[anth-volk]

Fixes #4287

Summary

The live staging household checks added in #4274 failed in the release run on cd513dacc (push.yml run 43075278725), so no release of the Public API reaches production until they pass. Two parity cases failed, for two different reasons:

• us-single-california: /us/calculate answered 400, "Household variable income_tax belongs on tax_units, not people". policyengine_us defines income_tax on the tax unit, and /calculate has refused a variable on the wrong entity since 3a59b4d14. The simulation service's household function has the same check. The household now asks for income_tax on its tax unit.
• us-single-reform-added-variables: httpx.ReadTimeout on /us/calculate-full after 90 s, the test client's read timeout. The candidate's request log shows that request answered 200 after 223.8 s. Production's /calculate-full requests from the past week took between 50 s and 296 s, and some got Cloud Run's 504 at its 300 s request timeout. The parity check now waits 320 s (V1_CALCULATION_TIMEOUT_SECONDS) for /calculate and /calculate-full. That is longer than CLOUD_RUN_TIMEOUT, so the check gets the candidate's own answer.

The parity check also reads the v2 job's status before calling /calculate. A failed job is then reported without waiting minutes for the Public API's own calculation.

Known limit: the added-variables case now passes only if staging's /calculate-full answers within Cloud Run's 300 s request timeout. Its one staging measurement is 224 s; production has answered in up to 296 s, and has sometimes given a 504. A slower staging run would block a release on the Public API's own latency rather than on a parity difference.

Tests

• New tests/unit/v2/test_live_household_parity_requests.py, committed failing first:
  • test_the_public_api_accepts_each_parity_household sends each of the six parity households, as calculate_with_v1 sends it, to the real /calculate and /calculate-full routes, registered on a test Flask app with the installed US and UK models. The routes check the inputs and parse the situation as in the Public API; only the calculation is replaced, by an answer that returns the household as the routes prepared it. It failed on us-single-california with the staging message.
  • test_the_check_waits_longer_than_the_candidate_lets_a_request_run requires the read timeout of both requests to exceed CLOUD_RUN_TIMEOUT, read from .github/scripts/cloud_run_env.sh. It failed with the client's own timeout.
• After the fix, the 8 new test cases pass, each run by node ID.
• Review: two rounds, no bugs. Round 1 found that the unit test skipped the routes' situation parsing, so it now goes through the real routes; round 2's wording findings are fixed.
• alembic -c alembic-v2.ini check on a disposable PostgreSQL 17: no new operations, one head 0c136a278f29. lint-imports, the rules registry, the quality guards, ruff, ruff format, shellcheck, uv lock --check and mypy pass.

Environment variables and secrets

The change reads no new variable. run_live_staging_tests.sh changed only in a comment. The live step receives API_BASE_URL, CI_SERVER_ACCESS_TOKEN and STAGING_API_TEST_PROBE_ID through its env:; the failed run's log shows all three set. The new unit test reads CLOUD_RUN_TIMEOUT from cloud_run_env.sh's default, which no job in push.yml overrides.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.74%. Comparing base (cd513da) to head (46cf2f4).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4288      +/-   ##
==========================================
+ Coverage   92.69%   92.74%   +0.05%     
==========================================
  Files         329      329              
  Lines       18783    18783              
  Branches     3114     3114              
==========================================
+ Hits        17410    17420      +10     
+ Misses        818      809       -9     
+ Partials      555      554       -1     

Flag	Coverage Δ
integration	53.77% <ø> (ø)
unit	90.50% <ø> (+0.05%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.