Replace the full-catalog integration tests with selected folders of the installed models

.github/workflows/v2-integration-check.yml

@@ -76,12 +76,8 @@ jobs:
         run: uv run --frozen --extra dev python scripts/reconcile_v2_privileges.py apply --create-missing-roles
       - name: Verify the managed v2 roles hold their declared privileges
         run: uv run --frozen --extra dev python scripts/reconcile_v2_privileges.py verify
-      - name: Verify the installed PolicyEngine.py catalog interface
-        run: uv run --frozen --extra dev coverage run --branch -m pytest -q tests/integration/test_v2_catalog_installed.py
-        env:
-          RUN_V2_CATALOG_COMPATIBILITY: "1"
       - name: Test v2 metadata publication, resource routes and the countries list
-        run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_v2_catalog_publication.py tests/integration/test_v2_metadata_routes.py tests/integration/test_v2_country_routes.py
+        run: uv run --frozen --extra dev coverage run --branch -m pytest -q tests/integration/test_v2_catalog_publication.py tests/integration/test_v2_metadata_routes.py tests/integration/test_v2_country_routes.py
       - name: Test v2 policy persistence and immediate v1 mirroring
         run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_v2_policy_persistence.py tests/integration/test_v1_policy_dual_write.py tests/integration/test_mysql_policy_dual_write.py tests/integration/test_v2_user_policy_mirroring.py tests/integration/test_v1_user_policy_dual_write.py
       - name: Test signed-in identity provisioning and user associations
@@ -92,10 +88,10 @@ jobs:
         run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_v2_household_persistence.py tests/integration/test_v1_household_dual_write.py tests/integration/test_mysql_household_spm_dual_write.py
       - name: Test the row counts v1 ORM updates report on MySQL
         run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_mysql_update_row_counts.py
-      - name: Qualify production-scale v2 metadata publication
-        run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_v2_catalog_publication_qualification.py
-        env:
-          RUN_V2_CATALOG_PUBLICATION_QUALIFICATION: "1"
+      # The real US and UK models, reduced to a few parameter folders: the
+      # complete installed catalog is never loaded in CI.
+      - name: Publish selected parameter folders of the installed PolicyEngine.py models
+        run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q --durations=10 tests/integration/test_v2_catalog_installed_folders.py
       - name: Test real Redis cross-instance semantics
         run: uv run --frozen --extra dev coverage run -a --branch -m pytest -q tests/integration/test_runtime_cache_redis.py
       - name: Export v2 integration coverage

changelog.d/4149.changed.md

@@ -0,0 +1 @@
+CI no longer loads the complete installed PolicyEngine.py catalog. The cross-database integration tests publish a few parameter folders of the installed US and UK models, and a small catalog of edge-case values, and check that every value comes back from PostgreSQL unchanged.

docs/engineering/skills/testing.md

@@ -203,6 +203,24 @@ Otherwise the release run runs them, and `publish-git-tag` requires lint, both
 Alembic checks and the integration tests to pass, as before. The release test
 suite always runs.
 
+CI never loads the complete installed PolicyEngine.py catalog (about 1.2
+million parameter values). Two tests in the cross-database check cover
+publishing it instead, on every run:
+
+- `test_v2_catalog_installed_folders.py` runs the real US and UK models through
+  extraction and publication, reduced to the parameter folders in its
+  `SELECTED_FOLDERS`, and reads every value back. Between them the folders
+  hold every value shape the installed models hold, and the test fails when
+  one is missing: replace a renamed folder with one that holds the same
+  values.
+- `test_v2_catalog_publication.py` publishes
+  `tests/fixtures/v2_catalog_value_shapes.py`, which holds one example of each
+  value shape extraction accepts, and reads every value back. When extraction
+  accepts a new shape, add an example there.
+
+Each release publishes and validates the complete catalog against staging,
+then production, in `seed-v2-database.yml`.
+
 Staging deployment checks run the live integration suite against the tagged
 Cloud Run staging URL before promoting the exact tested revision to the
 service URL. No production deployment may begin until the

docs/migration/stage-9-v2-metadata.md

@@ -158,17 +158,13 @@ regions. Publication took 24.120 seconds and added 7,968,103 bytes of measured
 peak publisher memory. The US fallback summary reported 436 congressional
 districts, 333 places, and 51 states.
 
-Re-run the production-scale test only against disposable Postgres:
-
-```bash
-RUN_V2_CATALOG_PUBLICATION_QUALIFICATION=1 \
-V2_ALEMBIC_DISPOSABLE_TEST=1 \
-V2_MIGRATION_DATABASE_URL="<disposable-postgres-url>" \
-uv run pytest -q tests/integration/test_v2_catalog_publication_qualification.py
-```
-
-Do not point this qualification command at a persistent staging or production
-target.
+The production-scale test that measured this,
+`tests/integration/test_v2_catalog_publication_qualification.py`, was removed
+with PolicyEngine/policyengine-api#4149: CI no longer loads the complete
+catalog. Each release publishes and validates it against staging, then
+production (`seed-v2-database.yml`), and the cross-database integration tests
+publish selected parameter folders of the installed models and a catalog of
+edge-case values (`docs/engineering/skills/testing.md`).
 
 ## Preview verification