diff --git a/pyproject.toml b/pyproject.toml
index 2e9f399..f45a25c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -23,7 +23,8 @@ classifiers = [
 requires-python = ">=3.9"
 dependencies = [
     "attrs >=24.1",
-    "bidsschematools >= 1.0.10",
+    "bidsschematools @ git+https://github.com/bids-standard/bids-specification.git@refs/pull/2133/head#subdirectory=tools/schemacode",
+    "universal_pathlib >=0.2",
 ]
 
 [project.optional-dependencies]
diff --git a/src/bids_validator/context.py b/src/bids_validator/context.py
new file mode 100644
index 0000000..434018c
--- /dev/null
+++ b/src/bids_validator/context.py
@@ -0,0 +1,272 @@
+"""Classes and functions for populating a validation context."""
+
+from __future__ import annotations
+
+import itertools
+import json
+from functools import cache
+
+import attrs
+from bidsschematools.types import Namespace
+from bidsschematools.types import context as ctx
+from upath import UPath
+
+from .types import _typings as t
+from .types.files import FileTree
+
+TYPE_CHECKING = False
+if TYPE_CHECKING:
+    from collections.abc import Generator
+
+    from bidsschematools.types import protocols as proto
+
+    # pyright does not treat cached_property like property
+    cached_property = property
+else:
+    from functools import cached_property
+
+
+# Design strategy for these classes.
+#
+# *  Attribute access is essential. Dictionary lookups must become either
+#    dataclasses or Namespace-like objects.
+# *  Objects should have a single source of truth, such as a FileTree object.
+#    When defining derived fields, prefer @property if the lookup is trivial
+#    or cached_property, if it is nontrivial.
+# *  Hitting the filesystem should be minimized. FileTree already caches os.stat calls,
+#    but reading file contents should never happen more than once. Therefore, if file
+#    contents are accessed in the context.associations or context.dataset, they should
+#    use caching loaders.
+# *  If the full contents of an object will be known on its first instantiation,
+#    prefer to use the dataclasses in bidsschematools.types.context. Lazy fields
+#    need custom classes.
+
+
+class ValidationError(Exception):
+    """TODO: Add issue structure."""
+
+
+@cache
+def load_tsv(file: FileTree, *, max_rows=0) -> Namespace:
+    """Load TSV contents into a Namespace."""
+    with open(file) as fobj:
+        if max_rows > 0:
+            fobj = itertools.islice(fobj, max_rows)
+        contents = (line.rstrip('\r\n').split('\t') for line in fobj)
+        # Extract headers then transpose rows to columns
+        return Namespace(zip(next(contents), zip(*contents)))
+
+
+@cache
+def load_json(file: FileTree) -> dict[str]:
+    """Load JSON file contents."""
+    with open(file) as fobj:
+        return json.load(fobj)
+
+
+class Subjects:
+    """Collections of subjects in the dataset."""
+
+    def __init__(self, tree: FileTree):
+        self._tree = tree
+
+    @cached_property
+    def sub_dirs(self) -> list[str]:
+        """Subjects as determined by sub-* directories."""
+        return [
+            child.name
+            for child in self._tree.children.values()
+            if child.is_dir and child.name.startswith('sub-')
+        ]
+
+    @property
+    def participant_id(self) -> list[str] | None:
+        """The participant_id column of participants.tsv."""
+        if 'participants.tsv' not in self._tree.children:
+            return None
+
+        return self._get_participant_id(self._tree.children['participants.tsv'])
+
+    @cached_property
+    def phenotype(self) -> list[str] | None:
+        """The union of participant_id columns in phenotype files."""
+        if 'phenotype' not in self._tree.children:
+            return None
+
+        subjects = set()
+        for phenotype_file in self._tree.children['phenotype'].children:
+            if phenotype_file.name.endswith('.tsv'):
+                subjects.update(self._get_participant_id(phenotype_file))
+
+        return sorted(subjects)
+
+    @staticmethod
+    def _get_participant_id(phenotype_file: FileTree) -> list[str] | None:
+        columns = load_tsv(phenotype_file)
+        if 'participant_id' not in columns:
+            return None
+        return list(columns['participant_id'])
+
+
+@attrs.define
+class Dataset:
+    """A dataset object that loads properties on first access."""
+
+    tree: FileTree
+    ignored: list[str] = attrs.field(factory=list)
+    subjects: Subjects = attrs.field(init=False)
+
+    def __attrs_post_init__(self):
+        self.subjects = Subjects(self.tree)
+
+    @cached_property
+    def dataset_description(self) -> Namespace:
+        """Contents of '/dataset_description.json'."""
+        return Namespace.from_json(
+            UPath(self.tree.children['dataset_description.json']).read_text()
+        )
+
+    @cached_property
+    def modalities(self) -> list[str]:
+        """List of modalities found in the dataset."""
+        ...
+        return []
+
+    @cached_property
+    def datatypes(self) -> list[str]:
+        """List of datatypes found in the dataset."""
+        ...
+        return []
+
+
+@attrs.define
+class Association:
+    """Generic association, exposing the associated file's path."""
+
+    _file: FileTree
+
+    @property
+    def path(self):
+        """Dataset-relative path of the associated file."""
+        return self._file.relative_path
+
+
+def load_file(file: FileTree, dataset: proto.Dataset) -> ctx.Context:
+    """Load a full context for a given file."""
+    associations = load_associations(file, dataset)
+    _ = associations
+
+
+def load_associations(file: FileTree, dataset: proto.Dataset) -> ctx.Associations:
+    """Load all associations for a given file."""
+    # If something fails, return None.
+    # Uses walk back algorithm
+    # https://bids-validator.readthedocs.io/en/latest/validation-model/inheritance-principle.html
+    # Stops on first success
+
+
+def load_events(file: FileTree) -> ctx.Events:
+    """Load events.tsv file."""
+
+
+def load_sidecar(file: FileTree) -> dict[str, t.Any]:
+    """Load sidecar metadata, using the inheritance principle."""
+    # Uses walk back algorithm
+    # https://bids-validator.readthedocs.io/en/latest/validation-model/inheritance-principle.html
+    # Accumulates all sidecars
+
+
+def walk_back(
+    source: FileTree,
+    inherit: bool,
+    target_extensions: tuple[str, ...] = ('.json',),
+    target_suffix: str | None = None,
+    target_entities: tuple[str, ...] = (),
+) -> Generator[FileTree] | Generator[list[FileTree, ...]]:
+    """Walk up the file tree to find associated files."""
+    for file_group in _walk_back(
+        source, inherit, target_extensions, target_suffix, target_entities
+    ):
+        if target_entities:
+            yield file_group
+        elif len(file_group) == 1:
+            yield file_group[0]
+        else:
+            raise ValidationError('Multiple matching files.')
+
+
+def _walk_back(
+    source: FileTree,
+    inherit: bool,
+    target_extensions: tuple[str, ...],
+    target_suffix: str | None,
+    target_entities: tuple[str, ...],
+) -> Generator[list[FileTree, ...]]:
+    file_parts = FileParts.from_file(source.relative_path)
+
+    if target_suffix is None:
+        target_suffix = file_parts.suffix
+
+    tree = source.parent
+    while tree:
+        matches = []
+        for child in tree.children:
+            if child.is_dir:
+                continue
+            parts = FileParts.from_file(child.relative_path)
+            if parts.extension != target_extensions:
+                continue
+            if parts.suffix != target_suffix:
+                continue
+            if all(
+                key in target_entities or file_parts.entities.get(key) == value
+                for key, value in parts.entities.items()
+            ):
+                matches.append(child)
+
+        yield matches
+        if not inherit:
+            break
+        tree = tree.parent
+
+
+@attrs.define
+class FileParts:
+    """BIDS-relevant components of a file path."""
+
+    path: str
+    stem: str
+    entities: dict[str, str | None]
+    datatype: str | None
+    suffix: str | None
+    extension: str | None
+
+    @classmethod
+    def from_file(cls, file: FileTree, schema: Namespace) -> t.Self:
+        """Parse file parts from FileTree object."""
+        stem, _, extension = file.name.partition('.')
+
+        if extension:
+            extension = f'.{extension}'
+        if file.is_dir:
+            extension = f'{extension}/'
+
+        datatype = None
+        if file.parent:
+            if any(file.parent.name == dtype.value for dtype in schema.objects.datatypes.values()):
+                datatype = file.parent.name
+
+        *entity_strings, suffix = stem.split('_')
+        entities = {
+            key: vals[0] if vals else None
+            for key, *vals in (string.split('-', 1) for string in entity_strings)
+        }
+
+        return cls(
+            path=f'/{file.relative_path}',
+            stem=stem,
+            entities=entities,
+            datatype=datatype,
+            suffix=suffix,
+            extension=extension,
+        )
diff --git a/src/bids_validator/types/files.py b/src/bids_validator/types/files.py
index 18f04e4..a3a6f62 100644
--- a/src/bids_validator/types/files.py
+++ b/src/bids_validator/types/files.py
@@ -112,6 +112,16 @@ def __contains__(self, relpath: os.PathLike) -> bool:
     def __fspath__(self):
         return self.direntry.path
 
+    def __hash__(self):
+        return hash(self.direntry.path)
+
+    def __truediv__(self, relpath: str | os.PathLike) -> t.Self:
+        parts = Path(relpath).parts
+        child = self
+        for part in parts:
+            child = child.children[part]
+        return child
+
     @cached_property
     def relative_path(self) -> str:
         """The path of the current FileTree, relative to the root.
diff --git a/tests/conftest.py b/tests/conftest.py
index 99dbd5c..12fa33f 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -5,6 +5,8 @@
 from pathlib import Path
 
 import pytest
+from bidsschematools.schema import load_schema
+from bidsschematools.types import Namespace
 
 
 @pytest.fixture(scope='session')
@@ -31,3 +33,9 @@ def gitignore_test() -> Path:
     else:  # pragma: no cover
         pass
     return Path(ret)
+
+
+@pytest.fixture(scope='session')
+def schema() -> Namespace:
+    """Load BIDS schema for tests."""
+    return load_schema()
diff --git a/tests/test_context.py b/tests/test_context.py
new file mode 100644
index 0000000..c58c6f7
--- /dev/null
+++ b/tests/test_context.py
@@ -0,0 +1,26 @@
+from bids_validator import context
+from bids_validator.types.files import FileTree
+
+
+def test_load(examples):
+    tree = FileTree.read_from_filesystem(examples / 'synthetic')
+    ds = context.Dataset(tree)
+
+    assert ds.dataset_description.Name.startswith('Synthetic dataset')
+    assert ds.subjects.participant_id == [f'sub-{i:02d}' for i in range(1, 6)]
+    assert sorted(ds.subjects.sub_dirs) == [f'sub-{i:02d}' for i in range(1, 6)]
+
+
+def test_fileparts(examples, schema):
+    tree = FileTree.read_from_filesystem(examples / 'synthetic')
+
+    T1w = tree / 'sub-01' / 'ses-01' / 'anat' / 'sub-01_ses-01_T1w.nii'
+    parts = context.FileParts.from_file(T1w, schema)
+    assert parts == context.FileParts(
+        path='/sub-01/ses-01/anat/sub-01_ses-01_T1w.nii',
+        stem='sub-01_ses-01_T1w',
+        entities={'sub': '01', 'ses': '01'},
+        datatype='anat',
+        suffix='T1w',
+        extension='.nii',
+    )