diff --git a/changelog.d/954.changed.md b/changelog.d/954.changed.md
new file mode 100644
index 000000000..ad5a9a9fd
--- /dev/null
+++ b/changelog.d/954.changed.md
@@ -0,0 +1 @@
+Made only US state runs compute congressional district impacts; national runs (no region, or `us`) and region-group runs, and so segmented national runs, their reduce and Stage 12 aggregation of a national report, now compute none: `congressional_district_impact` is null in their output, and the gateway's job-status response leaves that null field out (Stage 12 aggregation of a state report spelled `state/...` still computes that state's districts), because nationwide reports take each district from a separate run of its state on that state's dataset, which matches the state's own report; the beta integration check of a US national result now expects no `congressional_district_impact`, whether absent or null, and the state checks still expect the state's districts.
diff --git a/projects/policyengine-apis-integ/tests/simulation/test_calculate.py b/projects/policyengine-apis-integ/tests/simulation/test_calculate.py
index 105d7bcb6..d61607a3b 100644
--- a/projects/policyengine-apis-integ/tests/simulation/test_calculate.py
+++ b/projects/policyengine-apis-integ/tests/simulation/test_calculate.py
@@ -34,7 +34,6 @@
     "no_change_percentage",
     "population",
 }
-_AT_LARGE_DISTRICT_IDS = {"AK-01", "DC-01", "DE-01", "ND-01", "SD-01", "VT-01", "WY-01"}
 # From the generated client, so a stage added to the schema is accepted here.
 _JOB_STAGES = frozenset(stage.value for stage in JobStage)
 
@@ -238,14 +237,30 @@ def assert_congressional_district_results(
         )
 
     if expected_district_ids is not None:
-        district_ids = set(_district_ids(district_results))
-        missing = expected_district_ids - district_ids
-        assert not missing, (
-            f"Missing expected district IDs {sorted(missing)} "
-            f"from result IDs: {sorted(district_ids)}"
+        district_ids = _district_ids(district_results)
+        # Sorted lists, not sets, so a district listed twice fails too.
+        assert sorted(district_ids) == sorted(expected_district_ids), (
+            f"Expected district IDs {sorted(expected_district_ids)}, "
+            f"got {sorted(district_ids)}"
         )
 
 
+def assert_no_congressional_district_results(economy_result: object) -> None:
+    """Check a US national result has no congressional_district_impact.
+
+    Only state runs compute districts: nationwide reports take them from a
+    separate run of each state. The field is null in the run's output, and the
+    gateway's job-status response leaves null fields out, so a deployed result
+    has no such field.
+    """
+    economy_result = _as_mapping(economy_result, label="economy result")
+    district_impact = economy_result.get("congressional_district_impact")
+    assert district_impact is None, (
+        "Expected no congressional_district_impact in a US national result, "
+        f"got {district_impact!r}"
+    )
+
+
 _UK_AREA_RESULT_KEYS = ("constituency_impact", "local_authority_impact")
 _PUBLISHED_DISCLOSURE_TIERS = {"reported", "approximate"}
 
@@ -323,6 +338,74 @@ def test_result_assertions_accept_generated_client_objects() -> None:
     assert_congressional_district_results(result, expected_district_prefix="UT-")
 
 
+def _state_result_with_districts(*district_ids: str) -> dict:
+    return {
+        "congressional_district_impact": {
+            "districts": [
+                {
+                    "district": district_id,
+                    "average_household_income_change": 0.0,
+                    "relative_household_income_change": 0.0,
+                    "winner_percentage": 0.0,
+                    "loser_percentage": 0.0,
+                    "no_change_percentage": 1.0,
+                    "population": 1.0,
+                }
+                for district_id in district_ids
+            ]
+        }
+    }
+
+
+def test_state_assertions_accept_an_at_large_state_with_its_one_district() -> None:
+    assert_congressional_district_results(
+        _state_result_with_districts("VT-01"),
+        expected_district_prefix="VT-",
+        expected_district_ids={"VT-01"},
+    )
+
+
+@pytest.mark.parametrize(
+    "district_ids",
+    [("VT-00",), ("VT-01", "VT-02"), ("VT-01", "VT-01")],
+    ids=["another-id", "an-extra-district", "a-repeated-district"],
+)
+def test_state_assertions_reject_other_district_ids(district_ids) -> None:
+    with pytest.raises(AssertionError, match="district IDs"):
+        assert_congressional_district_results(
+            _state_result_with_districts(*district_ids),
+            expected_district_prefix="VT-",
+            expected_district_ids={"VT-01"},
+        )
+
+
+def test_national_us_assertions_accept_a_result_without_districts() -> None:
+    assert_no_congressional_district_results(
+        _GeneratedClientObject({"budget": {}, "congressional_district_impact": None})
+    )
+    assert_no_congressional_district_results({"congressional_district_impact": None})
+    # The gateway's job-status response leaves null fields out, so a deployed
+    # national result has no district field at all.
+    assert_no_congressional_district_results({"budget": {}})
+    assert_no_congressional_district_results(_GeneratedClientObject({"budget": {}}))
+
+
+@pytest.mark.parametrize(
+    "district_impact",
+    [
+        {"districts": []},
+        {"districts": [{"district": "UT-01"}]},
+        _GeneratedClientObject({"districts": [{"district": "UT-01"}]}),
+    ],
+    ids=["no-districts-listed", "districts-listed", "generated-client-object"],
+)
+def test_national_us_assertions_reject_district_results(district_impact) -> None:
+    with pytest.raises(AssertionError, match="congressional_district_impact"):
+        assert_no_congressional_district_results(
+            {"congressional_district_impact": district_impact}
+        )
+
+
 _BREAKDOWN_CONSTITUENCY = {
     "constituency_code": "E14001063",
     "constituency_name": "Aldershot",
@@ -613,9 +696,10 @@ def test_calculate_default_model(
 ):
     """A simulation on the default model version returns economic impact data.
 
-    Given a simulation request with default model version
+    Given a US national simulation request with default model version
     When the simulation is submitted and polled to completion
-    Then the result contains expected economic impact data.
+    Then the result contains expected economic impact data and no
+    congressional districts, which only state runs compute.
     """
     # Given
     request = SimulationRequest.from_dict(
@@ -649,10 +733,7 @@ def test_calculate_default_model(
 
     economy_result = result.result
     assert_economy_result_sections(economy_result)
-    assert_congressional_district_results(
-        economy_result,
-        expected_district_ids=_AT_LARGE_DISTRICT_IDS,
-    )
+    assert_no_congressional_district_results(economy_result)
 
 
 @pytest.mark.beta_only
@@ -719,16 +800,18 @@ def test_calculate_specific_model(
 ):
     """A simulation pinned to a specific US model version returns impact data.
 
-    Given a simulation request with a specific model version
+    Given a simulation request with a specific model version, for Vermont
     When the simulation is submitted and polled to completion
-    Then the result contains expected economic impact data.
+    Then the result contains expected economic impact data, and Vermont's one
+    at-large district: nationwide reports take each at-large district from its
+    state's own run, as here.
     """
     # Given
     request = SimulationRequest.from_dict(
         {
             "country": "us",
             "version": us_model_version,
-            "region": "state/ut",
+            "region": "state/vt",
             "scope": "macro",
             "reform": {
                 "gov.irs.credits.ctc.refundable.fully_refundable": {
@@ -763,7 +846,8 @@ def test_calculate_specific_model(
     assert_economy_result_sections(economy_result)
     assert_congressional_district_results(
         economy_result,
-        expected_district_prefix="UT-",
+        expected_district_prefix="VT-",
+        expected_district_ids={"VT-01"},
     )
 
 
diff --git a/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/segmented_national_reduce.py b/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/segmented_national_reduce.py
index 26ba85cac..4d059e96f 100644
--- a/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/segmented_national_reduce.py
+++ b/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/segmented_national_reduce.py
@@ -142,8 +142,9 @@ def build_national_output(
     """Reduce the children's microdata into the full national macro output dict.
 
     Runs the existing ``SimulationOutputBuilder`` over stand-in simulations
-    with ``resolved_region_code=country`` so the national congressional
-    district table populates. No model recompute occurs.
+    with ``resolved_region_code=country``, as for any national run, so the
+    output has no congressional districts: nationwide reports take those from
+    separate state runs. No model recompute occurs.
     """
     country = inputs.country
     country_module = inputs.country_module
diff --git a/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/simulation_output_geographic.py b/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/simulation_output_geographic.py
index 4ce4a49b8..7f76ec7d3 100644
--- a/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/simulation_output_geographic.py
+++ b/projects/policyengine-simulation-executor/src/policyengine_simulation_executor/simulation_output_geographic.py
@@ -223,12 +223,19 @@ def _integer_record_value(record: Mapping[str, Any], key: str) -> int | None:
 
 
 def _should_build_us_congressional_district_impact(region_code: str | None) -> bool:
-    if region_code is None:
-        return True
-    normalized_region_code = region_code.lower()
-    return normalized_region_code == "us" or normalized_region_code.startswith(
-        ("state/", "region_group/")
-    )
+    """Return whether a US run for this region computes congressional districts.
+
+    Only state runs do, such as ``state/ca`` in any case. National runs (no
+    region, or ``us``) and region-group runs compute none, so neither do
+    segmented national runs, their reduce or Stage 12 aggregation of a
+    national report: a nationwide report takes its districts from a separate
+    run of each state on that state's dataset, which gives each district the
+    same figures as its state's own report. Stage 12 aggregation of a state
+    report still computes that state's districts when the request spells the
+    region ``state/...``: that path passes the region as sent, so a bare code
+    such as ``ut`` computes none there. District and place runs compute none.
+    """
+    return region_code is not None and region_code.lower().startswith("state/")
 
 
 def build_congressional_district_impact(
@@ -238,10 +245,11 @@ def build_congressional_district_impact(
     *,
     region_code: str | None = None,
 ) -> CongressionalDistrictImpactOutput | None:
-    """Compute US congressional district impacts for national, state or group runs.
+    """Compute US congressional district impacts for a state run.
 
-    None for other countries or regions, and when the calculation fails (the
-    failure is logged; SPM selection errors still propagate).
+    None for other countries and other regions, including national and
+    region-group runs, and when the calculation fails (the failure is logged;
+    SPM selection errors still propagate).
     """
     if country != "us":
         return None
diff --git a/projects/policyengine-simulation-executor/tests/test_region_group_resolution.py b/projects/policyengine-simulation-executor/tests/test_region_group_resolution.py
index 3c366b068..01d3830b4 100644
--- a/projects/policyengine-simulation-executor/tests/test_region_group_resolution.py
+++ b/projects/policyengine-simulation-executor/tests/test_region_group_resolution.py
@@ -1,4 +1,4 @@
-"""Unit tests for region-group resolution (B2) and the CD-impact gate (B3)."""
+"""Unit tests for region-group resolution (B2) and the district-impact regions (B3)."""
 
 from unittest.mock import MagicMock
 
@@ -11,7 +11,7 @@
 from policyengine_simulation_executor import region_resolution
 from policyengine_simulation_executor import simulation_runtime as sr
 from policyengine_simulation_executor.simulation_output_geographic import (
-    _should_build_us_congressional_district_impact as gate,
+    _should_build_us_congressional_district_impact as should_build_districts,
 )
 
 
@@ -119,13 +119,16 @@ def test__resolve_region_dispatches_to_group(self, monkeypatch):
         assert isinstance(res.scoping_strategy, RegionGroupStrategy)
 
 
-class TestCongressionalDistrictGate:
-    def test__allows_national_state_and_group(self):
-        assert gate(None) is True
-        assert gate("us") is True
-        assert gate("state/ca") is True
-        assert gate("region_group/state/hi+state/ia") is True
+class TestCongressionalDistrictRegions:
+    def test__allows_states(self):
+        assert should_build_districts("state/ca") is True
+        assert should_build_districts("STATE/CA") is True
+
+    def test__rejects_national_and_group(self):
+        assert should_build_districts(None) is False
+        assert should_build_districts("us") is False
+        assert should_build_districts("region_group/state/hi+state/ia") is False
 
     def test__rejects_district_and_place(self):
-        assert gate("congressional_district/CA-01") is False
-        assert gate("place/CA-12345") is False
+        assert should_build_districts("congressional_district/CA-01") is False
+        assert should_build_districts("place/CA-12345") is False
diff --git a/projects/policyengine-simulation-executor/tests/test_simulation_output_builder.py b/projects/policyengine-simulation-executor/tests/test_simulation_output_builder.py
index 7166450fb..b75e5e92a 100644
--- a/projects/policyengine-simulation-executor/tests/test_simulation_output_builder.py
+++ b/projects/policyengine-simulation-executor/tests/test_simulation_output_builder.py
@@ -199,6 +199,51 @@ def fake_build_congressional_district_impact(
     assert calls == [("us", baseline, reform, "state/ut")]
 
 
+@pytest.fixture
+def district_calculations(monkeypatch):
+    """Record each call to policyengine.py's district calculation, in its place."""
+    calculations = []
+
+    def record_district_calculation(baseline, reform):
+        calculations.append((baseline, reform))
+        return SimpleNamespace(district_results=[])
+
+    monkeypatch.setattr(
+        "policyengine.outputs.congressional_district_impact."
+        "compute_us_congressional_district_impacts",
+        record_district_calculation,
+    )
+    return calculations
+
+
+@pytest.mark.parametrize(
+    "region_code",
+    [None, "us", "region_group/state/hi+state/ia"],
+)
+def test_builder_skips_districts_for_national_and_region_group_runs(
+    district_calculations, region_code
+):
+    baseline, reform = macro_baseline_reform()
+    builder = make_output_builder(
+        "us", baseline, reform, BuilderOptions(resolved_region_code=region_code)
+    )
+
+    assert builder._build_congressional_district_impact() is None
+    assert district_calculations == []
+
+
+def test_builder_builds_districts_for_state_runs(district_calculations):
+    baseline, reform = macro_baseline_reform()
+    builder = make_output_builder(
+        "us", baseline, reform, BuilderOptions(resolved_region_code="state/ut")
+    )
+
+    assert builder._build_congressional_district_impact() == (
+        CongressionalDistrictImpactOutput(districts=[])
+    )
+    assert district_calculations == [(baseline, reform)]
+
+
 def test_builder_records_output_timings_without_real_calculation(monkeypatch):
     baseline = object()
     reform = object()