Models

Models are filed under the calculation setting they advise, not the algorithm that produced them.

Setting	Predicts	Model	Status
k-point mesh	k_distance	QRF95	published, historical
k-point mesh	k_index	k-index forest	published
k-point mesh	k_index_dense	Dense-mesh screen	draft, serves the dataset
Metallicity	is_metal	CGCNN	published
Metallicity	a representation	CGCNN	published, historical
Magnetism	is_magnetic	mMACE+MLP	published
Hubbard U	—	—	planned
Pseudopotentials	—	—	planned
Exchange-correlation	—	—	planned
Smearing	—	—	planned

Historical means the record's latest version is its last. Both were fitted before this repository existed, so neither carries a training run anyone can repeat. They stay loadable and citable.

How a model is named

k_points . k_distance . qrf . goldilocks_kdist_ultra . v1
└ setting   └ quantity   └ family └ dataset            └ version

Part	Answers
setting	which calculation input this advises
quantity	what the number is, which decides how it is converted
family	the kind of model fitted
dataset	the snapshot's own record id
version	bumped when the same combination is trained again

Setting and quantity are separate because one setting can be reached from more than one quantity — a k-point mesh from either a k-distance or a k-index — and those need different conversions.

Every deposit definition lives in deposits/, and is the example to copy when you publish your own.