diff --git a/src/nxmx/__init__.py b/src/nxmx/__init__.py
index 5fc185a..515f3f0 100644
--- a/src/nxmx/__init__.py
+++ b/src/nxmx/__init__.py
@@ -12,14 +12,15 @@
 import dateutil.parser
 import h5py
 import numpy as np
+import numpy.typing as npt
 import pint
 from scipy.spatial.transform import Rotation
 
 # NeXus field type for type annotations
 # https://manual.nexusformat.org/nxdl-types.html#nxdl-field-types-and-units
-NXBoolT: TypeAlias = bool | np.ndarray
-NXFloatT: TypeAlias = float | np.ndarray
-NXIntT: TypeAlias = int | np.ndarray
+NXBoolT: TypeAlias = np.bool_ | npt.NDArray[np.bool_]
+NXFloatT: TypeAlias = np.floating | npt.NDArray[np.floating]
+NXIntT: TypeAlias = np.integer | npt.NDArray[np.integer]
 NXNumberT: TypeAlias = NXFloatT | NXIntT
 
 
@@ -46,20 +47,52 @@ def __len__(self):
         return len(self._handle)
 
 
-def h5scalar(ds: h5py.Dataset):
-    """Read a scalar value from an HDF5 dataset
+def h5scalar(ds: h5py.Dataset) -> np.generic:
+    """
+    Read a scalar value from an HDF5 dataset
 
-    Sometimes scalars are stored as a length-1 1D dataset instead of a proper
-    scalar. This function allows for that, since NumPy has got stricter about
-    converting to scalars.
+    Raises: ValueError, if the dataset does not contain a scalar.
     """
-    if ds.size != 1:
-        raise ValueError("only length-1 arrays can be converted to Python scalars")
-    arr = np.squeeze(ds[()])
-    return arr.item()
+    value = h5_maybe_scalar(ds)
+    if not isinstance(value, np.generic):
+        raise ValueError("Cannot be converted to numpy scalar")
+    return value
+
+
+def h5_maybe_scalar(ds: h5py.Dataset) -> np.generic | npt.NDArray[np.generic]:
+    """
+    Coerce an HDF5 dataset to a NumPy scalar, if appropriate.
+
+    Sometimes scalars are stored as a length-1 1D dataset instead of a
+    scalar, this converts those instances to a scalar.
+
+    Only changes when:
+    - It is already a 0-D array
+    - It is a 1-D array of length 1
+
+    Otherwise, the array is returned as-is, with dtype preserved.
+    """
+    data = ds[()]
+
+    if isinstance(data, np.ndarray):
+        if data.ndim == 0:
+            # 0-D array → NumPy scalar
+            return data[()]
+        elif data.ndim == 1 and data.size == 1:
+            # 1-D length-1 array → NumPy scalar
+            return data[0]
+        else:
+            # multi-element array → leave as-is
+            return data
+    else:
+        if isinstance(data, np.generic):
+            # already a NumPy scalar → leave as-is
+            return data
+        # Explicitly promote this into a numpy scalar
+        return np.asarray(data).flat[0]
 
 
-def h5str(h5_value: str | np.bytes_ | bytes | None) -> str | None:
+def h5str(h5_value: str | np.bytes_ | np.str_ | bytes | None) -> str | None:
     """
     Convert a value returned from an h5py attribute to str.
 
@@ -69,6 +102,9 @@ def h5str(h5_value: str | np.bytes_ | bytes | None) -> str | None:
     """
     if isinstance(h5_value, np.bytes_ | bytes):
         return h5_value.decode("utf-8")
+    if isinstance(h5_value, np.str_):
+        return str(h5_value)
+    assert isinstance(h5_value, str) or h5_value is None
     return h5_value
 
 
@@ -246,7 +282,7 @@ def signal(self) -> str | None:
         return self._handle.attrs.get("signal")
 
     @cached_property
-    def data_scale_factor(self) -> str | None:
+    def data_scale_factor(self) -> NXNumberT | None:
         """
         An optional scaling factor to apply to the values in ``data``.
 
@@ -278,17 +314,17 @@ def data_scale_factor(self) -> str | None:
         When omitted, the scaling factor is assumed to be 1.
         """
         if "data_scale_factor" in self._handle:
-            return h5scalar(self._handle["data_scale_factor"])
+            return h5_maybe_scalar(self._handle["data_scale_factor"])
 
     @cached_property
-    def data_offset(self) -> str | None:
+    def data_offset(self) -> NXNumberT | None:
         """
         An optional offset to apply to the values in data.
 
         When omitted, the offset is assumed to be 0.
         """
         if "data_offset" in self._handle:
-            return h5scalar(self._handle["data_offset"])
+            return h5_maybe_scalar(self._handle["data_offset"])
 
 
 class NXtransformations(H5Mapping):
@@ -575,7 +611,7 @@ def depends_on(self) -> NXtransformationsAxis | None:
     def temperature(self) -> pint.Quantity | None:
         """The temperature of the sample."""
         if temperature := self._handle.get("temperature"):
-            return h5scalar(temperature) * units(temperature)
+            return h5_maybe_scalar(temperature) * units(temperature)
         return None
 
     @cached_property