diff --git a/src/pyrecest/evaluation/selection.py b/src/pyrecest/evaluation/selection.py
index db174f778..f6d7124cd 100644
--- a/src/pyrecest/evaluation/selection.py
+++ b/src/pyrecest/evaluation/selection.py
@@ -22,9 +22,13 @@ def _is_text_scalar(value) -> bool:
 
 
 def _normalize_nonnegative_integer(value, name: str) -> int:
-    value_array = np.asarray(value)
+    message = f"{name} must be a non-negative integer."
+    try:
+        value_array = np.asarray(value)
+    except (TypeError, ValueError, RuntimeError) as exc:
+        raise ValueError(message) from exc
     if value_array.shape != () or value_array.dtype == np.bool_:
-        raise ValueError(f"{name} must be a non-negative integer.")
+        raise ValueError(message)
 
     scalar = value_array.item()
     if (
@@ -32,35 +36,42 @@ def _normalize_nonnegative_integer(value, name: str) -> int:
         or _is_text_scalar(scalar)
         or not isinstance(scalar, Real)
     ):
-        raise ValueError(f"{name} must be a non-negative integer.")
+        raise ValueError(message)
     if isinstance(scalar, (int, np.integer)):
         integer = int(scalar)
     else:
         try:
             scalar_float = float(scalar)
         except (TypeError, ValueError, OverflowError) as exc:
-            raise ValueError(f"{name} must be a non-negative integer.") from exc
+            raise ValueError(message) from exc
         if not np.isfinite(scalar_float) or not scalar_float.is_integer():
-            raise ValueError(f"{name} must be a non-negative integer.")
+            raise ValueError(message)
         integer = int(scalar_float)
 
     if integer < 0:
-        raise ValueError(f"{name} must be a non-negative integer.")
+        raise ValueError(message)
     return integer
 
 
 def _normalize_bool_flag(value, name: str) -> bool:
-    value_array = np.asarray(value)
+    message = f"{name} must be a boolean."
+    try:
+        value_array = np.asarray(value)
+    except (TypeError, ValueError, RuntimeError) as exc:
+        raise ValueError(message) from exc
     if value_array.shape != ():
-        raise ValueError(f"{name} must be a boolean.")
+        raise ValueError(message)
     scalar = value_array.item()
     if not isinstance(scalar, (bool, np.bool_)):
-        raise ValueError(f"{name} must be a boolean.")
+        raise ValueError(message)
     return bool(scalar)
 
 
 def _normalize_finite_scalar(value, message: str) -> float:
-    value_array = np.asarray(value)
+    try:
+        value_array = np.asarray(value)
+    except (TypeError, ValueError, RuntimeError) as exc:
+        raise ValueError(message) from exc
     if value_array.shape != () or value_array.dtype == np.bool_:
         raise ValueError(message)
 
diff --git a/tests/evaluation/test_selection.py b/tests/evaluation/test_selection.py
index 15faf8c2b..9d1276fa5 100644
--- a/tests/evaluation/test_selection.py
+++ b/tests/evaluation/test_selection.py
@@ -15,6 +15,12 @@
 )
 
 
+class UncoercibleScalar:
+    def __array__(self, dtype=None):
+        del dtype
+        raise TypeError("cannot convert")
+
+
 def test_top_count_mask_is_deterministic_with_ties() -> None:
     mask = top_count_mask([1.0, 2.0, 2.0, 0.5], 2)
 
@@ -101,6 +107,23 @@ def test_selection_helpers_reject_text_scalar_fractions() -> None:
             tail_rescue_quota_count(3, rescue_fraction=fraction)
 
 
+def test_selection_helpers_report_value_error_for_uncoercible_scalars() -> None:
+    uncoercible = UncoercibleScalar()
+
+    with pytest.raises(ValueError, match="item_count"):
+        retained_count_from_fraction(uncoercible, 0.5)
+    with pytest.raises(ValueError, match="retention_fraction"):
+        retained_count_from_fraction(10, uncoercible)
+    with pytest.raises(ValueError, match="nonnegative"):
+        sanitized_score_vector([1.0], nonnegative=uncoercible)
+    with pytest.raises(ValueError, match="largest"):
+        top_count_mask([1.0], 1, largest=uncoercible)
+    with pytest.raises(ValueError, match="quantile"):
+        quantile_tail_threshold([0.0, 1.0], uncoercible)
+    with pytest.raises(ValueError, match="rescue_fraction"):
+        tail_rescue_quota_count(3, rescue_fraction=uncoercible)
+
+
 def test_quantile_tail_mask_selects_lower_tail() -> None:
     values = np.asarray([0.0, 1.0, 2.0, 3.0])