Skip to content

Bump PolicyEngine US for data builds#987

Merged
MaxGhenis merged 1 commit into
mainfrom
codex/pe-us-8298-data-refresh
May 14, 2026
Merged

Bump PolicyEngine US for data builds#987
MaxGhenis merged 1 commit into
mainfrom
codex/pe-us-8298-data-refresh

Conversation

@MaxGhenis
Copy link
Copy Markdown
Contributor

Summary

  • require PolicyEngine US 1.691.10 for data builds
  • resolve PolicyEngine US from exact upstream commit 4fd79e6608bc2dac3a7fde0be37191cb4870bd85 while PyPI publication is blocked
  • add a changelog note for the dependency bump

Why

PolicyEngine US PR #8298 fixed vectorized calculations that affect SNAP disability eligibility, Medicaid medically needy categories, TANF non-cash asset tests, and capital-gains flags. The latest PyPI release remains 1.691.3; PE-US 1.691.10 built but failed to upload because the PyPI project hit the 10 GB storage limit. I filed PolicyEngine/policyengine-us#8316 for that release blocker.

On the current enhanced_cps_2024.h5, comparing PE-US 1.691.3 to upstream 1.691.10 for 2024 shows the old vectorization bug was material before any recalibration:

  • is_usda_disabled weighted mean: 1.000 -> 0.126
  • has_qdiv_or_ltcg weighted mean: 1.000 -> 0.182
  • SNAP total: $109.4B -> $101.0B (-7.7%)
  • SNAP eligible SPM units: 36.7M -> 34.0M (-7.3%)

That means the ECPS should be regenerated/recalibrated against the fixed model, not only rescored downstream.

Tests

  • uv sync --extra calibration
  • uv run pytest tests/unit/test_import.py tests/unit/test_enhanced_cps.py -q
  • uv lock --check

@MaxGhenis MaxGhenis force-pushed the codex/pe-us-8298-data-refresh branch from e769786 to 8ecaf0b Compare May 14, 2026 20:10
@MaxGhenis MaxGhenis merged commit 610b76c into main May 14, 2026
12 checks passed
@MaxGhenis MaxGhenis deleted the codex/pe-us-8298-data-refresh branch May 14, 2026 20:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant