k-point mesh

How finely reciprocal space is sampled.

Core knows every mesh a structure can have, in order, with the k-distance, k-line-density and k-points-per-atom each one corresponds to. A model does not predict a mesh; it predicts one quantity on that ladder and Core finds the rung.

Quantity	What it is	Model
k_distance	Largest spacing between adjacent k-points, Å⁻¹	QRF95, historical
k_index	Position in the ordered table of meshes	none yet
k_line_density	k-points per unit reciprocal length	none
k_pra	k-points per reciprocal atom	none

QRF95 was fitted for a published study rather than in this repository, so it carries a citation — dataset, reference calculations and training code — in place of results measured here.

Core's side

Core owns the ladder and the conversion onto it, and nothing else. Feature extraction and inference belong here. Core's current k-index path does both itself, which is what the inference seam replaces.

Two datasets can both call a column "k-distance" and differ by a factor of 2π. Nothing in the number says which. This is why a model declares a target contract rather than a column name.