From 261bf61d18b48fe5b174e69351a9a3992392dbf0 Mon Sep 17 00:00:00 2001
From: Anthony Volk <14987227+anth-volk@users.noreply.github.com>
Date: Tue, 8 Sep 2026 18:24:09 +0400
Subject: [PATCH 1/4] Add parallel live simulation coverage

---
 .github/workflows/push.yml                    | 52 +++++++++++-
 changelog.d/2400.added.md                     |  1 +
 tests/data/california_eitc_reform.json        |  7 ++
 .../live_uk_universal_credit_household.json   | 48 +++++++++++
 tests/data/live_us_credits_household.json     | 85 +++++++++++++++++++
 tests/data/uk_universal_credit_reform.json    |  7 ++
 tests/integration/test_live_calculate.py      | 44 ++++++++++
 tests/integration/test_live_economy.py        | 72 ++++++++++++++++
 tests/unit/test_cloud_run_deploy_scripts.py   | 27 +++++-
 9 files changed, 336 insertions(+), 7 deletions(-)
 create mode 100644 changelog.d/2400.added.md
 create mode 100644 tests/data/california_eitc_reform.json
 create mode 100644 tests/data/live_uk_universal_credit_household.json
 create mode 100644 tests/data/live_us_credits_household.json
 create mode 100644 tests/data/uk_universal_credit_reform.json

diff --git a/.github/workflows/push.yml b/.github/workflows/push.yml
index 856b4f570..796d9b2b9 100644
--- a/.github/workflows/push.yml
+++ b/.github/workflows/push.yml
@@ -304,7 +304,7 @@ jobs:
         run: bash .github/scripts/health_check.sh "${{ steps.candidate.outputs.url }}/readiness-check"
 
   integration-tests-staging-cloud-run:
-    name: Run Cloud Run staging integration tests
+    name: Run core Cloud Run staging integration tests
     runs-on: ubuntu-latest
     needs: deploy-cloud-run-staging
     if: |
@@ -320,19 +320,65 @@ jobs:
           python-version: "3.12"
       - name: Install staging test dependencies
         run: pip install pytest httpx sqlalchemy psycopg[binary]
-      - name: Run staging smoke test
-        run: python -m pytest tests/integration/test_cloud_run_candidate.py tests/integration/test_live_v2_metadata.py tests/integration/test_live_v2_policies.py tests/integration/test_live_calculate.py tests/integration/test_live_economy.py tests/integration/test_live_budget_window_cache.py -v
+      - name: Run core staging integration tests
+        run: python -m pytest tests/integration/test_cloud_run_candidate.py tests/integration/test_live_v2_metadata.py tests/integration/test_live_v2_policies.py -v
         env:
           API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}
           STAGING_API_TEST_PROBE_ID: cloud-run-${{ needs.deploy-cloud-run-staging.outputs.tag }}
           V2_MIGRATION_DATABASE_URL: ${{ secrets.V2_MIGRATION_DATABASE_URL }}
 
+  parallel-live-simulation-tests-staging-cloud-run:
+    name: Run ${{ matrix.name }} against the staging candidate
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    needs:
+      - deploy-cloud-run-staging
+      - integration-tests-staging-cloud-run
+    if: |
+      (github.repository == 'PolicyEngine/policyengine-api')
+      && (github.event.head_commit.message == 'Update PolicyEngine API')
+    strategy:
+      fail-fast: false
+      max-parallel: 2
+      matrix:
+        include:
+          - name: budget-window simulations
+            test: tests/integration/test_live_budget_window_cache.py
+            probe: budget-window
+          - name: household calculations
+            test: tests/integration/test_live_calculate.py
+            probe: household
+          - name: Utah simulation
+            test: tests/integration/test_live_economy.py::test_live_utah_macro_reform
+            probe: utah
+          - name: California CalEITC simulation
+            test: tests/integration/test_live_economy.py::test_live_california_eitc_macro_reform
+            probe: california-eitc
+          - name: UK Scotland Universal Credit simulation
+            test: tests/integration/test_live_economy.py::test_live_uk_universal_credit_macro_reform_in_scotland
+            probe: uk-scotland-uc
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v4
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install live test dependencies
+        run: pip install pytest httpx
+      - name: Run live staging test selection
+        run: python -m pytest ${{ matrix.test }} -v
+        env:
+          API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}
+          STAGING_API_TEST_PROBE_ID: cloud-run-${{ needs.deploy-cloud-run-staging.outputs.tag }}-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.probe }}
+
   promote-cloud-run-staging:
     name: Promote staging Cloud Run traffic
     runs-on: ubuntu-latest
     needs:
       - deploy-cloud-run-staging
       - integration-tests-staging-cloud-run
+      - parallel-live-simulation-tests-staging-cloud-run
     if: |
       (github.repository == 'PolicyEngine/policyengine-api')
       && (github.event.head_commit.message == 'Update PolicyEngine API')
diff --git a/changelog.d/2400.added.md b/changelog.d/2400.added.md
new file mode 100644
index 000000000..e72c44258
--- /dev/null
+++ b/changelog.d/2400.added.md
@@ -0,0 +1 @@
+Add parallel live staging coverage for US federal and California tax credits and UK Universal Credit calculations.
diff --git a/tests/data/california_eitc_reform.json b/tests/data/california_eitc_reform.json
new file mode 100644
index 000000000..f64a5e6bb
--- /dev/null
+++ b/tests/data/california_eitc_reform.json
@@ -0,0 +1,7 @@
+{
+  "data": {
+    "gov.states.ca.tax.income.credits.earned_income.phase_in.rate[1].amount": {
+      "2025-01-01.2100-12-31": 0.35
+    }
+  }
+}
diff --git a/tests/data/live_uk_universal_credit_household.json b/tests/data/live_uk_universal_credit_household.json
new file mode 100644
index 000000000..41560f113
--- /dev/null
+++ b/tests/data/live_uk_universal_credit_household.json
@@ -0,0 +1,48 @@
+{
+  "household": {
+    "benunits": {
+      "benefit unit": {
+        "members": [
+          "parent",
+          "child"
+        ],
+        "universal_credit": {
+          "2025": null
+        },
+        "would_claim_uc": {
+          "2025": true
+        }
+      }
+    },
+    "households": {
+      "household": {
+        "members": [
+          "parent",
+          "child"
+        ],
+        "region": {
+          "2025": "SCOTLAND"
+        }
+      }
+    },
+    "people": {
+      "parent": {
+        "age": {
+          "2025": 35
+        },
+        "employment_income": {
+          "2025": 12000
+        }
+      },
+      "child": {
+        "age": {
+          "2025": 6
+        },
+        "employment_income": {
+          "2025": 0
+        }
+      }
+    }
+  },
+  "policy": {}
+}
diff --git a/tests/data/live_us_credits_household.json b/tests/data/live_us_credits_household.json
new file mode 100644
index 000000000..ca1e6d02a
--- /dev/null
+++ b/tests/data/live_us_credits_household.json
@@ -0,0 +1,85 @@
+{
+  "household": {
+    "families": {
+      "family": {
+        "members": [
+          "parent",
+          "child"
+        ]
+      }
+    },
+    "households": {
+      "household": {
+        "members": [
+          "parent",
+          "child"
+        ],
+        "state_name": {
+          "2025": "CA"
+        }
+      }
+    },
+    "marital_units": {
+      "parent's marital unit": {
+        "members": [
+          "parent"
+        ]
+      },
+      "child's marital unit": {
+        "marital_unit_id": {
+          "2025": 1
+        },
+        "members": [
+          "child"
+        ]
+      }
+    },
+    "people": {
+      "parent": {
+        "age": {
+          "2025": 35
+        },
+        "employment_income": {
+          "2025": 20000
+        }
+      },
+      "child": {
+        "age": {
+          "2025": 6
+        },
+        "employment_income": {
+          "2025": 0
+        },
+        "is_tax_unit_dependent": {
+          "2025": true
+        }
+      }
+    },
+    "spm_units": {
+      "spm unit": {
+        "members": [
+          "parent",
+          "child"
+        ]
+      }
+    },
+    "tax_units": {
+      "tax unit": {
+        "members": [
+          "parent",
+          "child"
+        ],
+        "eitc": {
+          "2025": null
+        },
+        "ctc": {
+          "2025": null
+        },
+        "ca_eitc": {
+          "2025": null
+        }
+      }
+    }
+  },
+  "policy": {}
+}
diff --git a/tests/data/uk_universal_credit_reform.json b/tests/data/uk_universal_credit_reform.json
new file mode 100644
index 000000000..28cae5218
--- /dev/null
+++ b/tests/data/uk_universal_credit_reform.json
@@ -0,0 +1,7 @@
+{
+  "data": {
+    "gov.dwp.universal_credit.standard_allowance.amount.SINGLE_OLD": {
+      "2025-01-01.2100-12-31": 500
+    }
+  }
+}
diff --git a/tests/integration/test_live_calculate.py b/tests/integration/test_live_calculate.py
index 092c464ae..af792d5d6 100644
--- a/tests/integration/test_live_calculate.py
+++ b/tests/integration/test_live_calculate.py
@@ -1,5 +1,8 @@
+import json
 from pathlib import Path
 
+import pytest
+
 
 def _load_payload(filename: str) -> str:
     return (Path(__file__).resolve().parents[1] / "data" / filename).read_text(
@@ -7,6 +10,10 @@ def _load_payload(filename: str) -> str:
     )
 
 
+def _load_json_payload(filename: str) -> dict:
+    return json.loads(_load_payload(filename))
+
+
 def test_live_calculate_us_1(api_client):
     response = api_client.post(
         "/us/calculate",
@@ -61,3 +68,40 @@ def test_live_calculate_drops_deprecated_medical_input(
         "medical_out_of_pocket_expenses" in warning and "deprecated" in warning.lower()
         for warning in payload["warnings"]
     )
+
+
+def test_live_calculate_us_federal_and_california_credits(
+    api_client,
+    integration_probe_id,
+):
+    request_payload = _load_json_payload("live_us_credits_household.json")
+    request_payload["integration_probe_id"] = f"{integration_probe_id}-us-credits"
+
+    response = api_client.post("/us/calculate", json=request_payload)
+
+    assert response.status_code == 200, response.text
+    payload = response.json()
+    assert payload["status"] == "ok", payload
+    tax_unit = payload["result"]["tax_units"]["tax unit"]
+    assert tax_unit["eitc"]["2025"] == pytest.approx(4328.0, abs=0.01)
+    assert tax_unit["ctc"]["2025"] == pytest.approx(2200.0, abs=0.01)
+    assert tax_unit["ca_eitc"]["2025"] == pytest.approx(395.86, abs=0.01)
+
+
+def test_live_calculate_uk_universal_credit_in_scotland(
+    api_client,
+    integration_probe_id,
+):
+    request_payload = _load_json_payload("live_uk_universal_credit_household.json")
+    request_payload["integration_probe_id"] = f"{integration_probe_id}-uk-uc"
+
+    response = api_client.post("/uk/calculate", json=request_payload)
+
+    assert response.status_code == 200, response.text
+    payload = response.json()
+    assert payload["status"] == "ok", payload
+    benefit_unit = payload["result"]["benunits"]["benefit unit"]
+    assert benefit_unit["universal_credit"]["2025"] == pytest.approx(
+        6229.80,
+        abs=0.01,
+    )
diff --git a/tests/integration/test_live_economy.py b/tests/integration/test_live_economy.py
index 7988f3afd..89cc082bb 100644
--- a/tests/integration/test_live_economy.py
+++ b/tests/integration/test_live_economy.py
@@ -75,3 +75,75 @@ def test_live_utah_macro_reform(api_client, integration_probe_id, poll_live_endp
     assert isinstance(lose_less_than_5, int | float), result
     assert math.isfinite(lose_less_than_5), result
     assert 0 <= lose_less_than_5 <= 1, result
+
+
+def _assert_live_macro_reform(
+    api_client,
+    integration_probe_id,
+    poll_live_endpoint,
+    *,
+    country_id: str,
+    reform_filename: str,
+    region: str,
+    probe_suffix: str,
+) -> None:
+    metadata_response = api_client.get(f"/{country_id}/metadata")
+    metadata_response.raise_for_status()
+    metadata = metadata_response.json()["result"]
+
+    policy_response = api_client.post(
+        f"/{country_id}/policy",
+        json=_load_reform_payload(reform_filename),
+    )
+    assert policy_response.status_code in (200, 201), policy_response.text
+    policy_id = policy_response.json()["result"]["policy_id"]
+
+    payload = poll_live_endpoint(
+        api_client,
+        f"/{country_id}/economy/{policy_id}/over/{metadata['current_law_id']}",
+        {
+            "region": region,
+            "time_period": _pick_time_period(metadata),
+            "staging_probe": f"{integration_probe_id}-{probe_suffix}",
+        },
+        route_name=f"{country_id}-{probe_suffix}-economy",
+    )
+
+    assert payload["status"] == "ok", payload
+    result = payload["result"]
+    assert result is not None, payload
+    budgetary_impact = result["budget"]["budgetary_impact"]
+    assert isinstance(budgetary_impact, int | float), result
+    assert math.isfinite(budgetary_impact), result
+
+
+def test_live_california_eitc_macro_reform(
+    api_client,
+    integration_probe_id,
+    poll_live_endpoint,
+):
+    _assert_live_macro_reform(
+        api_client,
+        integration_probe_id,
+        poll_live_endpoint,
+        country_id="us",
+        reform_filename="california_eitc_reform.json",
+        region="state/ca",
+        probe_suffix="california-eitc",
+    )
+
+
+def test_live_uk_universal_credit_macro_reform_in_scotland(
+    api_client,
+    integration_probe_id,
+    poll_live_endpoint,
+):
+    _assert_live_macro_reform(
+        api_client,
+        integration_probe_id,
+        poll_live_endpoint,
+        country_id="uk",
+        reform_filename="uk_universal_credit_reform.json",
+        region="country/scotland",
+        probe_suffix="uk-scotland-uc",
+    )
diff --git a/tests/unit/test_cloud_run_deploy_scripts.py b/tests/unit/test_cloud_run_deploy_scripts.py
index dd6842606..54e6c764f 100644
--- a/tests/unit/test_cloud_run_deploy_scripts.py
+++ b/tests/unit/test_cloud_run_deploy_scripts.py
@@ -1853,6 +1853,10 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
         workflow,
         "integration-tests-staging-cloud-run",
     )
+    parallel_live_tests = _workflow_job_block(
+        workflow,
+        "parallel-live-simulation-tests-staging-cloud-run",
+    )
     cloud_run_promotion = _workflow_job_block(workflow, "promote-cloud-run-staging")
     production_check = _workflow_job_block(
         workflow,
@@ -1861,10 +1865,7 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
     cloud_run_test_command = (
         "python -m pytest tests/integration/test_cloud_run_candidate.py "
         "tests/integration/test_live_v2_metadata.py "
-        "tests/integration/test_live_v2_policies.py "
-        "tests/integration/test_live_calculate.py "
-        "tests/integration/test_live_economy.py "
-        "tests/integration/test_live_budget_window_cache.py -v"
+        "tests/integration/test_live_v2_policies.py -v"
     )
 
     assert "make test" in cloud_run_deploy
@@ -1881,9 +1882,27 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
     )
     assert "environment: staging" in cloud_run_tests
     assert "V2_MIGRATION_DATABASE_URL" in cloud_run_tests
+    assert "fail-fast: false" in parallel_live_tests
+    assert "max-parallel: 2" in parallel_live_tests
+    assert "- integration-tests-staging-cloud-run" in parallel_live_tests
+    assert "tests/integration/test_live_calculate.py" in parallel_live_tests
+    assert "test_live_utah_macro_reform" in parallel_live_tests
+    assert "test_live_california_eitc_macro_reform" in parallel_live_tests
+    assert (
+        "test_live_uk_universal_credit_macro_reform_in_scotland" in parallel_live_tests
+    )
+    assert "tests/integration/test_live_v2_policies.py" not in parallel_live_tests
+    assert (
+        "API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}"
+        in parallel_live_tests
+    )
+    assert "github.run_id" in parallel_live_tests
+    assert "github.run_attempt" in parallel_live_tests
+    assert "matrix.probe" in parallel_live_tests
     assert "needs: promote-cloud-run-staging" in production_check
     assert "- integration-tests-staging-cloud-run" not in production_check
     assert "- integration-tests-staging-cloud-run" in cloud_run_promotion
+    assert "- parallel-live-simulation-tests-staging-cloud-run" in cloud_run_promotion
     assert "exercise-phase10-staging" not in workflow
     assert "run_phase10_staging_probe.sh" not in workflow
     assert "V2_FAILURE_DATABASE_URL_SECRET_RESOURCE" not in workflow

From 2c38a719e4fec3cde232066bd6dae1c589f93663 Mon Sep 17 00:00:00 2001
From: Anthony Volk <14987227+anth-volk@users.noreply.github.com>
Date: Tue, 8 Sep 2026 20:23:22 +0400
Subject: [PATCH 2/4] Run live tests with pytest workers

---
 .github/scripts/run_live_staging_tests.sh   | 11 ++++++++
 .github/workflows/push.yml                  | 30 ++++-----------------
 tests/unit/test_cloud_run_deploy_scripts.py | 20 +++++++-------
 3 files changed, 27 insertions(+), 34 deletions(-)
 create mode 100755 .github/scripts/run_live_staging_tests.sh

diff --git a/.github/scripts/run_live_staging_tests.sh b/.github/scripts/run_live_staging_tests.sh
new file mode 100755
index 000000000..39bc7a3cb
--- /dev/null
+++ b/.github/scripts/run_live_staging_tests.sh
@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+python -m pytest \
+  -n 2 \
+  --dist load \
+  --maxschedchunk=1 \
+  tests/integration/test_live_budget_window_cache.py \
+  tests/integration/test_live_calculate.py \
+  tests/integration/test_live_economy.py \
+  -v
diff --git a/.github/workflows/push.yml b/.github/workflows/push.yml
index 796d9b2b9..9f598e254 100644
--- a/.github/workflows/push.yml
+++ b/.github/workflows/push.yml
@@ -328,7 +328,7 @@ jobs:
           V2_MIGRATION_DATABASE_URL: ${{ secrets.V2_MIGRATION_DATABASE_URL }}
 
   parallel-live-simulation-tests-staging-cloud-run:
-    name: Run ${{ matrix.name }} against the staging candidate
+    name: Run live simulations against the staging candidate
     runs-on: ubuntu-latest
     timeout-minutes: 20
     needs:
@@ -337,26 +337,6 @@ jobs:
     if: |
       (github.repository == 'PolicyEngine/policyengine-api')
       && (github.event.head_commit.message == 'Update PolicyEngine API')
-    strategy:
-      fail-fast: false
-      max-parallel: 2
-      matrix:
-        include:
-          - name: budget-window simulations
-            test: tests/integration/test_live_budget_window_cache.py
-            probe: budget-window
-          - name: household calculations
-            test: tests/integration/test_live_calculate.py
-            probe: household
-          - name: Utah simulation
-            test: tests/integration/test_live_economy.py::test_live_utah_macro_reform
-            probe: utah
-          - name: California CalEITC simulation
-            test: tests/integration/test_live_economy.py::test_live_california_eitc_macro_reform
-            probe: california-eitc
-          - name: UK Scotland Universal Credit simulation
-            test: tests/integration/test_live_economy.py::test_live_uk_universal_credit_macro_reform_in_scotland
-            probe: uk-scotland-uc
     steps:
       - name: Checkout repo
         uses: actions/checkout@v4
@@ -365,12 +345,12 @@ jobs:
         with:
           python-version: "3.12"
       - name: Install live test dependencies
-        run: pip install pytest httpx
-      - name: Run live staging test selection
-        run: python -m pytest ${{ matrix.test }} -v
+        run: pip install pytest pytest-xdist httpx
+      - name: Run live staging tests in two pytest workers
+        run: bash .github/scripts/run_live_staging_tests.sh
         env:
           API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}
-          STAGING_API_TEST_PROBE_ID: cloud-run-${{ needs.deploy-cloud-run-staging.outputs.tag }}-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.probe }}
+          STAGING_API_TEST_PROBE_ID: cloud-run-${{ needs.deploy-cloud-run-staging.outputs.tag }}-${{ github.run_id }}-${{ github.run_attempt }}
 
   promote-cloud-run-staging:
     name: Promote staging Cloud Run traffic
diff --git a/tests/unit/test_cloud_run_deploy_scripts.py b/tests/unit/test_cloud_run_deploy_scripts.py
index 54e6c764f..26703a0e8 100644
--- a/tests/unit/test_cloud_run_deploy_scripts.py
+++ b/tests/unit/test_cloud_run_deploy_scripts.py
@@ -1867,6 +1867,7 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
         "tests/integration/test_live_v2_metadata.py "
         "tests/integration/test_live_v2_policies.py -v"
     )
+    live_test_script = (REPO / ".github/scripts/run_live_staging_tests.sh").read_text()
 
     assert "make test" in cloud_run_deploy
     assert cloud_run_deploy.index("make test") < cloud_run_deploy.index(
@@ -1882,15 +1883,17 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
     )
     assert "environment: staging" in cloud_run_tests
     assert "V2_MIGRATION_DATABASE_URL" in cloud_run_tests
-    assert "fail-fast: false" in parallel_live_tests
-    assert "max-parallel: 2" in parallel_live_tests
+    assert "strategy:" not in parallel_live_tests
+    assert "matrix:" not in parallel_live_tests
     assert "- integration-tests-staging-cloud-run" in parallel_live_tests
-    assert "tests/integration/test_live_calculate.py" in parallel_live_tests
-    assert "test_live_utah_macro_reform" in parallel_live_tests
-    assert "test_live_california_eitc_macro_reform" in parallel_live_tests
-    assert (
-        "test_live_uk_universal_credit_macro_reform_in_scotland" in parallel_live_tests
-    )
+    assert "pip install pytest pytest-xdist httpx" in parallel_live_tests
+    assert "bash .github/scripts/run_live_staging_tests.sh" in parallel_live_tests
+    assert "-n 2" in live_test_script
+    assert "--dist load" in live_test_script
+    assert "--maxschedchunk=1" in live_test_script
+    assert "tests/integration/test_live_budget_window_cache.py" in live_test_script
+    assert "tests/integration/test_live_calculate.py" in live_test_script
+    assert "tests/integration/test_live_economy.py" in live_test_script
     assert "tests/integration/test_live_v2_policies.py" not in parallel_live_tests
     assert (
         "API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}"
@@ -1898,7 +1901,6 @@ def test_push_workflow_runs_release_and_cloud_run_staging_tests():
     )
     assert "github.run_id" in parallel_live_tests
     assert "github.run_attempt" in parallel_live_tests
-    assert "matrix.probe" in parallel_live_tests
     assert "needs: promote-cloud-run-staging" in production_check
     assert "- integration-tests-staging-cloud-run" not in production_check
     assert "- integration-tests-staging-cloud-run" in cloud_run_promotion

From 6b5b92df60992c03727a30f31a467d8b1d256d04 Mon Sep 17 00:00:00 2001
From: Anthony Volk <14987227+anth-volk@users.noreply.github.com>
Date: Tue, 8 Sep 2026 21:20:53 +0400
Subject: [PATCH 3/4] Remove ambiguous core wording from workflow

---
 .github/workflows/push.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/push.yml b/.github/workflows/push.yml
index 9f598e254..a3a0f2dbe 100644
--- a/.github/workflows/push.yml
+++ b/.github/workflows/push.yml
@@ -304,7 +304,7 @@ jobs:
         run: bash .github/scripts/health_check.sh "${{ steps.candidate.outputs.url }}/readiness-check"
 
   integration-tests-staging-cloud-run:
-    name: Run core Cloud Run staging integration tests
+    name: Run Cloud Run staging integration tests
     runs-on: ubuntu-latest
     needs: deploy-cloud-run-staging
     if: |
@@ -320,7 +320,7 @@ jobs:
           python-version: "3.12"
       - name: Install staging test dependencies
         run: pip install pytest httpx sqlalchemy psycopg[binary]
-      - name: Run core staging integration tests
+      - name: Run staging integration tests
         run: python -m pytest tests/integration/test_cloud_run_candidate.py tests/integration/test_live_v2_metadata.py tests/integration/test_live_v2_policies.py -v
         env:
           API_BASE_URL: ${{ needs.deploy-cloud-run-staging.outputs.url }}

From ab8a9c171e4e1b59e703f0847bc49e718ffbb205 Mon Sep 17 00:00:00 2001
From: Anthony Volk <14987227+anth-volk@users.noreply.github.com>
Date: Fri, 11 Sep 2026 22:45:01 +0400
Subject: [PATCH 4/4] Handle structured errors in live polling

---
 tests/integration/conftest.py   | 13 ++++++++
 tests/unit/test_live_polling.py | 58 +++++++++++++++++++++++++++++++++
 2 files changed, 71 insertions(+)
 create mode 100644 tests/unit/test_live_polling.py

diff --git a/tests/integration/conftest.py b/tests/integration/conftest.py
index d3e3c40bc..78283bd40 100644
--- a/tests/integration/conftest.py
+++ b/tests/integration/conftest.py
@@ -85,6 +85,16 @@ def _response_summary(response: httpx.Response) -> str:
     return f"HTTP {response.status_code}: {response.text[:500]}"
 
 
+def _structured_error_payload(response: httpx.Response) -> dict | None:
+    try:
+        payload = response.json()
+    except ValueError:
+        return None
+    if isinstance(payload, dict) and payload.get("status") == "error":
+        return payload
+    return None
+
+
 def _poll_live_endpoint(
     api_client: httpx.Client,
     path: str,
@@ -102,6 +112,9 @@ def _poll_live_endpoint(
             last_response = f"{type(error).__name__}: {error}"
         else:
             if response.status_code in TRANSIENT_POLL_STATUS_CODES:
+                error_payload = _structured_error_payload(response)
+                if error_payload is not None:
+                    return error_payload
                 last_response = _response_summary(response)
             else:
                 response.raise_for_status()
diff --git a/tests/unit/test_live_polling.py b/tests/unit/test_live_polling.py
new file mode 100644
index 000000000..7dd33ec55
--- /dev/null
+++ b/tests/unit/test_live_polling.py
@@ -0,0 +1,58 @@
+from unittest.mock import Mock
+
+import httpx
+
+from tests.integration import conftest as live_polling
+
+
+def _response(
+    status_code: int,
+    *,
+    json_payload: dict | None = None,
+    text: str | None = None,
+) -> httpx.Response:
+    request = httpx.Request("GET", "https://example.test/economy")
+    if json_payload is not None:
+        return httpx.Response(status_code, json=json_payload, request=request)
+    return httpx.Response(status_code, text=text, request=request)
+
+
+def test_structured_bad_gateway_response_is_returned_immediately(monkeypatch):
+    payload = {
+        "status": "error",
+        "message": "Simulation worker exited",
+        "result": None,
+    }
+    client = Mock()
+    client.get.return_value = _response(502, json_payload=payload)
+    monkeypatch.setattr(live_polling, "INTEGRATION_TIMEOUT_SECONDS", 0)
+
+    result = live_polling._poll_live_endpoint(
+        client,
+        "/us/economy/1/over/2",
+        {},
+        route_name="economy",
+    )
+
+    assert result == payload
+    client.get.assert_called_once()
+
+
+def test_unstructured_bad_gateway_response_is_retried(monkeypatch):
+    success_payload = {"status": "ok", "result": {"budget": {}}}
+    client = Mock()
+    client.get.side_effect = [
+        _response(502, text="<html>Temporary proxy failure</html>"),
+        _response(200, json_payload=success_payload),
+    ]
+    monkeypatch.setattr(live_polling.time, "sleep", lambda _: None)
+
+    result = live_polling._poll_live_endpoint(
+        client,
+        "/us/economy/1/over/2",
+        {},
+        route_name="economy",
+    )
+
+    assert result == success_payload
+    assert client.get.call_count == 2