Frontend models have some peculiarities compared to some other popular web frameworks, so much so that they probably shouldn't be called "models" in the first place:
they are composed of two separate parts: internal data and metadata
they do no direct data access, instead serving more as simple data structures with some additional convenience methods
The reason for 1) is to make explicit the separation between data that is "owned" by the model item itself, and that which is relational and must be updated elsewhere. This has a practical component, in that whilst the model's internal data and metadata must be read from the backend service, only the internal data has to be written to it, which greatly simplifies serialization. Additionally, internal data often has to be deserialized from other sources, namely from web forms (and in some cases, client side JSON), and writing form deserializers is much simpler when all the item metadata can be ignored.
The way this internal/metadata duality manifests itself is best
described from example: the model for the documentary unit type is
DocumentaryUnit, which is a
model field containing their internal data, which for the
documentary unit is
standing for "form" (which is unsatisfactory and subject to change!).
DocumentaryUnit also has various
metadata fields, such as it's
holder (a repository) and its
parent (another documentary unit, if it exists), which are not
directly updatable when a particular documentary unit is modified.
The principal reason for 2) is that data access is assumed to be asynchronous (and slow) and taken care of by a separate service layer. This has the side effect of enforcing what is typically deemed a best-practice by separating logic (e.g. data access) and presentation.