Populate CPS inputs for SPM childcare formula

[MaxGhenis]

Summary

• carry CPS WKSWORK through the microdata build as weeks_worked
• stop rebuilding spm_unit_capped_work_childcare_expenses inside policyengine-us-data
• leave capped childcare to the model formula in policyengine-us and add regression coverage for the new input plumbing

Validation

• uv run pytest -q tests/unit/test_extended_cps.py tests/unit/test_weeks_worked.py
• uv run ruff check policyengine_us_data/datasets/cps/extended_cps.py tests/unit/test_extended_cps.py tests/unit/test_weeks_worked.py
• git diff --check

Notes

• On raw CPS 2024, SPM_WKXPNS is highly reproducible from other CPS inputs: MAE is about $12, 97.9% of units are within $1, and 99.1% are within $5.
• SPM_CAPWKCCXPNS is not reproducible nearly as cleanly from current public CPS inputs, so this PR intentionally stops short of reconstructing the capped value in us-data.
• Paired model PR: Add Census SPM work expense formulas policyengine-us#7960.

[baogorek]

I realize this is an early review and we're waiting on policyengine-us#7960, but I wanted to get some thoughts in here. Interesting that the tests are failing on state-level calibration of aca. (I need to add that to my scorecard.)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
pipeline-diagrams	Error		Apr 11, 2026 0:57am

[MaxGhenis]

Addressed the review comments in 27169f8:

• Replaced the brittle source-text tests with behavioral tests that call the CPS extraction functions.
• Added a named PERRP unmarried-partner code mapping with Census CPS ASEC 2024 technical documentation cited in code.
• Kept the WKSWORK test focused on generated weeks_worked output and clipping behavior.

Local checks:

• uv run pytest -q tests/unit/test_reference_partner.py tests/unit/test_weeks_worked.py
• uv run pytest -q tests/unit/test_extended_cps.py
• ruff check tests/unit/test_reference_partner.py tests/unit/test_weeks_worked.py

Note: ruff check policyengine_us_data/datasets/cps/cps.py still reports pre-existing lint debt in that module (star import/unused locals), not introduced by this patch.

[MaxGhenis]

Follow-up for the unit-test failure on the new run: 89b58ac extracts derive_weeks_worked and tests that helper directly, while production still assigns cps["weeks_worked"] = derive_weeks_worked(person.WKSWORK).

The CI failure came from the test calling the full add_personal_income_variables path after package-spec state had been changed elsewhere in the full unit suite. Local full unit verification now passes with CI-style env:

• HUGGING_FACE_TOKEN=dummy uv run pytest -q tests/unit -> 589 passed, 9 skipped

[MaxGhenis]

Integration follow-up: the Modal run failed in tests/integration/test_cps.py::test_add_personal_variables_maps_current_health_coverage_flags because that synthetic fixture does not include PERRP.

I pushed a narrow fix in this commit:

• default missing PERRP to all-false when building is_unmarried_partner_of_household_head
• added a unit test covering the missing-column fallback

Local verification:

• uv run pytest -q tests/unit/test_reference_partner.py tests/integration/test_cps.py -k 'missing_perrp_defaults_to_false or add_personal_variables_maps_current_health_coverage_flags'