diff --git a/RELEASE_NOTES.md b/RELEASE_NOTES.md
index e38a1d3..60ef358 100644
--- a/RELEASE_NOTES.md
+++ b/RELEASE_NOTES.md
@@ -8,6 +8,14 @@ validate, patch, and query merged config files.
 
 ## Upgrading
 
+- The legacy config layouts are no longer migrated on load: top-level microgrid
+  IDs and `meta`-nested microgrid fields are gone. A file with no `assets` table
+  now loads as empty and contributes nothing to a merge, rather than having its
+  top-level keys read as microgrid IDs. This lets a mixed list of files (some
+  carrying an `[app]` table for their own config, some an `assets` table) be
+  passed to `AssetsConfig.load_from_files` / `load_configs`, which reads only the
+  `assets`-bearing ones. Rebuild any legacy-layout files with `generate-config`.
+
 - `generate-config` now writes the `assets.microgrids` layout and stamps
   `assets.version`, instead of the legacy layout. Regenerate configs to get the
   current format; generated files then load back without a migration.
@@ -40,3 +48,10 @@ validate, patch, and query merged config files.
 - `generate-config --inplace` refreshes managed values while preserving
   formatting; `--fill-missing` only adds absent values. Generated patches are
   validated before writing.
+
+## Bug Fixes
+
+- Config files with validity periods now load under marshmallow 3, not only
+  marshmallow 4. `tomllib` yields native `datetime` objects, which marshmallow
+  3's `DateTime` field rejected. `marshmallow` is now a direct dependency so the
+  minimum-version test exercises this path.
diff --git a/pyproject.toml b/pyproject.toml
index 0ed785c..9244e65 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -26,6 +26,10 @@ classifiers = [
 ]
 requires-python = ">= 3.11, < 4"
 dependencies = [
+  # Used transitively via marshmallow-dataclass (every config schema is a
+  # `marshmallow.Schema`); declared directly so `pytest_min` pins its lower
+  # bound and the minimum-version path is actually tested.
+  "marshmallow >= 3.18.0, < 5",
   "marshmallow-dataclass >= 8.7.1, < 9",
   "typing-extensions >= 4.14.1, < 5",
   "frequenz-microgrid-component-graph >= 0.5.0, < 0.6",
diff --git a/src/frequenz/gridpool/cli/_dump_config.py b/src/frequenz/gridpool/cli/_dump_config.py
index abbec28..68104c9 100644
--- a/src/frequenz/gridpool/cli/_dump_config.py
+++ b/src/frequenz/gridpool/cli/_dump_config.py
@@ -17,6 +17,7 @@
 fields are done here, since TOML has no null and `tomlkit` rejects it.
 """
 
+from dataclasses import asdict
 from typing import Any
 
 import tomlkit
@@ -86,13 +87,10 @@ def dump_map(configs: dict[int, MicrogridConfig]) -> str:
         The TOML representation as a string, with one blank line between
         microgrids and entries sorted by numeric microgrid ID.
     """
-    schema = MicrogridConfig.Schema()
     doc = tomlkit.document()
     doc.append(tomlkit.key(["assets", "version"]), _CURRENT_VERSION)
     for mid in sorted(configs):
-        dumped = schema.dump(configs[mid])
-        assert isinstance(dumped, dict)
-        leaves = _iter_leaves([*_MICROGRID_PREFIX, str(mid)], dumped)
+        leaves = _iter_leaves([*_MICROGRID_PREFIX, str(mid)], asdict(configs[mid]))
         if not leaves:
             continue
         doc.add(tomlkit.nl())
diff --git a/src/frequenz/gridpool/cli/_patch_config.py b/src/frequenz/gridpool/cli/_patch_config.py
index d550691..cd996ef 100644
--- a/src/frequenz/gridpool/cli/_patch_config.py
+++ b/src/frequenz/gridpool/cli/_patch_config.py
@@ -10,6 +10,7 @@
 """
 
 from collections.abc import Mapping
+from dataclasses import asdict
 from pathlib import Path
 from typing import Any
 
@@ -54,7 +55,6 @@ def patch_text(
     """
     doc = tomlkit.parse(original)
     _reject_legacy_layout(doc)
-    schema = MicrogridConfig.Schema()
 
     # Leaves needing a whole new sub-table can't be inserted via item assignment
     # without tomlkit falling back to bracket-header syntax, so they are spliced
@@ -63,9 +63,7 @@ def patch_text(
 
     for microgrid_id, cfg in configs.items():
         mid = str(microgrid_id)
-        dumped = schema.dump(cfg)
-        assert isinstance(dumped, dict)
-        leaves = _iter_leaves([], dumped)
+        leaves = _iter_leaves([], asdict(cfg))
         if not leaves:
             continue
 
diff --git a/src/frequenz/gridpool/config/_assets.py b/src/frequenz/gridpool/config/_assets.py
index 3b8d179..d5b0b87 100644
--- a/src/frequenz/gridpool/config/_assets.py
+++ b/src/frequenz/gridpool/config/_assets.py
@@ -443,23 +443,28 @@ def _read_assets_table(cls, config_path: Path) -> dict[str, Any]:
         """Read the raw `assets` table from a TOML file.
 
         The document is migrated to the current format before its `assets`
-        table is returned for merging.
+        table is returned for merging. A file with no `assets` table contributes
+        nothing to the merge, so consumers can pass a mixed list of files and
+        gridpool reads only the `assets`-bearing ones.
 
         Args:
             config_path: The path to the TOML configuration file.
 
         Returns:
-            The raw `assets` table, unvalidated, for merging before it is loaded.
+            The raw `assets` table, unvalidated, for merging before it is loaded,
+            or an empty table if the file has none.
 
         Raises:
-            TypeError: If `assets` is not a table.
+            TypeError: If `assets` is present but not a table.
         """
         with config_path.open("rb") as f:
             data: dict[str, Any] = tomllib.load(f)
 
         data = migrate(data, config_path)
 
-        assets = data["assets"]
+        assets = data.get("assets")
+        if assets is None:
+            return {}
         if not isinstance(assets, dict):
             raise TypeError(
                 f"{config_path}: `assets` must be a table, got {type(assets)}"
diff --git a/src/frequenz/gridpool/config/_fields.py b/src/frequenz/gridpool/config/_fields.py
new file mode 100644
index 0000000..39b96b8
--- /dev/null
+++ b/src/frequenz/gridpool/config/_fields.py
@@ -0,0 +1,29 @@
+# License: MIT
+# Copyright © 2026 Frequenz Energy-as-a-Service GmbH
+
+"""Reusable marshmallow fields for config data models."""
+
+from datetime import datetime
+from typing import Any
+
+import marshmallow
+
+
+class TomlAwareDateTime(marshmallow.fields.DateTime):
+    """A `DateTime` that also accepts an already-parsed `datetime`.
+
+    marshmallow 3's `DateTime` deserialises only ISO strings, but `tomllib`
+    yields native `datetime` objects before the value reaches marshmallow, so
+    those inputs must pass through unchanged. marshmallow 4 accepts them itself;
+    this keeps both behaving the same.
+    """
+
+    def _deserialize(self, value: Any, *args: Any, **kwargs: Any) -> datetime:
+        if isinstance(value, datetime):
+            return value
+        return super()._deserialize(value, *args, **kwargs)
+
+
+def toml_datetime_metadata() -> dict[str, Any]:
+    """Field metadata mapping a TOML-sourced `datetime` to `TomlAwareDateTime`."""
+    return {"marshmallow_field": TomlAwareDateTime(allow_none=True)}
diff --git a/src/frequenz/gridpool/config/_microgrid.py b/src/frequenz/gridpool/config/_microgrid.py
index 33f8bac..82c3b6a 100644
--- a/src/frequenz/gridpool/config/_microgrid.py
+++ b/src/frequenz/gridpool/config/_microgrid.py
@@ -11,6 +11,8 @@
 from marshmallow import Schema
 from marshmallow_dataclass import dataclass
 
+from ._fields import toml_datetime_metadata
+
 ComponentType = Literal["grid", "pv", "battery", "consumption", "chp", "ev"]
 """Valid component types."""
 
@@ -98,10 +100,10 @@ def is_valid_type(cls, ctype: str) -> bool:
 class PVConfig:
     """Configuration of a PV system in a microgrid."""
 
-    start_time: datetime | None = None
+    start_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Start time of the PV system installation."""
 
-    end_time: datetime | None = None
+    end_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """End time of the PV system installation."""
 
     peak_power: float | None = None
@@ -119,10 +121,10 @@ class WindConfig:
     # pylint: disable=too-many-instance-attributes
     """Configuration of a wind turbine in a microgrid."""
 
-    start_time: datetime | None = None
+    start_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Start time of the wind turbine installation."""
 
-    end_time: datetime | None = None
+    end_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """End time of the wind turbine installation."""
 
     turbine_model: str | None = None
@@ -151,10 +153,10 @@ class WindConfig:
 class BatteryConfig:
     """Configuration of a battery in a microgrid."""
 
-    start_time: datetime | None = None
+    start_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Start time of the battery installation."""
 
-    end_time: datetime | None = None
+    end_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """End time of the battery installation."""
 
     capacity: float | None = None
@@ -187,10 +189,10 @@ class MicrogridConfig:
     altitude: float | None = None
     """Geographic altitude of the microgrid."""
 
-    start_time: datetime | None = None
+    start_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Start time of the microgrid operation."""
 
-    end_time: datetime | None = None
+    end_time: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """End time of the microgrid operation."""
 
     pv: dict[str, PVConfig] | None = None
diff --git a/src/frequenz/gridpool/config/_migrations.py b/src/frequenz/gridpool/config/_migrations.py
index bd9dfca..da44970 100644
--- a/src/frequenz/gridpool/config/_migrations.py
+++ b/src/frequenz/gridpool/config/_migrations.py
@@ -3,18 +3,16 @@
 
 """Format migrations for raw config documents.
 
-Each step rewrites the raw `dict` a file parses to, before it is loaded into the
-typed model. Steps are self-detecting and idempotent, so the `version` field
-does not gate whether a step runs; it only records how old a file may be, which
-bounds when a step may be dropped.
+Loading validates `assets.version` and stamps the current version, so a file
+written for a newer reader is rejected rather than silently misread. Each
+migration step rewrites the raw `dict` before it is loaded into the typed
+model; steps are self-detecting and idempotent, so the `version` field does not
+gate whether a step runs. No legacy layouts remain, so `_STEPS` is empty.
 """
 
-import logging
 from pathlib import Path
 from typing import Any, Callable
 
-_logger = logging.getLogger(__name__)
-
 _CURRENT_VERSION = 1
 
 
@@ -27,7 +25,7 @@ def migrate(raw: dict[str, Any], source: Path) -> dict[str, Any]:
 
     Args:
         raw: The raw document as parsed from a file.
-        source: The file it was parsed from, for warnings and errors.
+        source: The file it was parsed from, named in error messages.
 
     Returns:
         The document in the current format.
@@ -56,84 +54,5 @@ def migrate(raw: dict[str, Any], source: Path) -> dict[str, Any]:
     return raw
 
 
-def _nest_top_level_microgrids(raw: dict[str, Any], source: Path) -> dict[str, Any]:
-    """Wrap deprecated top-level microgrid entries under `assets.microgrids`.
-
-    Args:
-        raw: The raw document.
-        source: The file it was parsed from.
-
-    Returns:
-        The document with its entries under `assets`.
-
-    Raises:
-        ValueError: If both layouts are present, a half-migrated file rather
-            than a merge.
-    """
-    if "assets" in raw:
-        stray = sorted(k for k in raw if k != "assets")
-        if stray:
-            raise ValueError(
-                f"{source}: keys {stray} sit outside `assets` while the file "
-                "already has an `assets` table; move them under `assets.microgrids`."
-            )
-        return raw
-
-    _logger.warning(
-        "%s: top-level microgrid IDs are deprecated, "
-        "nest the entries under `assets.microgrids` instead.",
-        source,
-    )
-    return {"assets": {"microgrids": dict(raw)}}
-
-
-def _lift_microgrid_meta(raw: dict[str, Any], source: Path) -> dict[str, Any]:
-    """Merge a deprecated nested `meta` table onto its microgrid entry.
-
-    Earlier files nested a microgrid's fields under `meta`; they now sit
-    directly on the entry.
-
-    Args:
-        raw: The raw document.
-        source: The file it was parsed from.
-
-    Returns:
-        The document with microgrid fields directly on each entry.
-
-    Raises:
-        TypeError: If a `meta` value is not a table.
-        ValueError: If a field is set both under `meta` and on the entry.
-    """
-    assets = raw.get("assets")
-    if not isinstance(assets, dict):
-        return raw
-    microgrids = assets.get("microgrids")
-    if not isinstance(microgrids, dict):
-        return raw
-    for mid, entry in microgrids.items():
-        if not isinstance(entry, dict) or "meta" not in entry:
-            continue
-        meta = entry.pop("meta")
-        if not isinstance(meta, dict):
-            raise TypeError(f"{source}: microgrid {mid} `meta` must be a table")
-        _logger.warning(
-            "%s: microgrid %s nests fields under `meta`; that is deprecated, "
-            "set them directly on the entry instead.",
-            source,
-            mid,
-        )
-        for key, value in meta.items():
-            if key in entry:
-                raise ValueError(
-                    f"{source}: microgrid {mid} sets `{key}` both under `meta` "
-                    "and directly; keep only the direct one"
-                )
-            entry[key] = value
-    return raw
-
-
-_STEPS: list[tuple[int, Callable[[dict[str, Any], Path], dict[str, Any]]]] = [
-    (0, _nest_top_level_microgrids),
-    (0, _lift_microgrid_meta),
-]
+_STEPS: list[tuple[int, Callable[[dict[str, Any], Path], dict[str, Any]]]] = []
 """Migration steps as `(from_version, step)`, oldest first."""
diff --git a/src/frequenz/gridpool/config/_topology.py b/src/frequenz/gridpool/config/_topology.py
index c17909a..99c7e08 100644
--- a/src/frequenz/gridpool/config/_topology.py
+++ b/src/frequenz/gridpool/config/_topology.py
@@ -26,6 +26,8 @@
 from marshmallow import Schema
 from marshmallow_dataclass import dataclass
 
+from ._fields import toml_datetime_metadata
+
 _EIC_ALPHABET = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ-"
 _MARKET_AREA_EU_DE = 101
 
@@ -165,10 +167,10 @@ class ValidityConfig:
     participation: MarketParticipationType | None = None
     """The market use case, unset for a plain validity window."""
 
-    start: datetime | None = None
+    start: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Inclusive start of the period."""
 
-    end: datetime | None = None
+    end: datetime | None = field(default=None, metadata=toml_datetime_metadata())
     """Exclusive end of the period."""
 
     Schema: ClassVar[Type[Schema]] = Schema
diff --git a/tests/test_config.py b/tests/test_config.py
index ca8bd87..0e57b87 100644
--- a/tests/test_config.py
+++ b/tests/test_config.py
@@ -121,17 +121,17 @@ def test_microgrid_config_formula(valid_microgrid_config: MicrogridConfig) -> No
 def test_load_configs(mocker: MockerFixture) -> None:
     """Test loading configurations for multiple microgrids from mock TOML files."""
     toml_data = """
-    1.microgrid_id = 1
-    1.name = "Test Grid"
-    1.gid = 1
-    1.ctype.pv.meter = [101, 102]
-    1.ctype.battery.inverter = [201, 202, 203]
-    1.ctype.battery.component = [301, 302, 303, 304, 305, 306]
-    1.pv.PV1.peak_power = 5000
-    1.pv.PV1.rated_power = 4500
-    1.pv.PV2.peak_power = 8000
-    1.pv.PV2.rated_power = 7000
-    1.battery.BAT1.capacity = 10000
+    assets.microgrids.1.microgrid_id = 1
+    assets.microgrids.1.name = "Test Grid"
+    assets.microgrids.1.gid = 1
+    assets.microgrids.1.ctype.pv.meter = [101, 102]
+    assets.microgrids.1.ctype.battery.inverter = [201, 202, 203]
+    assets.microgrids.1.ctype.battery.component = [301, 302, 303, 304, 305, 306]
+    assets.microgrids.1.pv.PV1.peak_power = 5000
+    assets.microgrids.1.pv.PV1.rated_power = 4500
+    assets.microgrids.1.pv.PV2.peak_power = 8000
+    assets.microgrids.1.pv.PV2.rated_power = 7000
+    assets.microgrids.1.battery.BAT1.capacity = 10000
     """
     mock_file = mocker.mock_open(read_data=toml_data.encode("utf-8"))
     mocker.patch("pathlib.Path.open", mock_file)
@@ -175,12 +175,6 @@ def _assert_optional_field(value: float | None, expected: float) -> None:
 assets.microgrids.1.ctype.pv.meter = [101, 102]
 """
 
-_LEGACY_TOML = """
-1.microgrid_id = 1
-1.name = "Test Grid"
-1.ctype.pv.meter = [101, 102]
-"""
-
 
 def _write(tmp_path: Path, name: str, text: str) -> Path:
     """Write `text` to `name` under `tmp_path` and return the path."""
@@ -200,47 +194,18 @@ def test_load_prefixed(tmp_path: Path, caplog: pytest.LogCaptureFixture) -> None
     assert "deprecated" not in caplog.text
 
 
-def test_load_legacy_warns(tmp_path: Path, caplog: pytest.LogCaptureFixture) -> None:
-    """Top-level entries still load, but warn and name the file."""
-    path = _write(tmp_path, "legacy.toml", _LEGACY_TOML)
-
-    with caplog.at_level(logging.WARNING):
-        configs = AssetsConfig.load_from_files(path).microgrids
-
-    assert configs[1].name == "Test Grid"
-    assert "deprecated" in caplog.text
-    assert str(path) in caplog.text
-
-
-def test_load_meta_layout_warns(
-    tmp_path: Path, caplog: pytest.LogCaptureFixture
-) -> None:
-    """A microgrid nesting its fields under `meta` still loads, but warns."""
-    path = _write(
-        tmp_path,
-        "meta.toml",
-        "assets.microgrids.1.meta.microgrid_id = 1\n"
-        'assets.microgrids.1.meta.name = "Test Grid"\n',
-    )
-
-    with caplog.at_level(logging.WARNING):
-        configs = AssetsConfig.load_from_files(path).microgrids
-
-    assert configs[1].name == "Test Grid"
-    assert "meta" in caplog.text
-    assert "deprecated" in caplog.text
-
-
-def test_load_mixed_layouts_rejected(tmp_path: Path) -> None:
-    """A half-migrated file with both layouts is an error, not a merge."""
+def test_load_ignores_keys_outside_assets(tmp_path: Path) -> None:
+    """Top-level keys next to `assets` are ignored; only `assets` is read."""
     path = _write(
         tmp_path,
         "mixed.toml",
         _PREFIXED_TOML + '2.microgrid_id = 2\n2.name = "Other Grid"\n',
     )
 
-    with pytest.raises(ValueError, match="outside `assets`"):
-        AssetsConfig.load_from_files(path)
+    configs = AssetsConfig.load_from_files(path).microgrids
+
+    assert set(configs) == {1}
+    assert configs[1].name == "Test Grid"
 
 
 def test_load_assets_without_microgrids(tmp_path: Path) -> None:
@@ -250,18 +215,16 @@ def test_load_assets_without_microgrids(tmp_path: Path) -> None:
     assert not AssetsConfig.load_from_files(path).microgrids
 
 
-async def test_merge_prefixed_base_with_legacy_override(tmp_path: Path) -> None:
-    """Layers of either layout merge as usual, since layout is resolved on load."""
+async def test_merge_ignores_a_non_assets_override(tmp_path: Path) -> None:
+    """An override file with no `assets` table contributes nothing to the merge."""
     base = _write(tmp_path, "base.toml", _PREFIXED_TOML)
-    override = _write(
-        tmp_path, "override.toml", '1.microgrid_id = 1\n1.name = "Renamed"\n'
-    )
+    override = _write(tmp_path, "override.toml", '[app]\nname = "consumer"\n')
 
     configs = (
         await load_configs(default_files=base, override_files=override)
     ).microgrids
 
-    assert configs[1].name == "Renamed"
+    assert configs[1].name == "Test Grid"
     assert configs[1].component_type_ids("pv") == [101, 102]
 
 
diff --git a/tests/test_migrations.py b/tests/test_migrations.py
index 8e1f213..8a8ed59 100644
--- a/tests/test_migrations.py
+++ b/tests/test_migrations.py
@@ -11,31 +11,6 @@
 from frequenz.gridpool.config._migrations import _CURRENT_VERSION, migrate
 
 
-def test_migrate_nests_top_level_microgrids() -> None:
-    """A v0 top-level document is nested under `assets.microgrids` and stamped."""
-    raw = {"1": {"microgrid_id": 1, "name": "Grid"}}
-
-    migrated = migrate(raw, Path("v0.toml"))
-
-    assert migrated == {
-        "assets": {
-            "microgrids": {"1": {"microgrid_id": 1, "name": "Grid"}},
-            "version": _CURRENT_VERSION,
-        }
-    }
-
-
-def test_migrate_lifts_nested_meta() -> None:
-    """A microgrid nesting fields under `meta` has them lifted onto the entry."""
-    raw = {
-        "assets": {"microgrids": {"1": {"meta": {"microgrid_id": 1, "name": "Grid"}}}}
-    }
-
-    migrated = migrate(raw, Path("meta.toml"))
-
-    assert migrated["assets"]["microgrids"]["1"] == {"microgrid_id": 1, "name": "Grid"}
-
-
 def test_migrate_is_a_noop_on_a_current_document() -> None:
     """A current document passes through unchanged."""
     raw = {
@@ -48,20 +23,6 @@ def test_migrate_is_a_noop_on_a_current_document() -> None:
     assert migrate(raw, Path("current.toml")) == raw
 
 
-def test_migrate_rejects_mixed_layout_at_current_version() -> None:
-    """Layout validation still runs when a document declares the current version."""
-    raw = {
-        "assets": {
-            "version": _CURRENT_VERSION,
-            "microgrids": {"1": {"microgrid_id": 1}},
-        },
-        "2": {"microgrid_id": 2},
-    }
-
-    with pytest.raises(ValueError, match="outside `assets`"):
-        migrate(raw, Path("mixed.toml"))
-
-
 def test_migrate_rejects_a_newer_version() -> None:
     """A future document is not silently downgraded."""
     future_version = _CURRENT_VERSION + 1
@@ -80,12 +41,30 @@ def test_migrate_rejects_a_newer_version() -> None:
     assert raw["assets"]["version"] == future_version
 
 
-def test_v0_top_level_file_loads(tmp_path: Path) -> None:
-    """A v0 top-level file still loads through the public loader."""
-    path = tmp_path / "v0.toml"
-    path.write_text('1.microgrid_id = 1\n1.name = "Grid"\n')
+def test_file_without_assets_loads_empty(tmp_path: Path) -> None:
+    """A file with no `assets` table loads empty and contributes nothing to a merge."""
+    other = tmp_path / "app.toml"
+    other.write_text('[app]\nname = "consumer"\n')
+    real = tmp_path / "assets.toml"
+    real.write_text(
+        'assets.microgrids.1.microgrid_id = 1\nassets.microgrids.1.name = "Grid"\n'
+    )
+
+    assert AssetsConfig.load_from_files(other).microgrids == {}
+
+    merged = AssetsConfig.load_from_files([other, real])
+    assert merged.microgrids[1].name == "Grid"
+
+
+def test_file_with_app_and_assets_reads_only_assets(tmp_path: Path) -> None:
+    """A sibling `[app]` table next to `assets` is ignored; only `assets` is read."""
+    path = tmp_path / "mixed.toml"
+    path.write_text(
+        "assets.microgrids.1.microgrid_id = 1\n"
+        'assets.microgrids.1.name = "Grid"\n\n'
+        '[app]\nname = "consumer"\n'
+    )
 
     config = AssetsConfig.load_from_files(path)
 
     assert config.microgrids[1].name == "Grid"
-    assert config.version == _CURRENT_VERSION
diff --git a/tests/test_topology.py b/tests/test_topology.py
index 3c1a28c..d71c402 100644
--- a/tests/test_topology.py
+++ b/tests/test_topology.py
@@ -575,6 +575,27 @@ def test_period_and_query_require_a_utc_offset() -> None:
         relation.covers(naive)
 
 
+def test_validity_accepts_datetime_objects_and_iso_strings() -> None:
+    """A period loads from native datetimes, as `tomllib` yields, or ISO strings."""
+    aware = datetime(2026, 1, 1, tzinfo=timezone.utc)
+
+    from_objects = ValidityConfig.Schema().load(
+        {"start": aware, "end": aware.replace(month=6)}
+    )
+    assert isinstance(from_objects, ValidityConfig)
+    assert from_objects.start == aware
+
+    from_strings = ValidityConfig.Schema().load(
+        {"start": "2026-01-01T00:00:00+00:00", "end": "2026-06-01T00:00:00+00:00"}
+    )
+    assert isinstance(from_strings, ValidityConfig)
+    assert from_strings.start == aware
+
+    naive = datetime(2026, 1, 1)  # noqa: DTZ001
+    with pytest.raises(ValueError, match="UTC offset"):
+        ValidityConfig.Schema().load({"start": naive})
+
+
 def test_market_location_properties() -> None:
     """A location identifies its market and how to read its identifier."""
     config = _load(