@@ -619,8 +619,10 @@ async def list_efficiency_metrics(
     """List cost efficiency metrics from Cost Optimization Hub.
 
     Returns time-series efficiency data (efficiency score, estimated savings, and
-    spend) aggregated over the requested time period and optionally grouped by
-    account ID or Region.
+    spend), optionally grouped by account ID or Region. Each point is a one-day
+    snapshot (estimated savings of the recommendations open that day, against the
+    trailing 30 days of spend); a Monthly point is the snapshot for the last day
+    of the month that has data, with the trailing 30 days of spend at month-end.
 
     Fetches every page, or stops before exceeding ``max_items``, and returns the combined
     groups with a ``pagination`` block. ``orderBy`` is sent on every page. In ``performance``
 
@@ -397,8 +397,10 @@ def _validate_order_by(
 3. get_recommendation: Get detailed information about a specific recommendation
 4. list_efficiency_metrics: Cost efficiency score, estimated savings, and spend as
    a time series (optionally grouped by AccountId or Region). Use for questions
-   about the cost-efficiency score, its trend over time, month-over-month change,
-   cross-region/cross-account comparison, and top/worst performers.
+   about the cost-efficiency score, how it changed over time (use 'Daily' for
+   that), cross-region/cross-account comparison, and top/worst performers. Each
+   point is a one-day snapshot, not a period total: the estimated savings of the
+   recommendations open on that day, against the trailing 30 days of spend.
 5. list_enrollment_statuses: Cost Optimization Hub enrollment status (Active/Inactive)
    of the calling account. Optional 'account_id' returns the status of that account instead.
    An account that has never enrolled has no enrollment record; it is reported as
@@ -431,7 +433,15 @@ def _validate_order_by(
 - Service only available in us-east-1 region
 
 list_efficiency_metrics parameters:
-- 'granularity': 'Daily' or 'Monthly' (defaults to 'Monthly' when omitted)
+- 'granularity': 'Daily' or 'Monthly' (defaults to 'Monthly' when omitted).
+  A Monthly point ('YYYY-MM') is not a monthly average or total: it is the
+  snapshot for the last day of the month that has data, with the trailing 30
+  days of spend at month-end, so a change on that one day moves the month's
+  value. Ask only for completed months (end_date no later than the current
+  month, since end is exclusive); a current-month point would be partial. For a
+  trend or a change over time, use 'Daily' and read the daily series rather than
+  comparing Monthly values. When reporting a Monthly value, say it is the
+  month-end snapshot.
 - 'start_date' / 'end_date': window bounds as 'YYYY-MM-DD' (daily) or 'YYYY-MM'
   (monthly); end is exclusive. Both default to a ~90-day lookback when omitted.
   The API only serves data whose start is within the most recent 90 days (Daily)
@@ -519,7 +529,9 @@ async def cost_optimization_hub(
             Applies to list_recommendations, list_recommendation_summaries, and
             list_efficiency_metrics.
         granularity: Time granularity for list_efficiency_metrics ('Daily' or
-            'Monthly'). Defaults to 'Monthly' when omitted.
+            'Monthly'). Defaults to 'Monthly' when omitted. A Monthly point is the
+            month-end snapshot (the last day of the month that has data), not a
+            monthly average; use 'Daily' for trends.
         start_date: Inclusive window start for list_efficiency_metrics
             ('YYYY-MM-DD' or 'YYYY-MM'). Defaults to a ~90-day lookback when omitted.
         end_date: Exclusive window end for list_efficiency_metrics ('YYYY-MM-DD'
 
@@ -1858,3 +1858,41 @@ async def test_coh_summaries_group_by_tag_key_without_key(mock_context):
 
     assert res['error_type'] == 'validation_error'
     mock_summaries.assert_not_awaited()
+
+
+async def _coh_served_description() -> str:
+    """Return the cost-optimization description as served (reloaded with the real decorator).
+
+    Other tests reload the module with an identity decorator, which leaves a server
+    with no tools, so reload it here with the real one.
+    """
+    from awslabs.billing_cost_management_mcp_server.tools import (
+        cost_optimization_hub_tools as coh_mod,
+    )
+
+    coh_mod = importlib.reload(coh_mod)
+    tool = await coh_mod.cost_optimization_hub_server.get_tool('cost-optimization')
+    assert tool is not None
+    return ' '.join((tool.description or '').split())
+
+
+@pytest.mark.asyncio
+async def test_description_does_not_route_trends_to_monthly_efficiency():
+    """A Monthly efficiency point is one day's snapshot, so trend questions go to Daily."""
+    flat = await _coh_served_description()
+    assert 'month-over-month change' not in flat
+    assert 'its trend over time' not in flat
+    assert 'not a monthly average or total' in flat
+    assert 'last day of the month that has data' in flat
+    assert "For a trend or a change over time, use 'Daily'" in flat
+
+
+@pytest.mark.asyncio
+async def test_description_says_what_an_efficiency_point_covers():
+    """Each point is a snapshot over trailing-30-day spend; ask Monthly only for completed months."""
+    flat = await _coh_served_description()
+    assert 'one-day snapshot, not a period total' in flat
+    assert 'the trailing 30 days of spend' in flat
+    assert 'Ask only for completed months' in flat
+    assert 'a current-month point would be partial' in flat
+    assert 'Only completed months are returned' not in flat