diff --git a/ami/main/api/serializers.py b/ami/main/api/serializers.py
index 5e46a790d..acefa7785 100644
--- a/ami/main/api/serializers.py
+++ b/ami/main/api/serializers.py
@@ -1942,14 +1942,14 @@ class ModelAgreementSerializer(serializers.Serializer):
         max_value=1.0,
         allow_null=True,
         required=False,
-        help_text="Wilson 95% CI lower bound for agreed_exact_pct. Null when verified_with_prediction_count is 0.",
+        help_text="Wilson 95% CI lower bound for agreed_exact_pct. Null when comparable_count is 0.",
     )
     agreed_exact_ci_high = serializers.FloatField(
         min_value=0.0,
         max_value=1.0,
         allow_null=True,
         required=False,
-        help_text="Wilson 95% CI upper bound for agreed_exact_pct. Null when verified_with_prediction_count is 0.",
+        help_text="Wilson 95% CI upper bound for agreed_exact_pct. Null when comparable_count is 0.",
     )
     agreed_any_rank_count = serializers.IntegerField(
         help_text="Exact matches plus disagreements whose LCA is at any real rank (UNKNOWN excluded)."
@@ -1957,21 +1957,21 @@ class ModelAgreementSerializer(serializers.Serializer):
     agreed_any_rank_pct = serializers.FloatField(
         min_value=0.0,
         max_value=1.0,
-        help_text="agreed_any_rank_count / verified_with_prediction_count",
+        help_text="agreed_any_rank_count / comparable_count",
     )
     agreed_any_rank_ci_low = serializers.FloatField(
         min_value=0.0,
         max_value=1.0,
         allow_null=True,
         required=False,
-        help_text="Wilson 95% CI lower bound for agreed_any_rank_pct. Null when verified_with_prediction_count is 0.",
+        help_text="Wilson 95% CI lower bound for agreed_any_rank_pct. Null when comparable_count is 0.",
     )
     agreed_any_rank_ci_high = serializers.FloatField(
         min_value=0.0,
         max_value=1.0,
         allow_null=True,
         required=False,
-        help_text="Wilson 95% CI upper bound for agreed_any_rank_pct. Null when verified_with_prediction_count is 0.",
+        help_text="Wilson 95% CI upper bound for agreed_any_rank_pct. Null when comparable_count is 0.",
     )
     cohens_kappa = serializers.FloatField(
         min_value=-1.0,
@@ -2002,5 +2002,5 @@ class ModelAgreementSerializer(serializers.Serializer):
         max_value=1.0,
         allow_null=True,
         required=False,
-        help_text="agreed_coarser_rank_count / verified_with_prediction_count. Null when no threshold supplied.",
+        help_text="agreed_coarser_rank_count / comparable_count. Null when no threshold supplied.",
     )
diff --git a/ami/main/tests.py b/ami/main/tests.py
index 0a2ce6b7f..0b89ff686 100644
--- a/ami/main/tests.py
+++ b/ami/main/tests.py
@@ -6020,6 +6020,28 @@ def test_taxon_less_verification_excluded_from_denominator(self):
         # 1 exact agreement out of 1 comparable → 100%, not 50%.
         self.assertEqual(result["agreed_exact_count"], 1)
         self.assertAlmostEqual(result["agreed_exact_pct"], 1.0)
+        self.assertAlmostEqual(result["agreed_any_rank_pct"], 1.0)
+
+    def test_confidence_intervals_null_when_nothing_is_comparable(self):
+        """Wilson bounds are null once nothing is comparable, even though every
+        verified occurrence here carries a machine prediction.
+
+        comparable_count falls below verified_with_prediction_count whenever a
+        verification has no taxon, and the intervals key off the former.
+        """
+        from ami.main.models_future.occurrence import model_agreement_for_project
+
+        occurrences = list(Occurrence.objects.filter(project=self.project).order_by("pk"))
+        for occurrence in occurrences[:2]:
+            Identification.objects.create(user=self.user, occurrence=occurrence, taxon=None)
+
+        result = model_agreement_for_project(Occurrence.objects.filter(project=self.project))
+        self.assertEqual(result["verified_with_prediction_count"], 2)
+        self.assertEqual(result["comparable_count"], 0)
+        self.assertIsNone(result["agreed_exact_ci_low"])
+        self.assertIsNone(result["agreed_exact_ci_high"])
+        self.assertIsNone(result["agreed_any_rank_ci_low"])
+        self.assertIsNone(result["agreed_any_rank_ci_high"])
 
 
 class TestOccurrenceStatsViewSet(APITestCase):