MAST toroidal-field authority

validation/mast_toroidal_field_authority.py is the toroidal-field admission gate for the canonical disruption channel BT_T. It does not extract, interpolate, or rename a field. It answers the narrower question: does one verified FAIR-MAST source object carry enough physical authority to call a value canonical total toroidal field?

Quantity boundary

The pinned FAIR-MAST Level-2 mapping exposes separate bphi_rmag and bvac_rmag arrays. The direct route therefore requires all three source members on the same one-dimensional equilibrium timebase:

• equilibrium.bphi_rmag, in T, identified as EFM_BPHI_RMAG;
• equilibrium.magnetic_axis_r, in m, identified as EFM_MAGNETIC_AXIS_R; and
• equilibrium.time, in s.

The output quantity is the signed total toroidal magnetic field at the time-varying magnetic-axis major radius. Source sign is preserved. Observed negative or positive samples are evidence about values, not authority for the physical positive direction. bvac_rmag remains comparison evidence only and cannot be relabelled or absolute-valued into BT_T.

The array names and source relationship are pinned to the FAIR-MAST ingestion mapping. Dataset provenance follows the FAIR-MAST SoftwareX paper.

Required authority

Array metadata alone is insufficient. Admission additionally requires a content-digested declaration from an allowlisted UKAEA or FAIR-MAST primary source for:

1. total-versus-vacuum quantity semantics;
2. the physical positive toroidal-field direction; and
3. one-standard-deviation uncertainty, either as a positive constant in T or as a verified, non-negative per-sample array in T.

The alternative TF-current route is also declared, but remains blocked until the exact signed current signal, timebase, field-radius calibration K_TF in T m A⁻¹, polarity, positive reference radius, all standard uncertainties, and input covariance policy are source-attested. The gate does not substitute an ideal-solenoid constant.

Real two-shot result

The 2026-07-23 bounded run acquired a new SourceObjectManifest-v2 snapshot for MAST shots 30421 and 30424 into the external Samsung datasets tree. The snapshot is generation-pinned and contains bphi_rmag, bvac_rmag, magnetic_axis_r, and their equilibrium timebase without resampling.

Both shots have 107 joint finite field/radius/time samples. The total-field and vacuum-field candidates are demonstrably different: their maximum absolute differences are about 0.7774 T and 0.4328 T respectively. This rejects bvac_rmag as an identity substitute on the observed data as well as by contract.

The current result remains blocked: the public array metadata does not supply an authoritative physical positive direction or one-standard-deviation uncertainty, and the gate does not invent them. Canonical binding, training, scientific validation, facility prediction, and control admission remain false. The digest-bound report is retained as non-public operational evidence; the reproduction command below emits a reviewable report at a caller-selected external path.

Reproduction

Run the assessment only against a fully verified SourceObjectManifest-v2 root:

python -m validation.mast_toroidal_field_authority \
  --manifest /external/source-v3/source_object_manifest.json \
  --artifact-root /external/source-v3 \
  --shot-id 30421 --shot-id 30424 \
  --json-out /external/l2f12a-authority.json

The output is deterministic and self-digested. A source-member, metadata, shape, time-order, reference-radius, finite-coverage, sign-authority, or uncertainty failure always keeps canonical_bt_binding_admissible false.