diff --git a/soil_id/global_soil.py b/soil_id/global_soil.py
index 3155ad5..c4e415c 100644
--- a/soil_id/global_soil.py
+++ b/soil_id/global_soil.py
@@ -555,6 +555,27 @@ def convert_to_serializable(obj):
     )
 
 
+def _slice_gower_distance(slice_mat):
+    """Gower distance matrix for a single depth slice.
+
+    A slice can have zero feature columns when a depth has no usable measurements
+    — e.g. the user recorded a depth interval but left texture/rock-fragment/color
+    blank, or (after depth-aligning) the horizon data simply does not reach this
+    depth. ``gower_distances`` would feed a shape ``(n, 0)`` array into its mean
+    imputer and raise "Found array with 0 feature(s)…", failing the whole ranking.
+
+    For such a slice, return an all-NaN ``(n, n)`` matrix instead. Callers treat NaN
+    distances as "no information" (the masked average and NaN-infill steps), so the
+    slice is ignored and components rank on the depths that do have data. Returning a
+    matrix (rather than skipping) keeps the per-slice list aligned with soil_matrix
+    rows.
+    """
+    if slice_mat.shape[1] == 0:
+        n = slice_mat.shape[0]
+        return np.full((n, n), np.nan)
+    return gower_distances(slice_mat)
+
+
 ##############################################################################################
 #                                   rankPredictionGlobal                                     #
 ##############################################################################################
@@ -828,8 +849,11 @@ def rank_soils_global(
             else:
                 slice_mat = slice_df.drop("compname", axis=1)
 
-            # Compute the Gower distance on the prepared slice matrix.
-            D = gower_distances(slice_mat)
+            # Compute the Gower distance on the prepared slice matrix. A slice with
+            # zero usable feature columns is handled inside the helper (see its
+            # docstring) so the per-depth mean imputer can't crash on a 0-feature
+            # array; it returns an all-NaN matrix that the steps below then ignore.
+            D = _slice_gower_distance(slice_mat)
 
             dis_mat_list.append(D)
 
@@ -840,8 +864,10 @@ def rank_soils_global(
             "Not Ranked" if np.ma.is_masked(x) else "Ranked" for x in D_check[0][1:]
         ]
 
-        # Calculate max dissimilarity per depth slice
-        dis_max = max(map(np.nanmax, dis_mat_list))
+        # Calculate max dissimilarity across all depth slices. Use a single
+        # NaN-aware reduction over the stack so an all-NaN slice (a depth with no
+        # usable data) can't make the result NaN via max()'s ordering.
+        dis_max = np.nanmax(dis_mat_list)
 
         # Apply depth weight
         depth_weight = np.concatenate([np.repeat(0.2, 20), np.repeat(1.0, 180)])
diff --git a/soil_id/tests/test_global_rank_empty_slice.py b/soil_id/tests/test_global_rank_empty_slice.py
new file mode 100644
index 0000000..eeb654d
--- /dev/null
+++ b/soil_id/tests/test_global_rank_empty_slice.py
@@ -0,0 +1,64 @@
+# Copyright © 2026 Technology Matters
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published
+# by the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU Affero General Public License for more details.
+#
+# You should have received a copy of the GNU Affero General Public License
+# along with this program. If not, see https://www.gnu.org/licenses/.
+
+"""Regression tests for the empty-feature depth-slice guard in global rank.
+
+rank_soils_global builds a per-depth-slice feature matrix and feeds it to
+gower_distances. When a depth slice has zero usable feature columns (e.g. a gap in
+the user's horizons, or horizon data that doesn't reach the slice depth), the raw
+gower path passes a shape ``(n, 0)`` array into SimpleImputer and raises
+"Found array with 0 feature(s)…", failing the whole ranking. ``_slice_gower_distance``
+guards that case. These tests cover the guard directly so they need no database or
+network — unlike the live ``test_global_integration`` path.
+"""
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from soil_id.global_soil import _slice_gower_distance
+from soil_id.utils import gower_distances
+
+
+def test_zero_feature_slice_returns_all_nan_matrix():
+    # 11 components, but no feature columns — the exact shape from the Sentry crash.
+    slice_mat = pd.DataFrame(index=range(11))
+    assert slice_mat.shape == (11, 0)
+
+    result = _slice_gower_distance(slice_mat)
+
+    # An (n, n) all-NaN matrix: same shape gower would have returned, and the
+    # downstream masked-average/NaN-infill steps treat it as "no information".
+    assert result.shape == (11, 11)
+    assert np.isnan(result).all()
+
+
+def test_zero_feature_slice_would_otherwise_crash_gower():
+    # Characterization: the raw gower path crashes on a 0-feature array. This is
+    # exactly what the guard above prevents; if gower ever stops raising here, the
+    # guard's rationale should be revisited.
+    slice_mat = pd.DataFrame(index=range(11))
+    with pytest.raises(ValueError):
+        gower_distances(slice_mat)
+
+
+def test_nonempty_slice_passes_through_to_gower():
+    # With at least one feature column, the helper must be a transparent passthrough.
+    slice_mat = pd.DataFrame({"sand": [10.0, 20.0, 30.0], "clay": [5.0, 15.0, 25.0]})
+
+    result = _slice_gower_distance(slice_mat)
+    expected = gower_distances(slice_mat)
+
+    assert np.array_equal(result, expected, equal_nan=True)