Fix SNAP rounding for income limits and minimum allotment

[hua7450]

Summary

PolicyEngine compared unrounded income-to-poverty ratios against 1.3 / 1.0 and paid an unrounded minimum allotment. USDA rounds both: monthly income eligibility standards round up to the next whole dollar, and the minimum allotment rounds to the nearest whole dollar. Households at exactly the published limit therefore failed the tests in PolicyEngine and received $0 instead of a positive benefit.

Legal basis (verified against statute and regulation text)

Change	Authority	Verbatim rule
Gross income limit: ceil(1.3 × monthly FPG)	7 CFR 273.9(a)(3)(i)	"130 percent of the annual income poverty guidelines shall be divided by 12 to determine the monthly gross income standards, rounding the results upwards as necessary."
Net income limit: ceil(monthly FPG)	7 CFR 273.9(a)(3)(ii)	"The annual income poverty guidelines shall be divided by 12 to determine the monthly net income eligibility standards, rounding the results upward as necessary."
Minimum allotment: round(0.08 × one-person max)	7 CFR 273.10(e)(2)(ii)(C); 7 U.S.C. 2017(a)	"The minimum benefit is 8 percent of the maximum allotment for a household of one, rounded to the nearest whole dollar."

The statute (7 U.S.C. 2014(c)) contains no rounding rule for the income standards; the regulation supplies it.

Verification against USDA published values

• FY 2026, household of 3, 48 states: gross limit $2,888 (from $2,887.08), net limit $2,221 (from $2,220.83) — both now reproduced exactly.
• Minimum allotment: FY 2025 $23 (8% × $292 = $23.36 rounds down) and FY 2026 $24 (8% × $298 = $23.84 rounds up) — pinning "nearest" in both directions; ceiling or flooring would each fail one year.
• np.round half-to-even is safe here: 8% of an integer allotment always has a fractional part that is a multiple of $0.04, so an exact 50-cent tie cannot occur.
• The limits pre-round to 4 decimals before the ceiling so float error on an exact whole-dollar standard (e.g., $2,600 for a household of 4 in FY 2025) cannot add a dollar.

Implementation notes

• The income tests now compare monthly income against a rounded dollar limit instead of comparing unrounded ratios. snap_fpg stays unrounded because rounding does not commute with the 130% multiplication and snap_fpg also feeds BBCE.
• The gross-test exemption for households with an elderly or disabled member is unchanged.
• State minimum-allotment overrides (DC, MD, NJ) are unchanged.

Known limitations (documented, not fixed here)

• FY 2023 minimum allotment is $22 in PolicyEngine vs USDA's published $23. USDA applies the statutory 8% to the unrounded thrifty food plan cost ($281.70 → $22.54 → $23), while this implementation applies the regulation's paraphrase to the rounded published maximum ($281 → $22.48 → $22). FY 2023 is the only year in FY 2019–2026 where the two bases diverge; matching it would require storing unrounded TFP costs.
• Households of 9+: the regulation rounds the per-person increment separately before adding it to the eight-person standard; this implementation ceilings the total, which can be $1 stricter (e.g., 10-person FY 2026 gross limit $7,032 vs USDA $7,033).
• BBCE (meets_tanf_non_cash_*) still compares unrounded ratios — same class of issue, left for a follow-up.

Test plan

• Rewrote meets_snap_net_income_test.yaml to drive the formula through snap_net_income and snap_fpg (the ratio variable is no longer read by the formula but remains for BBCE).
• Added boundary tests at the published FY 2026 limits for a 3-person Missouri household (non-BBCE state, monthly period), passing at the limit and failing $1 above.
• Added synthetic round-up cases and the FY 2025/FY 2026 minimum-allotment nearest-dollar cases.
• Changelog fragment included.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (4c6ac7e) to head (74f045a).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main     #9162   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            3         3           
  Lines           49        63   +14     
  Branches         0         3    +3     
=========================================
+ Hits            49        63   +14     

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.