Statewide ABAWD waiver map (FNS FY2025 Q1) + Sec. 10102 discretionary-cut parameter

[daphnehanse11]

Standalone on main — previously stacked on #8961, but rebased off it after main merged the SNAP ineligible-member income proration (snap-ineligible-member-income). This PR now contains only the three ABAWD waiver-map commits and none of #8961's SNAP work-requirement income changes. On the income-counting question that separated the two: 7 CFR 273.11(c)(2) (and the explicit cross-reference in 273.24) prorates the income of an ABAWD-time-limit-ineligible member, which is exactly what main now implements — so this PR adopts main's proration rather than #8961's full-count.

The in-model ABAWD waiver surface was 29 Alaska county FIPS — but the FNS FY2025 Q1 status report shows nine statewide waivers as of October 1, 2024 (CA, DE, DC, GU, IL, NV, NM, NY, VI) plus 22 partial-waiver states. CA+NY+IL alone are roughly a third of the national caseload, fully shielded pre-HR1. With no waiver map, HR1's waiver terminations — the largest single driver in CBO/CBPP loss estimates — scored near zero in release-scale reform runs (#8961 discussion).

• New waived_states parameter (postal-code-keyed; state_fips is an input defaulting to 6 and cannot key policy), FY25-Q1-sourced, ending at P.L. 119-21 Sec. 10102 enactment with the ran-to-expiration simplification documented.
• is_in_snap_abawd_waived_area resolves statewide OR county waivers.
• Sec. 10102's discretionary-exemption cut (8% → 1%) recorded on discretionary_exemption_rate, with a NOTE that runtime consumption requires a persisted seed quantile on the dataset (populace-side follow-up).
• Partial-waiver states (22) are a documented follow-up — their sub-state area lists live in individual waiver approvals, not the quarterly report.
• Two existing ABAWD tests pinned to TX (the default-CA household became waiver-exempt).

Caveat: the partial-state county map rides on imputed county geography

Populace datasets impute each household's county via the geography ladder (CPS does not identify counties for most households). County assignment is validated for presence and state-consistency, but has not been validated against county-level benchmarks (populace#241/#292). The exposure decomposes cleanly in release-scale runs (Build K × #8961, 2026 HR1 repeal):

• ~$4.0B of the $4.18B/yr estimate keys on statewide waivers or non-geographic channels — state is observed, so no county risk.
• Only the ~$150M partial-state increment depends on imputed counties, and most of it is robust to within-state misassignment because the big partial states are lopsided (MI 82/83 counties waived, KY 117/120, WA 38/39, AZ 14/15 — a misassigned household almost always lands in another waived county).
• The genuinely county-sensitive slice is the even-split states — CO (18/64), MN (15/87), VA (12/133) — together roughly $80M.

Any published use of the partial-state increment should carry this caveat until county assignment gets a benchmark (populace#292).

Tests: 286 passed across the SNAP work-requirements/income suites rebased on main (main's proration tests and the new waiver tests pass together); new is_in_snap_abawd_waived_area.yaml covers statewide, never-waived, post-HR1, and county cases.

🤖 Generated with Claude Code

[daphnehanse11]

Second commit lands the partial-state county map: 436 county FIPS across 16 states at full-county grain, each state's list read from its FNS approval letter (references on the parameter; five research passes over the approval PDFs, verified area-by-area). Sub-county waivers (ME's 213 towns, CT's 68, RI/NH towns, MT's reservation-only waiver) are deliberately omitted and documented as an understatement — county-grain inclusion would overstate them.

Release-scale validation (Build K artifact × #8961, 2026, repeal counterfactual restoring the pre-HR1 waiver map by reading the parameters' own 2025-01 vintages):

waiver map	HR1 provisions' effect	households affected
none (29 AK counties)	$3.06B/yr	1.29M
+ 9 statewide	$4.03B/yr	1.63M
+ 436 partial-state counties	$4.18B/yr	1.71M

Largest state effects: NY $1.04B, CA $764M, PA $297M, FL $286M, OR $145M, MI $119M, WA $112M. Dataset county coverage verified (e.g., KY records span 106 distinct counties), so the map binds geographically; the moderate partial-state increment reflects the model's remaining exemption shields (annual-hours proxy, discretionary seeds at the pre-HR1 8% cap), tracked separately.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (db118db) to head (4abbeea).
⚠️ Report is 34 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main     #9030   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            1         1           
  Lines           17        15    -2     
=========================================
- Hits            17        15    -2     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

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.