Replication Data for: Causal Intervention Models for Regulator-Facing Plasma Control: Do-Calculus Prediction and Per-Decision Explainability in a Compact Spherical-Tokamak Breeder
Year: 2026
Type: dataset
Abstract: Replication data for the paper Causal Intervention Models for Regulator-Facing Plasma Control: Do-Calculus Prediction and Per-Decision Explainability in a Compact Spherical-Tokamak Breeder (Kronos Fusion Energy 2026 Publication Series, paper 2.82), published at doi:10.5281/zenodo.22645825. Abstract of the paper. A controller that holds a nuclear plasma must answer, on every cycle, two questions a regulator will ask: what will this action do, and why did you take it. Correlational machine learning answers neither; both are causal questions, and a control action is not an observation but an intervention. We formalise the Kronos digital twin of the Hyperion breeder (design point I_p=9.66, Q=3.076, P_ fus=85.04, negative triangularity δ=-0.30, B_0=8, elongation κ=2.0, major radius =1.20, aspect ratio A=2.5, centrepost peak field 16.84) as a differentiable structural causal model (SCM): its nodes are physical states, its edges are the physics couplings, and its mechanisms are physics operators, so a control action is an executable do-calculus intervention do(u) whose interventional distribution the twin computes directly, and every decision carries a gradient-based attribution to the states that drove it. We write the SCM, the truncated-factorisation identity for P(y do(u)), the control-barrier quadratic program that certifies the intervention, and the integrated-gradient and Sobol attributions, as explicit equations, and we discretise and solve each on the NVIDIA GPU high-performance-computing (HPC) campaign. Three subsystems are reduced to practice against pre-registered acceptance tests. What this dataset contains data/key_values.csv and data/anchors.json: the frozen design-point values quoted in the paper (symbol, value, unit, provenance), in machine-readable form. data/verification.csv: the validator gates the paper cites, each with its verdict and a link to a live recompute page. figures/: the figures from the paper, as vector PDF, with an inventory. causal-intervention-models-for-regulator-facing-plasma-.data-methods.pdf: methods, the computational codes used, and the full value table. data/data_dictionary.md: column definitions. Every value is a frozen design-point number that is independently re-run and recorded in the Kronos Physics De-Risking Register (doi:10.5281/zenodo.22645689). The paper itself is not duplicated here; read it at the DOI above.
Source: Harvard Dataverse
DOI: 10.7910/dvn/qlc0gc
Author: Ford, P I
Language: English
Subfield: Nuclear and High Energy Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7214606629