@@ -242,6 +242,11 @@ def _starts_before_default_history(today: date, *starts: str) -> bool:
 - RECORD_TYPE: Record type (e.g., Usage, Credit, Tax)
 - LEGAL_ENTITY_NAME: AWS seller of record
 
+Costs with no value for a grouped dimension come back under a placeholder such as NoRegion or
+NoTenancy, including in each comparison's cost_and_usage_selector. The placeholder is not a
+value: a filter on it matches nothing. To filter on those costs, pass the empty string:
+`{"Dimensions": {"Key": "REGION", "Values": [""]}}`.
+
 Note:
 - Time periods must start and end on the first day of a month, with a duration of exactly one month
 - The getCostComparisonDrivers operation automatically includes SERVICE and USAGE_TYPE dimensions
 
@@ -95,6 +95,12 @@ def _with_example_dates(description: str) -> str:
    with usage in the period, so absence is not proof the value is invalid. If you also need the
    spend, re-run grouped by that dimension to get the value and its cost in one call.
 2. If you GROUPED and the line genuinely shows $0, report "$0.00", not "no data found".
+3. A grouped result shows costs with no value for the dimension under a placeholder key such
+   as NoRegion or NoTenancy. The placeholder is not a value: a filter on it matches nothing
+   and returns $0, and getDimensionValues lists the empty value as "". To filter on those
+   costs, pass the empty string: `{"Dimensions": {"Key": "REGION", "Values": [""]}}`. For
+   costs without a tag or cost category key, use `ABSENT` instead (see FILTER EXPRESSION
+   GRAMMAR).
 
 PAGINATION (max_items, next_token: getCostAndUsage, getCostAndUsageWithResources, getDimensionValues, getTagsOrValues, getCostCategories only; the other operations take neither):
 - max_items is the total number of items returned (getCostAndUsage: one item per group per period, or per period without group_by). Omit it to fetch every page (no limit). Pass it, or narrow the time range, filter or group_by, for queries that can be large (e.g. daily data grouped by RESOURCE_ID, USAGE_TYPE or LINKED_ACCOUNT).
@@ -145,7 +151,9 @@ def _with_example_dates(description: str) -> str:
 to these filters.
 
 Values per leaf type:
-- Dimensions: `Values` is required and must not be empty.
+- Dimensions: `Values` is required and must not be empty. `[""]` (one empty string) is not
+  empty: in getCostAndUsage it matches costs with no value for the dimension (see
+  REPORTING $0).
 - Tags / CostCategories: `Values` is required, except with `ABSENT`. `ABSENT` matches costs
   where the tag or cost category key is not present, so there is no value to compare against:
   omit `Values` entirely (an empty `Values` list is rejected too).
 
@@ -1757,6 +1757,40 @@ def test_dimension_examples_use_exact_ce_values(
             # A wrong value only counts as a whole list item, so 'Aurora MySQL' is not 'Aurora'.
             assert not re.search(rf'[ (]{re.escape(value)}(?=[,);.]|$)', entry), value
 
+    @pytest.mark.parametrize(
+        'module_name, tool_name, server_attr',
+        [
+            ('cost_explorer_tools', 'cost-explorer', 'cost_explorer_server'),
+            ('cost_comparison_tools', 'cost-comparison', 'cost_comparison_server'),
+        ],
+    )
+    def test_empty_value_placeholder_is_filtered_as_empty_string(
+        self, module_name, tool_name, server_attr
+    ):
+        """Grouped results label an empty value NoRegion etc.; a filter must use "" instead.
+
+        Cost Explorer reports a cost with no value for a grouped dimension under a
+        placeholder key (NoRegion, NoTenancy, ...), but its filters match the stored empty
+        value, so a filter on the placeholder silently returns $0.
+        """
+        module = importlib.reload(
+            importlib.import_module(
+                f'awslabs.billing_cost_management_mcp_server.tools.{module_name}'
+            )
+        )
+        description = _registered_tool(getattr(module, server_attr), tool_name).description or ''
+        flat = ' '.join(description.split())
+        assert 'NoRegion' in flat
+        assert 'NoTenancy' in flat
+        assert 'a filter on it matches nothing' in flat
+        examples = [json.loads(m) for m in re.findall(r'`(\{"Dimensions": [^`]*\})`', description)]
+        assert {'Dimensions': {'Key': 'REGION', 'Values': ['']}} in examples
+
+    def test_filter_grammar_allows_one_empty_string_value(self, tool):
+        """The grammar's "Values must not be empty" rule does not forbid the [""] filter."""
+        grammar = tool.description.split('Values per leaf type:')[1].split('\n- Tags')[0]
+        assert '`[""]` (one empty string) is not' in grammar
+
     def test_description_prefers_monthly_for_whole_month_rankings(self, tool):
         """Whole-month totals and rankings are steered to MONTHLY; the code default stays DAILY."""
         assert 'DAILY granularity for periods <3 months' not in tool.description