diff --git a/autolens/point/dataset.py b/autolens/point/dataset.py
index c14ef46bc..55d2b722e 100644
--- a/autolens/point/dataset.py
+++ b/autolens/point/dataset.py
@@ -31,6 +31,7 @@
 _BASE_HEADERS = ["name", "y", "x", "positions_noise"]
 _FLUX_HEADERS = ["flux", "flux_noise"]
 _TIME_DELAY_HEADERS = ["time_delay", "time_delay_noise"]
+_REDSHIFT_HEADERS = ["redshift"]
 
 
 class PointDataset:
@@ -45,6 +46,7 @@ def __init__(
         time_delays_noise_map: Optional[
             Union[float, aa.ArrayIrregular, List[float]]
         ] = None,
+        redshift: Optional[float] = None,
     ):
         """
         A collection of the data component that can be used for point-source model-fitting, for example fitting the
@@ -73,6 +75,9 @@ def __init__(
             The time delays of each observed point-source of light in days.
         time_delays_noise_map
             The noise-value of every observed time delay, which is typically measured from the time delay analysis.
+        redshift
+            The redshift of the source. Optional; when provided it is carried through CSV round-trips alongside
+            the positions so cluster-scale workflows can encode per-source redshifts in a single spreadsheet.
         """
 
         self.name = name
@@ -111,6 +116,8 @@ def convert_to_array_irregular(values):
         self.time_delays = convert_to_array_irregular(time_delays)
         self.time_delays_noise_map = convert_to_array_irregular(time_delays_noise_map)
 
+        self.redshift = float(redshift) if redshift is not None else None
+
     @property
     def info(self) -> str:
         """
@@ -125,6 +132,7 @@ def info(self) -> str:
         info += f"fluxes_noise_map : {self.fluxes_noise_map}\n"
         info += f"time_delays : {self.time_delays}\n"
         info += f"time_delays_noise_map : {self.time_delays_noise_map}\n"
+        info += f"redshift : {self.redshift}\n"
         return info
 
     def extent_from(self, buffer: float = 0.1):
@@ -202,9 +210,12 @@ def output_to_csv(datasets: List[PointDataset], file_path: str):
     image.
 
     The base columns (``name, y, x, positions_noise``) are always written.  The
-    optional ``flux``/``flux_noise`` and ``time_delay``/``time_delay_noise`` columns
-    are included when *any* dataset in ``datasets`` carries those values; datasets
-    that do not carry them leave those cells blank.
+    optional ``flux``/``flux_noise``, ``time_delay``/``time_delay_noise`` and
+    ``redshift`` columns are included when *any* dataset in ``datasets`` carries
+    those values; datasets that do not carry them leave those cells blank.
+
+    When written, every row in a given ``name`` group repeats the same ``redshift``
+    value — the source redshift is a per-source property, not per-image.
 
     This is the hand-editable / spreadsheet form preferred for strong-lens cluster
     workflows with tens or hundreds of multiply-imaged sources.  For exact
@@ -212,12 +223,15 @@ def output_to_csv(datasets: List[PointDataset], file_path: str):
     """
     include_flux = any(d.fluxes is not None for d in datasets)
     include_time_delay = any(d.time_delays is not None for d in datasets)
+    include_redshift = any(d.redshift is not None for d in datasets)
 
     headers = list(_BASE_HEADERS)
     if include_flux:
         headers += _FLUX_HEADERS
     if include_time_delay:
         headers += _TIME_DELAY_HEADERS
+    if include_redshift:
+        headers += _REDSHIFT_HEADERS
 
     rows = []
     for dataset in datasets:
@@ -247,6 +261,10 @@ def output_to_csv(datasets: List[PointDataset], file_path: str):
                 row["time_delay_noise"] = (
                     "" if time_delays_noise is None else time_delays_noise[i]
                 )
+            if include_redshift:
+                row["redshift"] = (
+                    "" if dataset.redshift is None else dataset.redshift
+                )
             rows.append(row)
 
     csvable.output_to_csv(rows, file_path, headers=headers)
@@ -270,17 +288,47 @@ def _float_column(
     return [float(v) for v in raw]
 
 
+def _group_redshift(
+    group_rows: List[dict], group_name: str
+) -> Optional[float]:
+    raw = [row.get("redshift", "") for row in group_rows]
+    populated = [v for v in raw if v not in ("", None)]
+
+    if not populated:
+        return None
+
+    if len(populated) != len(raw):
+        raise ValueError(
+            f"CSV group {group_name!r} has partially populated column "
+            f"'redshift'; every row in the group must have a value or all be blank."
+        )
+
+    values = [float(v) for v in populated]
+    if any(v != values[0] for v in values):
+        raise ValueError(
+            f"CSV group {group_name!r} has inconsistent 'redshift' values "
+            f"{values!r}; a source redshift must be identical across all of its "
+            f"image rows."
+        )
+
+    return values[0]
+
+
 def list_from_csv(file_path: str) -> List[PointDataset]:
     """
     Load a list of ``PointDataset`` objects from a CSV written by
     :func:`output_to_csv` (or :meth:`PointDataset.to_csv`).
 
     Rows are grouped by their ``name`` column — one ``PointDataset`` per distinct
-    name, preserving the order of first appearance.  Optional columns
+    name, preserving the order of first appearance.  Optional per-image columns
     (``flux``/``flux_noise``, ``time_delay``/``time_delay_noise``) are carried through
     per-group: if every row in a group populates the column the values are loaded,
     if every row leaves it blank the corresponding attribute is set to ``None``, and
     any partial-population is rejected with a ``ValueError``.
+
+    The optional ``redshift`` column is per-source (not per-image): every row within
+    a group must share the same value.  A group with mixed or differing redshifts is
+    rejected with a ``ValueError``.
     """
     rows = csvable.list_from_csv(file_path)
 
@@ -304,6 +352,7 @@ def list_from_csv(file_path: str) -> List[PointDataset]:
     has_flux_noise_column = "flux_noise" in headers
     has_time_delay_column = "time_delay" in headers
     has_time_delay_noise_column = "time_delay_noise" in headers
+    has_redshift_column = "redshift" in headers
 
     datasets: List[PointDataset] = []
     for name, group_rows in groups.items():
@@ -332,6 +381,11 @@ def list_from_csv(file_path: str) -> List[PointDataset]:
             if has_time_delay_noise_column
             else None
         )
+        redshift = (
+            _group_redshift(group_rows, name)
+            if has_redshift_column
+            else None
+        )
 
         datasets.append(
             PointDataset(
@@ -342,6 +396,7 @@ def list_from_csv(file_path: str) -> List[PointDataset]:
                 fluxes_noise_map=fluxes_noise_map,
                 time_delays=time_delays,
                 time_delays_noise_map=time_delays_noise_map,
+                redshift=redshift,
             )
         )
 
diff --git a/test_autolens/point/test_dataset.py b/test_autolens/point/test_dataset.py
index e4da0b0e2..4470ed8e0 100644
--- a/test_autolens/point/test_dataset.py
+++ b/test_autolens/point/test_dataset.py
@@ -29,6 +29,10 @@ def _assert_dataset_equal(actual: al.PointDataset, expected: al.PointDataset):
     _assert_array_close(actual.fluxes_noise_map, expected.fluxes_noise_map)
     _assert_array_close(actual.time_delays, expected.time_delays)
     _assert_array_close(actual.time_delays_noise_map, expected.time_delays_noise_map)
+    if expected.redshift is None:
+        assert actual.redshift is None
+    else:
+        assert actual.redshift == pytest.approx(expected.redshift)
 
 
 def test__csv_round_trip__positions_only(tmp_path):
@@ -133,6 +137,70 @@ def test__csv_list_round_trip__heterogeneous_optional_columns(tmp_path):
     assert loaded[1].fluxes_noise_map is None
 
 
+def test__csv_round_trip__redshift(tmp_path):
+    dataset = al.PointDataset(
+        name="source_0",
+        positions=[(0.5, 1.0), (-0.25, 2.0), (1.5, -1.0)],
+        positions_noise_map=[0.05, 0.05, 0.1],
+        redshift=2.5,
+    )
+
+    file_path = os.path.join(tmp_path, "point_dataset.csv")
+    dataset.to_csv(file_path)
+
+    loaded = al.PointDataset.from_csv(file_path)
+
+    _assert_dataset_equal(loaded, dataset)
+    assert loaded.redshift == pytest.approx(2.5)
+
+
+def test__csv_list_round_trip__mixed_redshift_presence(tmp_path):
+    with_redshift = al.PointDataset(
+        name="source_0",
+        positions=[(0.0, 0.0), (1.0, 1.0)],
+        positions_noise_map=[0.05, 0.05],
+        redshift=1.8,
+    )
+    without_redshift = al.PointDataset(
+        name="source_1",
+        positions=[(2.0, 0.5), (-1.0, 0.5)],
+        positions_noise_map=[0.1, 0.1],
+    )
+
+    file_path = os.path.join(tmp_path, "point_datasets.csv")
+    al.output_to_csv([with_redshift, without_redshift], file_path)
+
+    loaded = al.list_from_csv(file_path)
+
+    assert [d.name for d in loaded] == ["source_0", "source_1"]
+    _assert_dataset_equal(loaded[0], with_redshift)
+    _assert_dataset_equal(loaded[1], without_redshift)
+    assert loaded[0].redshift == pytest.approx(1.8)
+    assert loaded[1].redshift is None
+
+
+def test__list_from_csv__inconsistent_redshift_raises(tmp_path):
+    file_path = os.path.join(tmp_path, "point_datasets.csv")
+    with open(file_path, "w") as f:
+        f.write("name,y,x,positions_noise,redshift\n")
+        f.write("source_0,0.0,0.0,0.05,1.5\n")
+        f.write("source_0,1.0,1.0,0.05,2.0\n")
+
+    with pytest.raises(ValueError, match="inconsistent 'redshift'"):
+        al.list_from_csv(file_path)
+
+
+def test__list_from_csv__partial_redshift_raises(tmp_path):
+    file_path = os.path.join(tmp_path, "point_datasets.csv")
+    with open(file_path, "w") as f:
+        f.write("name,y,x,positions_noise,redshift\n")
+        f.write("source_0,0.0,0.0,0.05,1.5\n")
+        f.write("source_0,1.0,1.0,0.05,\n")
+
+    with pytest.raises(ValueError, match="partially populated column 'redshift'"):
+        al.list_from_csv(file_path)
+
+
 def test__from_csv__multiple_groups_requires_name(tmp_path):
     datasets = [
         al.PointDataset(