Canonical semantic model

FullMag uses one semantic flow:

Python or browser authoring → ProblemIR → validation and normalization → capability planning → runtime execution → fields, observables and artifacts.

The public contract preserves requested intent and resolved execution. Automatic fallback, when allowed, must be visible in provenance. An explicitly requested GPU path must fail clearly when its requirements are unavailable.

Concept	Public meaning
ProblemIR	canonical lowered meaning of the physical problem
Study	declared execution and analysis intent
Session	user-visible execution context
Run	one execution attempt
Stage	one physical or numerical phase
Field	spatial quantity such as magnetization or effective field
Observable	scalar, vector or derived quantity exposed for inspection
Artifact	reproducible output with provenance

Detailed API pages are published only after the corresponding contract and validation scope are stable enough to publish.