Fixes #620

Summary

• Add ACA take-up values for 2023 through 2026 using APTC plan selections from the CMS open enrollment period (OEP) state-level public use files, not total Marketplace plan selections.
• Compute each rate as APTC plan selections divided by (APTC plan selections + the 2023 tax-credit-eligible uninsured count from KFF).
• Add a regression test for year-specific load_take_up_rate("aca", year) values.

Methodology notes

• Numerator: APTC plan selections summed from the CMS state-level OEP PUFs: 14,797,921 (2023), 19,743,689 (2024), 22,380,137 (2025), 20,035,756 (2026). All four values were re-derived directly from the PUF CSVs and match exactly. Plan selections are used rather than effectuated enrollment so all four years come from the same source; 2026 effectuated data is not yet published.
• Denominator: KFF publishes the tax-credit-eligible uninsured count (8,163,000) for 2023 only, so the same count is reused for 2024–2026. Because later enrollment growth draws people out of that uninsured pool, this likely understates take-up in 2024–2026. This assumption is documented in the parameter file.
• 2021–2022 stay on the 2018 baseline (0.672): the KFF denominator reflects the 2023 post-Medicaid-unwinding uninsured population and does not backcast cleanly to earlier years.

Sources

• KFF State Health Facts: Distribution of Eligibility for ACA Health Coverage Among the Remaining Uninsured (2023)
• CMS: 2023, 2024, 2025, and 2026 Marketplace Open Enrollment Period Public Use Files (state-level PUFs)

Tests

• uv run pytest tests/unit/test_stochastic_variables.py -q (27 passed)
• uv run ruff check tests/unit/test_stochastic_variables.py
• uv run ruff format --check .
• uv run towncrier check --compare-with origin/main
• uv run --no-sync --with pyyaml python scripts/run_quality_guards.py
• git diff --check

🤖 Generated with Claude Code