@@ -237,7 +237,7 @@ def _starts_before_default_history(today: date, *starts: str) -> bool:
 - REGION: AWS Region
 - USAGE_TYPE: Type of usage (e.g., BoxUsage:t2.micro)
 - INSTANCE_TYPE: EC2 instance type (e.g., t2.micro, m5.large)
-- PLATFORM: Operating system (e.g., Windows, Linux)
+- PLATFORM: Operating system (e.g., Linux/UNIX, Windows (Amazon VPC))
 - TENANCY: Instance tenancy. Exact, case-sensitive values: Shared, Dedicated, Host
 - RECORD_TYPE: Record type (e.g., Usage, Credit, Tax)
 - LEGAL_ENTITY_NAME: AWS seller of record
 
@@ -219,18 +219,20 @@ def _with_example_dates(description: str) -> str:
 - AZ: The Availability Zone (e.g., us-east-1a)
 - BILLING_ENTITY: Who bills the charge - "AWS" for AWS services, "AWS Marketplace" for
   third-party Marketplace purchases. This is the key for Marketplace spend, NOT LEGAL_ENTITY_NAME
-- DATABASE_ENGINE: The Amazon RDS database (e.g., Aurora, MySQL)
-- DEPLOYMENT_OPTION: RDS deployment scope (SingleAZ, MultiAZ)
+- DATABASE_ENGINE: The database engine (e.g., Aurora MySQL, Aurora PostgreSQL, MySQL, PostgreSQL)
+- DEPLOYMENT_OPTION: Deployment option (e.g., Single-AZ, Multi-AZ)
 - INSTANCE_TYPE: The EC2 instance type (e.g., m4.xlarge)
-- INSTANCE_TYPE_FAMILY: Family of instances (e.g., Compute Optimized, Memory Optimized)
+- INSTANCE_TYPE_FAMILY: Instance family (e.g., m5, t3, c6g); the case can vary by service (e.g., r5
+  and R5 are separate values)
 - INVOICING_ENTITY: The AWS entity that issued the invoice (e.g., Amazon Web Services, Inc.)
 - LEGAL_ENTITY_NAME: The AWS seller of record (e.g., Amazon Web Services, Inc.); it does not
   separate Marketplace from AWS charges
 - LINKED_ACCOUNT: AWS member accounts
-- OPERATING_SYSTEM: OS type (e.g., Windows, Linux)
+- OPERATING_SYSTEM: OS type (e.g., Linux/UNIX, Windows (Amazon VPC))
 - OPERATION: The API operation behind a charge (e.g., RunInstances, PutObject; for Amazon Bedrock, model-invocation operations). The dimension is OPERATION — there is no API_OPERATION dimension.
-- PLATFORM: EC2 operating system
-- PURCHASE_TYPE: Reservation type (e.g., On-Demand, Reserved)
+- PLATFORM: EC2 operating system (e.g., Linux/UNIX, Windows (Amazon VPC))
+- PURCHASE_TYPE: Purchase option (e.g., On Demand Instances, Standard Reserved Instances, Savings
+  Plans)
 - REGION: AWS Region
 - RESOURCE_ID: Resource identifier. Only for getCostAndUsageWithResources (last 14 days)
 - SERVICE: AWS service (e.g., Amazon DynamoDB)
@@ -239,7 +241,18 @@ def _with_example_dates(description: str) -> str:
   Host (Dedicated Host)
 - USAGE_TYPE: Usage type (e.g., DataTransfer-In-Bytes)
 - USAGE_TYPE_GROUP: Group of related usage types (e.g., EC2: Running Hours)
-- RECORD_TYPE: Charge types (e.g., RI fees, usage costs)
+- RECORD_TYPE: Charge type. Values are case-sensitive: Usage, DiscountedUsage (RI-covered
+  usage), SavingsPlanCoveredUsage, SavingsPlanNegation, SavingsPlanRecurringFee,
+  SavingsPlanUpfrontFee, Recurring (RI recurring fee), Upfront (RI upfront fee), Support, Other
+  (out-of-cycle and non-RI subscription charges), Tax, Credit, Refund, and these discounts by full
+  name: Bundled Discount, Distributor Discount, Enterprise Discount Program Discount, Marketplace
+  Discount, Private Rate Card Discount, RI Volume Discount, Solution Provider Program Discount.
+  These differ from the CUR and Cost Categories line-item types: Cost Explorer reports RIFee as
+  Recurring, so a RIFee filter matches nothing, and reports most CUR Fee rows as Upfront, Other or
+  Support, so a Fee filter misses them (use Upfront for RI upfront fees, Other for other
+  subscription fees). Any other CUR line-item type, such as a Fee with no subscription that is not
+  a support charge, passes through unchanged. Call getDimensionValues with RECORD_TYPE for the
+  values present in the account.
 
 BILLING VIEW SUPPORT:
 Most Cost Explorer operations support an optional billing_view_arn parameter that allows you to scope
 
@@ -1635,6 +1635,128 @@ def test_tenancy_reference_uses_exact_ce_values(self, module_name, tool_name, se
         for wrong in ('shared', 'dedicated', 'default', 'SHARED', 'DEDICATED', 'HOST'):
             assert not re.search(rf'\b{wrong}\b', tenancy)
 
+    def test_record_type_reference_uses_exact_ce_values(self, tool):
+        """RECORD_TYPE lists the values Cost Explorer matches, not CUR codes or prose."""
+        match = re.search(r'^- RECORD_TYPE:.*(?:\n  .*)*', tool.description, re.MULTILINE)
+        assert match is not None
+        # Re-join wrapped lines so multi-word values split across a line break still match.
+        record_type = ' '.join(match.group(0).split())
+        assert 'case-sensitive' in record_type
+        for value in (
+            'Usage',
+            'DiscountedUsage',
+            'SavingsPlanCoveredUsage',
+            'SavingsPlanNegation',
+            'SavingsPlanRecurringFee',
+            'SavingsPlanUpfrontFee',
+            'Recurring',
+            'Upfront',
+            'Support',
+            'Other',
+            'Tax',
+            'Credit',
+            'Refund',
+            'Bundled Discount',
+            'Enterprise Discount Program Discount',
+            'Private Rate Card Discount',
+            'RI Volume Discount',
+            'Solution Provider Program Discount',
+            'Distributor Discount',
+            'Marketplace Discount',
+        ):
+            assert re.search(rf'\b{value}\b', record_type), value
+        # The old prose guesses are gone, and the CUR codes are named only as non-values.
+        assert 'e.g., RI fees, usage costs' not in record_type
+        assert 'reports RIFee as Recurring, so a RIFee filter matches nothing' in record_type
+        assert 'most CUR Fee rows as Upfront, Other or Support' in record_type
+        assert 'use Upfront for RI upfront fees, Other for other subscription fees' in record_type
+        assert 'RIFee and Fee match nothing' not in record_type
+        # The list is not exhaustive: any unmapped CUR line-item type passes through unchanged.
+        assert 'Exact' not in record_type
+        assert 'Any other CUR line-item type' in record_type
+        assert 'passes through unchanged' in record_type
+        assert 'getDimensionValues with RECORD_TYPE' in record_type
+
+    @pytest.mark.parametrize(
+        'module_name, tool_name, server_attr, key, values, wrong',
+        [
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'DATABASE_ENGINE',
+                ('Aurora MySQL', 'Aurora PostgreSQL', 'MySQL', 'PostgreSQL'),
+                ('Aurora',),
+            ),
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'DEPLOYMENT_OPTION',
+                ('Single-AZ', 'Multi-AZ'),
+                ('SingleAZ', 'MultiAZ'),
+            ),
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'INSTANCE_TYPE_FAMILY',
+                ('m5', 't3', 'c6g', 'r5', 'R5'),
+                ('Compute Optimized', 'Memory Optimized'),
+            ),
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'OPERATING_SYSTEM',
+                ('Linux/UNIX', 'Windows (Amazon VPC)'),
+                ('Windows', 'Linux'),
+            ),
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'PLATFORM',
+                ('Linux/UNIX', 'Windows (Amazon VPC)'),
+                ('Windows', 'Linux'),
+            ),
+            (
+                'cost_explorer_tools',
+                'cost-explorer',
+                'cost_explorer_server',
+                'PURCHASE_TYPE',
+                ('On Demand Instances', 'Standard Reserved Instances', 'Savings Plans'),
+                ('On-Demand',),
+            ),
+            (
+                'cost_comparison_tools',
+                'cost-comparison',
+                'cost_comparison_server',
+                'PLATFORM',
+                ('Linux/UNIX', 'Windows (Amazon VPC)'),
+                ('Windows', 'Linux'),
+            ),
+        ],
+    )
+    def test_dimension_examples_use_exact_ce_values(
+        self, module_name, tool_name, server_attr, key, values, wrong
+    ):
+        """Dimension examples are values Cost Explorer returns, not console or pricing labels."""
+        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 ''
+        match = re.search(rf'^- {key}:.*(?:\n  .*)*', description, re.MULTILINE)
+        assert match is not None
+        entry = ' '.join(match.group(0).split())
+        for value in values:
+            assert re.search(rf'(?<![\w-]){re.escape(value)}(?![\w-])', entry), value
+        for value in wrong:
+            # 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
+
     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