.0, major radius =1.20m, aspect ratio A=2.5, centrepost peak field 16.84T) and a D–^3He tandem-mirror burner (Q_E=1.318, neutron fraction f_n=5.44%, end-plug field 26.49T, direct-conversion efficiency η_ DEC=0.49). The backbone is a catalog of 47 named solvers across nine physics domains, each assigned a formal role (reference, surrogate, or gate), a pinned version (git commit, build environment, and data library), and a reference benchmark it must reproduce within a stated tolerance. We give the mathematics that makes the backbone auditable: a decomposition of total predictive error into surrogate, discretization, and model-form terms; Richardson extrapolation and the grid-convergence index for solution verification; the method-of-manufactured-solutions order test for code verification; the Monte-Carlo standard-error law for statistical convergence; the Sobol variance decomposition for sensitivity; a Minkowski-L^1 area validation metric; a hash-chained provenance ledger with a tamper-evidence guarantee; and a distribution-free acceptance layer that upgrades the surrogate-promotion rule with a finite-sample conformal coverage guarantee P(y C(x))1-α and an out-of-distribution epistemic gate, deployed behind a formally verified Simplex runtime-assurance envelope. We then show the backbone at work: the formal verification-and-validation suite passes 19 of 19 frozen anchors across breeder and burner; the CGYRO linear gyrokinetic reference returns a growth rate =+0.3179 (in a/c_s units) byte-identically to its shipped benchmark; a CAD-faithful DAGMC neutronics chain reproduces the net tritium-breeding ratio to 1.014±0.004 against the 1.026 of the frozen recipe; and two independent density-functional-theory codes agree on a divertor-alloy formation energy to 0.002eV/atom (+0.0147 versus +0.016). Every headline number in the campaign thus carries a traceable computational provenance and an independent check behind it. All results were produced on the NVIDIA GPU HPC campaign using the cataloged community reference codes. Key results (frozen anchors): fus = 85.04 MW; I_p = 9.66 MA; B_0 = 8 T; Rz = 1.20 m; f_n = 5.44 %; N = 1.2 ×10^8. Methods & codes: FreeGSNKE, FreeGS, NEO, CGYRO, TGYRO, TGLF, ASCOT5, OpenMC, NIMROD, GENE, SOLPS-ITER, SOLPS, EIRENE, DAGMC, ENDF/B-VIII.0, STEP, R2S. Live verification: 13 gate validator(s) with live-recompute cards (11 reproduced, 2 revised). See the Verification section and data/verification.csv. Related: Paper page · De-risking register · 3D model · Learn more about Kronos Part of the 2026 Kronos publication series; independently re-run and stamped in the Kronos de-risking register (DOI 10.5281/zenodo.22645689). All numerical values are frozen design-point anchors; see the register. Public research artifact. No proprietary, financial, or supply-chain information is included.
No takes yet. Share an insight, caveat, or question.
Priyanca Ford (2026) studied this question.