**We want to make NWB a seed format in an interoperable, peer-to-peer
graph of research data**
NWB is written with its own [{index}`schema language`](https://schema-language.readthedocs.io/en/latest/)
(And see [the next section](nwb) for more information). It seems to have been created
primarily because other schema languages at the time couldn't easily handle array
specifications with fine-grained control of numerical format and shape.
The schema language is now relatively stable and does what it's designed to do,
but it being a domain-specific language rather than a general one makes it very
difficult to use NWB data alongside other formats.
`nwb_linkml` translates NWB to [linkml](https://linkml.io/), a schema language
for declaring **{index}`Linked Data`** schema. Linked Data schema consist of
semantic triplets, rather than an object hierarchy, and can make use of
**controlled vocabularies** to reuse terms and classes from other
schemas and ontologies.
## Storage Format Flexibility
**We want to use NWB in lots of different ways**
NWB as a format is designed with the intention for use with multiple
storage backends, but patterns and features of HDF5 have made their way
into the schema and the schema language, making direct translation to
other storage systems difficult. This is a problem for practical usage of
NWB data, since HDF5 files don't lend themselves to querying across many
files - eg. to find datasets that have some common piece of metadata, one
would have to download them all in full first. Having a whole hierarchy of
data in a single file is convenient in some ways, but this also makes them
difficult to share or split between computers which is a common need
when collecting data across multiple instruments and computers.
NWB, currently, lends itself towards being an **archival** format --- where
data is converted as a last step before publishing --- rather than a
**experimental** or **computational** format that can be used as a convenient
container of heterogeneous data during collection and analysis.
The LinkML team has also made a large number of [generators](https://linkml.io/linkml/generators/index.html)
to convert LinkML schema to different formats, including JSON Schema, GraphQL, SPARQL,
SQL/SQLAlchemy, and {mod}`~nwb_linkml.generators.pydantic`.
Since we have to use LinkML in a somewhat nonstandard way to accommodate
NWB's arrays, references, and naming conventions, these generators won't be
immediately available for use, but with some minor modification we should
be able to get NWB out of HDF5 files and into other formats.
## Zero-code Schema Extensions
**We want every researcher and every tool to have their own schemas.**
pynwb makes use of NWB Schema internally, but [schema extensions](https://pynwb.readthedocs.io/en/stable/tutorials/general/extensions.html#sphx-glr-tutorials-general-extensions-py)
require a decent amount of adjoining code to use. The underlying hdmf library
is relatively complex, and so to use a schema extension one must also
program the python classes or mappings to python class attributes
needed to use them, configuration for getter and setter methods,
i/o routines, etc. Since schema extensions are relatively hard to make,
