diff --git a/.github/workflows/push.yml b/.github/workflows/push.yml
index ea9658024..7cecdb294 100644
--- a/.github/workflows/push.yml
+++ b/.github/workflows/push.yml
@@ -172,10 +172,12 @@ jobs:
     secrets: inherit
 
   seed-v2-staging-database:
+    # Runs beside migrate-v1-staging-cloud-sql: the seeding job connects only to
+    # the v2 database and the v1 upgrade only to Cloud SQL, so neither needs the
+    # other. The staging deploy needs both.
     name: Seed staging v2 database
     needs:
       - publish-git-tag
-      - migrate-v1-staging-cloud-sql
     if: |
       (github.repository == 'PolicyEngine/policyengine-api')
       && (github.event.head_commit.message == 'Update PolicyEngine API')
@@ -515,10 +517,11 @@ jobs:
     secrets: inherit
 
   seed-v2-production-database:
+    # Runs beside migrate-v1-production-cloud-sql, as in staging. The
+    # production deploy needs both.
     name: Seed production v2 database
     needs:
       - ensure-production-model-version-aligns-with-sim-api
-      - migrate-v1-production-cloud-sql
     if: |
       (github.repository == 'PolicyEngine/policyengine-api')
       && (github.event.head_commit.message == 'Update PolicyEngine API')
@@ -530,7 +533,9 @@ jobs:
   deploy-cloud-run-candidate:
     name: Deploy production Cloud Run candidate
     runs-on: ubuntu-latest
-    needs: seed-v2-production-database
+    needs:
+      - migrate-v1-production-cloud-sql
+      - seed-v2-production-database
     if: |
       (github.repository == 'PolicyEngine/policyengine-api')
       && (github.event.head_commit.message == 'Update PolicyEngine API')
diff --git a/changelog.d/parallel-database-steps.changed.md b/changelog.d/parallel-database-steps.changed.md
new file mode 100644
index 000000000..332e21ef6
--- /dev/null
+++ b/changelog.d/parallel-database-steps.changed.md
@@ -0,0 +1 @@
+The release run upgrades the v1 Cloud SQL schema and seeds the v2 database side by side, in staging and in production, instead of one after the other; each Cloud Run candidate deploy still waits for both.
diff --git a/docs/engineering/skills/alembic-migrations.md b/docs/engineering/skills/alembic-migrations.md
index 9bcf0139a..e8cda9a72 100644
--- a/docs/engineering/skills/alembic-migrations.md
+++ b/docs/engineering/skills/alembic-migrations.md
@@ -344,6 +344,13 @@ shared or deployed database.
   unversioned or invalid history, takes a backup only when revisions are
   pending, upgrades under the advisory lock, and requires zero post-migration
   drift.
+- In each environment the v1 release migration and the v2 seeding job
+  (`seed-v2-database.yml`, which upgrades the v2 schema and publishes the
+  catalog) run side by side: the v1 job connects only to Cloud SQL and the
+  seeding job only to the v2 database. Each Cloud Run candidate deploy needs
+  both, so a candidate is never built until both databases are upgraded. When
+  one of them fails, the release stops with the other already applied; the
+  serving revision must work with either database upgraded on its own.
 - Database credentials and derived URLs must never be printed. The application
   runtime account must not be granted schema-migration privileges.
 - Production downgrades are never automatic. MySQL DDL is non-transactional, so
diff --git a/tests/unit/test_app_engine_decommission.py b/tests/unit/test_app_engine_decommission.py
index 926d04727..a078ceef0 100644
--- a/tests/unit/test_app_engine_decommission.py
+++ b/tests/unit/test_app_engine_decommission.py
@@ -77,7 +77,10 @@ def test_cloud_run_is_the_complete_release_sequence() -> None:
     docker_publish = _job_block(workflow, "docker")
     release_tests = _job_block(workflow, "release-tests")
 
-    assert "migrate-v1-staging-cloud-sql" in staging_seed
+    # The v1 schema upgrade and the v2 seeding run side by side, and each
+    # candidate deploy needs both (tests/unit/v2/test_metadata_deployment.py).
+    assert "- migrate-v1-staging-cloud-sql" not in staging_seed
+    assert "- migrate-v1-staging-cloud-sql" in staging_deploy
     assert "- seed-v2-staging-database" in staging_deploy
     # The release tests run in their own job, and the staging deploy needs it.
     assert "- release-tests" in staging_deploy
@@ -93,8 +96,9 @@ def test_cloud_run_is_the_complete_release_sequence() -> None:
     assert "exercise-phase10-staging" not in workflow
     assert "exercise-phase11-staging" not in workflow
     assert "needs: promote-cloud-run-staging" in production_check
-    assert "migrate-v1-production-cloud-sql" in production_seed
-    assert "needs: seed-v2-production-database" in production_deploy
+    assert "- migrate-v1-production-cloud-sql" not in production_seed
+    assert "- migrate-v1-production-cloud-sql" in production_deploy
+    assert "- seed-v2-production-database" in production_deploy
     assert "needs: deploy-cloud-run-candidate" in docker_publish
 
 
diff --git a/tests/unit/test_cloud_run_push_workflow.py b/tests/unit/test_cloud_run_push_workflow.py
index e0c41371d..c4ed071b1 100644
--- a/tests/unit/test_cloud_run_push_workflow.py
+++ b/tests/unit/test_cloud_run_push_workflow.py
@@ -392,7 +392,8 @@ def test_push_workflow_staging_precedes_all_production_deployments():
     docker_publish = _workflow_job_block(workflow, "docker")
     cloud_run_production = _workflow_job_block(workflow, "deploy-cloud-run-candidate")
 
-    assert "needs: seed-v2-production-database" in cloud_run_production
+    assert "- seed-v2-production-database" in cloud_run_production
+    assert "- migrate-v1-production-cloud-sql" in cloud_run_production
     assert "needs: deploy-cloud-run-candidate" in docker_publish
     assert "compute_cloud_run_metadata.sh stage3-prod" in cloud_run_production
     assert "Build and push Cloud Run image" not in cloud_run_production
diff --git a/tests/unit/v2/test_metadata_deployment.py b/tests/unit/v2/test_metadata_deployment.py
index c752b46c0..eaf4809e7 100644
--- a/tests/unit/v2/test_metadata_deployment.py
+++ b/tests/unit/v2/test_metadata_deployment.py
@@ -6,8 +6,28 @@
 import subprocess
 from pathlib import Path
 
+from tests.unit.cloud_run_deploy_support import job_needs, push_workflow_jobs
+
 REPO = Path(__file__).resolve().parents[3]
 
+# Per environment: the v1 Cloud SQL schema upgrade, the v2 seeding, the job
+# that deploys the Cloud Run candidate, and the job both database jobs start
+# after.
+DATABASE_JOBS_BY_ENVIRONMENT = {
+    "staging": {
+        "v1_upgrade": "migrate-v1-staging-cloud-sql",
+        "v2_seeding": "seed-v2-staging-database",
+        "candidate_deploy": "deploy-cloud-run-staging",
+        "start_after": "publish-git-tag",
+    },
+    "production": {
+        "v1_upgrade": "migrate-v1-production-cloud-sql",
+        "v2_seeding": "seed-v2-production-database",
+        "candidate_deploy": "deploy-cloud-run-candidate",
+        "start_after": "ensure-production-model-version-aligns-with-sim-api",
+    },
+}
+
 
 def _read(relative_path: str) -> str:
     return (REPO / relative_path).read_text(encoding="utf-8")
@@ -142,23 +162,39 @@ def test_privilege_steps_connect_with_the_migration_credential() -> None:
 
 
 def test_seeding_success_is_required_before_candidate_creation() -> None:
-    workflow = _read(".github/workflows/push.yml")
-    staging_seed = _job(workflow, "seed-v2-staging-database")
-    production_seed = _job(workflow, "seed-v2-production-database")
-
-    assert "deployment_environment: staging" in staging_seed
-    assert "migrate-v1-staging-cloud-sql" in staging_seed
-    assert "seed-v2-staging-database" in _job(
-        workflow,
-        "deploy-cloud-run-staging",
-    )
-
-    assert "migrate-v1-production-cloud-sql" in production_seed
-    assert "deployment_environment: production" in production_seed
-    assert "needs: seed-v2-production-database" in _job(
-        workflow,
-        "deploy-cloud-run-candidate",
-    )
+    jobs = push_workflow_jobs()
+
+    for environment, database_jobs in DATABASE_JOBS_BY_ENVIRONMENT.items():
+        seeding = jobs[database_jobs["v2_seeding"]]
+        candidate_deploy = jobs[database_jobs["candidate_deploy"]]
+
+        assert seeding["with"] == {"deployment_environment": environment}
+        assert seeding["secrets"] == "inherit"
+        # The candidate deploy waits for both database jobs.
+        assert database_jobs["v2_seeding"] in job_needs(candidate_deploy)
+        assert database_jobs["v1_upgrade"] in job_needs(candidate_deploy)
+
+
+def test_v1_upgrade_and_v2_seeding_run_side_by_side() -> None:
+    # Neither job reads the other's database: the v1 upgrade connects only to
+    # Cloud SQL through its proxy, and the seeding job only to the v2 database,
+    # with no Cloud SQL proxy or v1 credentials.
+    jobs = push_workflow_jobs()
+
+    for environment, database_jobs in DATABASE_JOBS_BY_ENVIRONMENT.items():
+        v1_upgrade = jobs[database_jobs["v1_upgrade"]]
+        seeding = jobs[database_jobs["v2_seeding"]]
+
+        assert database_jobs["v1_upgrade"] not in job_needs(seeding), environment
+        assert database_jobs["v2_seeding"] not in job_needs(v1_upgrade), environment
+        assert database_jobs["start_after"] in job_needs(seeding), environment
+        assert database_jobs["start_after"] in job_needs(v1_upgrade), environment
+        # Each keeps the GitHub environment it reads its database target from.
+        assert v1_upgrade["with"] == {
+            "deployment_environment": environment,
+            "github_environment": f"{environment}-database",
+        }
+        assert v1_upgrade["secrets"] == "inherit"
 
 
 def test_stage_9_deployment_shell_script_is_syntax_valid() -> None: