[anth-volk]

Fixes #4013

Summary

Stacked on #4007, which removed the models of the 15 simulation-domain v2 tables and stopped mapping the retired country and label association columns. It kept the tables and columns as schema-only declarations so the metadata still matched the schema. This PR deletes those declarations and adds the revision that drops them. The simulation service owns these tables in its own simulation schema, with its own replacements for all of them except aggregates and change_aggregates (v2 alpha's custom totals, which have no replacement).

Revision 363f34954a14 was generated with uv run alembic -c alembic-v2.ini revision --autogenerate against a disposable PostgreSQL database at fbb7488ebe8a. Autogeneration detected exactly the 15 tables (with their indexes) and the 4 columns. The file was then formatted with ruff format, like the other v2 revisions, and given three post-generation corrections. Each correction is commented in the revision and documented in docs/engineering/skills/alembic-migrations.md; the rule there allows corrections for reversibility or an objectively wrong generated operation:

1. Drop and create order. Alembic emits removed tables in arbitrary order. The generated upgrade dropped reports before report_runs, which references it, and the generated downgrade created output tables before simulations and report_runs, so both failed on foreign keys. The upgrade now drops each table before the tables it references; the downgrade creates them in the reverse order.
2. Enum types. Alembic drops the tables but leaves their 8 native enum types (v2_aggregate_type, v2_decile_type, v2_household_job_status, v2_output_status, v2_report_run_status, v2_report_run_trigger, v2_simulation_status, v2_simulation_type), and the generated downgrade could not create them again while they existed. The upgrade drops them.
3. One check constraint. Reflection strips the outer parentheses of ck_simulations_filter_pair ((filter_field IS NULL) = (filter_value IS NULL)), which left invalid SQL in the generated downgrade. They are restored as the model declared them.

Apart from those corrections, the downgrade statements are exactly the generated ones. The downgrade recreates the tables and columns empty.

Other changes:

• V2_ALEMBIC_HEAD_REVISION moves to 363f34954a14, and the privilege manifest no longer lists the 15 tables.
• The v2 metadata validation rejects the 15 table names.
• New rule api.v2.simulation-tables-stay-removed. tests/unit/v2/test_simulation_domain_boundaries.py fails when any module under policyengine_api, scripts or tests, or the v2 migration environment, imports a removed model or one of the removed modules. That includes imports from the parent package and relative imports. It also fails when the metadata declares one of the tables. The rule has proof fixtures and is rendered into docs/generated/rules.md and AGENTS.md. Rule changes may be approved by any core developer, so this PR requests a review from core-developers.
• Documentation. The Alembic migrations skill gains a section on this revision. The Stage 12A rollback steps say that once this revision has run, a rollback to a pre-Stage 12A release first needs the v2 schema downgraded to fbb7488ebe8a. The Stage 12 worker contract no longer names the dropped tables.

Migration scope: no route group, workflow or user-visible API contract changes. Nothing moves off the Flask/API v1 path. No inline suppressions are added; the one #4007 added goes away with the module it imported.

Environment variables

None. The changed code reads and declares no environment variable. The revision runs in the existing release migration job, and the privilege manifest reaches the deploy through the existing scripts/reconcile_v2_privileges.py step in .github/workflows/seed-v2-database.yml. This PR changes neither, nor their variables.

Merge requirements

Do not merge until all of these hold:

1. Operator step: row counts of the 15 tables in the v2 Supabase database show every table empty in both staging and production. The downgrade recreates the tables empty, so rows dropped here cannot be restored without a backup.
2. Stop using the simulation-domain v2 tables and retired association columns #4007 is merged and deployed to staging and production. The release workflow runs this revision before the new Cloud Run revision serves traffic, so the release serving traffic while it runs must be Stop using the simulation-domain v2 tables and retired association columns #4007's release or later.
3. CI on this PR (it runs in full once Stop using the simulation-domain v2 tables and retired association columns #4007 merges and this PR is rebased onto master) and an approving review, including one from a core developer for the rule change.

If another v2 revision merges first (the planned countries table revision uses the same Alembic chain), this PR is rebased and its revision regenerated on the new head.

Testing

• Against disposable PostgreSQL: alembic -c alembic-v2.ini check reports no new operations at head. tests/integration/test_alembic_v2_lifecycle.py passes, and covers:

  • upgrade to head, then downgrade to fbb7488ebe8a, which restores the 15 tables, the 8 enum types and the 4 nullable columns, then upgrade again;
  • a comparison of the schema the downgrade recreates with the one the earlier revisions created: columns with types, nullability and defaults, keys, unique and check constraints, indexes, and enum values. Changing one check in the downgrade makes it fail;
  • the full base-to-head cycle;
  • head serving the previous application release's use of every table it maps: it reflects fbb7488ebe8a without the dropped tables and columns, then inserts, reads, rewrites and deletes a row in each table, and deletes a user who owns association rows;
  • the same check for the pre-Stage 12A release against fbb7488ebe8a, which the updated rollback steps rely on.

  These tests fail without the revision (Can't locate revision identified by '363f34954a14').

• Also against disposable PostgreSQL and MySQL: test_v2_catalog_publication.py, test_v2_user_identity_persistence.py, test_v2_privilege_reconciliation.py (after reconcile_v2_privileges.py apply and verify), test_v2_stage12_comparison_persistence.py, test_v2_household_persistence.py, test_v2_policy_persistence.py, the cross-database dual-write tests and test_alembic_mysql_lifecycle.py: all passed.

• Unit tests check the revision text: it drops exactly the 15 tables and 4 columns, and recreates them in the downgrade; it has exactly these three corrections; it drops the 8 enum types; and the drop and create order respects every foreign key. Other unit tests check the metadata, the manifest and the rule. uv run pytest tests/unit/v2 tests/contract -q: 1042 passed. The rest of tests/unit and tests/to_refactor: 2331 passed.

• python -m scripts.rules guard --base-ref origin/master fails while the new guard test is unregistered and passes with the registry entry. A temporary module importing Report from the v2 models makes the guard test fail.

• mypy, ruff format --check, ruff check, scripts/run_quality_guards.py and scripts/export_migration_contracts.py (no changes): all passed.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.69%. Comparing base (943601c) to head (c61fbe6).

Additional details and impacted files

@@                              Coverage Diff                               @@
##           v2plan/1h-130-stop-using-simulation-tables    #4014      +/-   ##
==============================================================================
- Coverage                                       87.72%   87.69%   -0.04%     
==============================================================================
  Files                                             257      256       -1     
  Lines                                           15768    15723      -45     
  Branches                                         2616     2614       -2     
==============================================================================
- Hits                                            13833    13788      -45     
  Misses                                           1265     1265              
  Partials                                          670      670              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.