Skip to content

Correct countable income for school meals - #10873

Merged
MaxGhenis merged 9 commits into
mainfrom
school-meal-countable-income
Nov 16, 2026
Merged

MaxGhenis merged 9 commits into
mainfrom
school-meal-countable-income

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Correct school meal countable income under 7 CFR 245.6(a)(5) and USDA's 2026–27 income notice.

Exclude educational assistance, net each person's business, rental and estate income within its category before flooring it at zero, and count omitted cash income. Count state supplemental cash assistance once per recipient or assistance unit.

USDA instructs administrators through its Eligibility Manual to account for cash, business losses, educational aid and noncash payments on printed pages 26, 28, 30–31 and 33. Administrators of federally funded benefits must disregard Title IV aid under 20 USC 1087uu.

Estate income can include business and rental losses passed through estates and trusts. The IRS describes those amounts in boxes 6–8 of its Schedule K-1 instructions.

We give federal CSFP, Texas CSFP and California CVRP separate income measures. We retain their previous source sums and existing approximations, and defer substantive corrections under each program's income definition:

  • State CSFP agencies calculate gross income, apply statutory exclusions and choose permitted state options under 7 CFR 247.9.
  • Texas CSFP administrators accept qualifying program participation for adjunctive eligibility and exclude SSI under the Texas Department of Agriculture's April 2026 handbook, pages 13–14.
  • CVRP administrators count educational assistance, apply Schedule 1 loss rules and use adult household income under the California Air Resources Board's Clean Vehicle Rebate Project implementation manual, pages 12–15 and 38.

Closes #10801

School meal income now counts estate and trust income, which #10063 raises for other federal benefit lists. The Manual's printed page 26 counts railroad retirement with retirement Social Security, and the school meal list still leaves out railroad_benefits, which #10551 tracks.

Draft #10766, "Count Head Start and school meal values from the first qualifying month," changes school meal tiers.

Invariants

  • Each person's net income in each business category contributes at least zero. No loss reduces another category or another person's income.
  • Holding other counted amounts fixed, educational assistance contributes zero to school meal countable income.
  • Holding other counted amounts fixed, an additional dollar of included cash raises countable income by one dollar.
  • With educational assistance and every newly counted source at zero, nonnegative original sources retain their previous sum. TANF and assistance-unit supplements enter once.
  • Federal CSFP, Texas CSFP and CVRP retain their previous source sum for raw household income inputs, including signed business amounts and educational assistance.
  • Oregon's excluded placement and facility payments stay out of the annual school meal income sum.

test_school_meal_countable_income.py checks the first five invariants with Hypothesis, on six generated examples per property. Its loss-floor property draws one to four SPM units of one to three people, each with signed amounts in every business source. Its cash property raises one of the five newly counted cash inputs: alimony, workers' compensation, disability benefits, military retirement pay and military survivor payments. The school income YAML state vector checks assistance-unit supplements counted once.

Compared with main, school meal income falls for households with educational assistance, and for a person whose farm rent loss offsets their rental income. It rises for households with other business losses, newly counted cash or state supplements.

Microsimulation impact

We estimate that this branch lowers free and reduced-price school meal subsidies (school_meal_net_subsidy) by $303.1m in 2026 on the default dataset, and household benefits and net income by the same amount. The subsidy changes for 160 sampled SPM units (242,418 weighted): it falls for 126 and rises for 34. The value of free meals falls by $309.7m across 100 changed SPM units (138,949 weighted), and the value of reduced-price meals rises by $6.6m across 78 (134,042 weighted). The income the model tests against the meal limits, school_meal_countable_income, rises by $1,321.2bn across 10,985 changed SPM units (28,307,346 weighted): it rises for 8,143 and falls for 2,842. We sum that income over every SPM unit, with or without a child in school. CSFP, Texas CSFP and CVRP eligibility and federal and state income tax don't change. We compared main at 0ae3d78441 with this branch at 72c30a762a.

Output Main total Change Records changed (weighted)
school_meal_countable_income $18,477.8bn +$1,321.2bn 10,985 (28,307,346)
free_school_meals $35.40bn -$309.7m 100 (138,949)
reduced_price_school_meals $1.52bn +$6.6m 78 (134,042)
school_meal_net_subsidy $36.92bn -$303.1m 160 (242,418)
commodity_supplemental_food_program_eligible 8,474,800 people 0 0 (0)
is_ca_cvrp_increased_rebate_eligible 23,107,593 people 0 0 (0)
tx_dta_csfp_income_eligible 4,875,121 people 0 0 (0)
state_income_tax $576.6bn $0 0 (0)
income_tax $2,594.9bn $0 0 (0)
household_benefits $2,060.1bn -$303.1m 160 (242,418)
household_net_income $17,246.4bn -$303.1m 160 (242,418)

Validation

Pull request run 43891208201 at 72c30a762a passed all 37 jobs on GitHub's merge with main fd0a748698.

File Cases What it checks
school_meal_countable_income.yaml 38 Issue households, each newly counted source, loss floors at −1, 0 and 1, losses and gains across people and categories, same-person SSTB and rental netting, state supplement totals
or_ssp_countable_for_school_meals.yaml 7 Oregon's excluded placement categories and their annual sum
csfp_countable_income.yaml 6 Federal CSFP, Texas CSFP and CVRP measures keep educational assistance, signed losses and their previous omissions
commodity_supplemental_food_program_eligible.yaml and commodity_supplemental_food_program.yaml 21 and 6 Federal CSFP eligibility and benefit read the CSFP measure
tx_dta_csfp_income_eligible.yaml 2 Texas CSFP reads its own measure
is_ca_cvrp_increased_rebate_eligible.yaml 6 CVRP reads its own measure
Kansas CSFP integration.yaml and ks_dcf_csfp_categorically_eligible.yaml 16 and 2 Kansas CSFP reads the federal CSFP measure
test_school_meal_countable_income.py 5 properties Invariants 1–5

The property file runs in the Rest Python core group. In Rest shard 7 of CI run 43655281957, at c19c796174, each of the five properties took 0.6 to 1.1 seconds.

Methodology choices

  • Retain the existing SPM unit and annual-income framework, with main's school meal household membership and recertification timing.
  • Exclude the existing educational assistance aggregate. PolicyEngine has no input that separates excluded aid from assistance available for meals.
  • Net each person's gains and losses within each of five income categories, then floor each category at zero: nonfarm self-employment, rental income, farm operations, partnership and S corporation income, and estate and trust income.
  • Count Schedule C income from specified service businesses and from other businesses as one nonfarm self-employment category, through total_self_employment_income. USDA's notice lists "(2) net income from nonfarm self-employment" as one item, and the Manual's printed page 27 credits the self-employed "with net income rather than gross income."
  • Net farm_rent_income with rental_income in school_meal_net_rental_income, rather than with farm operations. USDA's notice lists "(6) net rental income" apart from "(3) net income from farm self-employment."
    • The Manual's printed page 26 lists "Net rental income" apart from "Net income from self-owned business, including farms."
    • On printed page 28, the Manual counts "Money received from the rental of land, buildings, or equipment to others" among a self-employed person's gross receipts, so the entry covers a business that rents out property.
    • farm_rent_income holds Form 4835 rent that a landowner receives from a farm they don't materially participate in. The Manual names no farm rent, so we treat it as rental income.
  • Net partnership and S corporation components within their existing aggregate before the floor, and floor estate_income as entered, so both variables keep their meaning.
  • Keep partnership and S corporation income as its own category, apart from nonfarm self-employment. USDA's notice names neither partnerships nor S corporations, and partnership_s_corp_income combines both.
  • Count modeled state supplemental cash assistance as the Manual's "Regular cash assistance from State or local governments" (printed page 26). Under 42 USC 1382a, Congress excludes state need-based aid from SSI income without extending that exclusion across federal programs.
  • Leave one question open for maintainers. On printed page 30 of the Manual, USDA adopts "the income exclusions provided under the Social Security Act and other Federal laws", and in section 1382a(b)(6) Congress excludes need-based state assistance from SSI income. If that exclusion carries into school meals, administrators leave need-based state supplements out of income. The source-conflict candidate us:school-meal-state-ssi-supplements records both readings with state open.
  • Use top-level state cash totals, including Colorado OAP and broader aged, blind and disabled assistance. Add no child component separately when a total already includes it.
  • Exclude Oregon initial placement and absence payments under 461-155-0500 and 461-155-0630. Treat incapacity placement payments as unavailable cash while the recipient awaits a representative. Count other recurring cash allowances.
  • Preserve ancillary programs' existing income approximations through their own parameters, and defer their substantive income corrections to program-specific follow-ups. An input of school_meal_countable_income or school_meal_fpg_ratio no longer changes federal CSFP, Texas CSFP or CVRP results.
  • Retain the existing source list's 2009 effective date for the new lists.
  • Use *_before_lsr earnings inputs and singleton marital units in YAML to match the issue households and US Simulation's input normalization. Set school meal income separately in ancillary routing fixtures to check each program's own measure. Use Core's entity builder and canonical earnings for generated income checks.

axiom: TheAxiomFoundation/rulespec-us#2293 queued (7 CFR 245.6(a)(5) and the income definition at 91 FR 17933; the corpus lacks 7 CFR part 245, that notice and the Eligibility Manual, so the signed encoder can't run yet)

🤖 Generated with Claude Code

MaxGhenis and others added 4 commits November 13, 2026 08:23
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Floor estate and trust pass-through income before adding it to household income. Check that ancillary income measures retain their original sums with generated households.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Use earnings before labor supply responses and explicit household groups in the issue cases. Check Oregon exclusions through the annual income sum and reuse one simulation for ancillary income assertions.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Build generated households with Core's entity builder and canonical earnings inputs.
Set independent school meal inputs in the CSFP and CVRP fixtures.
Match Oregon's state vector to nine recipients.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MaxGhenis and others added 3 commits November 14, 2026 14:51
Resolve source_conflicts/candidates.yaml: keep main's new South Dakota child
care entry (#10815) and append the school meal entry after it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… meal floor

USDA's income definition lists net income from nonfarm self-employment and net
rental income as single items, and its Manual credits the self-employed with
net income. PolicyEngine now floors each person's Schedule C and SSTB income
once, through total_self_employment_income, and nets rental income with farm
rent in school_meal_net_rental_income before the floor. Two new cases cover a
same-person SSTB loss and a same-person rental loss, and the property test
groups its expectation the same way.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n the supplement candidate

USDA names no state supplement, so the source list now says PolicyEngine counts
these totals as regular cash assistance from State or local governments. The
state SSI supplement candidate returns to open, and its note states what the
code counts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 14, 2026 21:39
MaxGhenis and others added 2 commits November 14, 2026 18:51
The formula bound `income` to the array `add()` returns and then wrote into it with `+=` twice. `add()` returns its first variable's array unchanged when a source list holds one variable on the same entity, and the code-health guard `test_no_in_place_writes_to_cached_arrays` fails on both lines.

Sum the two source lists in one expression and rebind `income` in the loop, so the formula builds new arrays. No expected value changes.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Main added source-conflict candidates beside this branch's entries. Keep main's
entries in main's order and put this branch's entries after main's last entry
for the same jurisdiction.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

US + core hub merge audit, us#10873 at 72c30a762a8931933c810a8357a9fd73a179a55f (squash): Correct countable income for school meals.

  • Reviews:
    • The PR has no GitHub reviews. Three independent review rounds ran outside GitHub, with a fix round after the first and after the second. A merge of main followed the third round, and no review read it. Of this PR's 27 files, that merge changed the source-conflict register alone, and a script checked the merged register.
    • The first round, at a590e468d8, read 7 CFR 245.2 and 245.6(a)(5), USDA's income notice at 91 FR 17932, six pages of USDA's Eligibility Manual and 20 USC 1087uu. It ran 24 households on the head and on the merge base a93cc58d7e, and its hand figures agreed with the PR's expected values. CSFP, Texas CSFP and CVRP returned the same results on both.
    • It requested three changes. The PR floored a person's two Schedule C amounts separately, so a Tennessee parent with $30,000 of wages, a $10,000 gain in one and a $10,000 loss in the other counted $40,000. USDA lists nonfarm self-employment as one item, which gives $30,000. The description had no population estimate, and CI had skipped every job on the draft, so no test had run at that head.
    • It also listed minor findings. Two parameter descriptions said USDA counts the state supplement totals and partnership and S corporation income. USDA names no state supplement, partnership or S corporation. The register entry for state supplements read ruled. The PR excludes educational assistance in full, where USDA excludes Title IV aid and aid that a household can't use for meals.
    • The fix round, through bef34c0691, replaced the two Schedule C entries with total_self_employment_income. It added school_meal_net_rental_income, which nets a person's rental income with their farm rent before the floor. It added a case for each: $30,000 for the Schedule C household, and $32,000 for wages of $30,000 with a $4,000 rental loss and $6,000 of farm rent. Each of four mutants failed a case. The round rewrote both descriptions with PolicyEngine as the subject and set the register entry to open.
    • The first of two merges of main came in that round. In 678b352a00, Pay South Dakota CCA from the July 2024 hourly county rates #10815 had added a register entry at the same place as this PR's, and the merge keeps both.
    • The second round, at bef34c0691, found each first-round finding on the law fixed, and it recomputed every changed and new expected value by hand. It read the merge hunk by hunk and fetched the notice and the Manual again. It requested one change: CI at that head failed the guard against writes into cached arrays, because the formula wrote into the array that add() returns.
    • It listed two corrections to the description. A person whose farm rent loss offsets their rental income counts less than on main, and the description said income rises for households with business losses. The description also didn't say that partnership and S corporation income stays its own category.
    • The next fix, c19c796174, sums the two source lists in one expression and rebinds income in the loop. It changes one file and no expected value.
    • The third round approved c19c796174 with no blocking finding. It read the commit against Core's add(). CI run 43655281957 at that head passed all 37 jobs on main at 7c516c74bb, including the guard, the five properties and every changed YAML file: 38, 7, 6, 21, 6, 2, 6, 16 and 2 cases. It also fetched the Texas Department of Agriculture's CSFP handbook and checked the description's page references against it.
    • It listed two new minor findings. The description cited a local timing for the property file, where CI ran each property in 0.6 to 1.1 seconds. The Texas CSFP and CVRP source lists title their links without the agency's name, which stays open under Next.
    • The reviews listed corrections to the description, and I made each before merge. The description now states which households count less than on main, names the partnership and S corporation category, gives the property file's CI time and describes the open register question under Methodology choices. It also uses an accepted axiom: form.
    • After the third round, 72c30a762a merged main at 0ae3d78441, the branch's second merge of main. Correct 2018–2020 Medicaid parent limits in UT, TN and VA and adult-group starts in UT and VA #10874 had added three entries at the end of source_conflicts/candidates.yaml, where this PR's entry sat, so git's merge conflicted there and in no other file. A script resolved that file. It kept main's 478 entries in main's order and put this PR's entry after main's last entry with a us: id. A separate script confirmed that the merged register holds 479 unique ids and that each entry equals its own side's text. Git merged every other file without conflict, and this PR's other 26 files match c19c796174.
    • CI at c19c796174 tested main at 7c516c74bb, which already held Apply SSI's general exclusion once under spousal deeming #10837's SSI deeming change and Finish South Carolina TANF boundary and source follow-ups #10829's South Carolina TANF change. Since 7c516c74bb, main's side of the merge changed inputs that this PR's code reads or reads through, and outputs that its cases test. Read monthly SSI support in Maryland and California benefits #10809 reads each month's SSI in-kind support in ca_state_supplement_person.py and md_paa_countable_income.py, which feed two of the state totals. Correct historical Colorado Works grant amounts and dates #10892 changes Colorado's TANF grant standards from September 2018 through June 2022, and school meal income counts tanf. Use exact 2016 HHS poverty guideline tables #10886 gives spm_unit_fpg HHS's exact table for 2016, and this PR's CSFP, Texas CSFP and CVRP formulas divide by that variable, as school_meal_fpg_ratio does. Count strike benefits other than gifts in federal gross income #10727 added documentation to strike_benefits, a counted input, and changed no value. Limit school meal categorical eligibility to each qualifying child #10838 takes child-specific categories out of meets_school_meal_categorical_eligibility and edits school_meal_tier's documentation. This PR's 23 cases that test that variable expect false, and no case enters a child-specific category, so Limit school meal categorical eligibility to each qualifying child #10838 can't change them. No case computes California's or Maryland's total for an SSI recipient, the state case enters TANF as zero for both Colorado households, and the cases run for 2019, 2022, 2024, 2025 and 2026.
  • Methodology choices for maintainers:
    • The PR counts 46 state cash totals as the Manual's "Regular cash assistance from State or local governments" (printed page 26). source_conflicts/candidates.yaml records a competing reading as us:school-meal-state-ssi-supplements, with state open. On printed page 30, USDA adopts the Social Security Act's income exclusions, and in 42 USC 1382a(b)(6) Congress excludes need-based state assistance from SSI income. A maintainer who carries that exclusion into school meals would take need-based supplements out of state_supplement_sources. If Wisconsin's supplement falls under it, the Wisconsin case would expect $50,042 and reduced-price meals in place of $51,114.40 and paid meals. The question stays open.
    • The PR excludes the whole educational_assistance input, since the model has no input that separates excluded aid from aid a household can use for meals. USDA excludes Title IV aid and aid unavailable for meals, so the model undercounts income for a household with other aid.
    • The floor applies to each person in each of five categories. A person's Schedule C gains and losses net across both Schedule C inputs, and rental income nets with farm rent. Partnership and S corporation income net with each other and stay apart from Schedule C income, since USDA names neither. A maintainer who nets farm rent with farm operations would change a household with a farm operations loss and farm rent.
    • Oregon's room and board payments at an initial placement, during an absence and during incapacity stay out of the state total. Under OAR 461-155-0630, the Department may authorize the first for the month of admission, pay the second to the care facility and authorize the third to secure a placement.
    • CSFP, Texas CSFP and CVRP now read their own income measures, each with main's source list, so their results don't change. An input of school_meal_countable_income or school_meal_fpg_ratio no longer reaches them.
    • The new lists start in 2009, like the list they replace. The PR cites USDA's notice for July 2026 through June 2027 and the 2017 Manual for every year.
    • School meal income still follows the SPM unit and the calendar year. USDA's rule reads the month before the application, or a projected annual amount.
  • Gates checked live at merge:
    • gh pr checks exits 0;
    • the latest Pull request run at the head, 43891208201, passed and tested this head merged into main at fd0a748698;
    • MERGEABLE;
    • not a draft;
    • no reviewer's latest review requests changes;
    • the semantic-overlap guard: since fd0a748698, main has changed no area this PR touches. This head merged main at 0ae3d78441. In source_conflicts/candidates.yaml, that merge kept main's 478 ids in main's order and this PR's one added id, us:school-meal-state-ssi-supplements. Between 0ae3d78441 and main at fd0a748698, main changed 280 files in 115 commits, and source_conflicts/candidates.yaml alone among this PR's 27 files. Main added five entries there and edited one, and none of the six names school meals, CSFP or CVRP. Merged with main at fd0a748698, the register holds 484 unique ids, main's 483 and this PR's one. Under states/ks and states/or, Correct Kansas's 2015-2017 child and dependent care credit #10968 set Kansas's dependent care credit match to zero for 2013 to 2017, and Correct Oregon EITC match rates for 2015–2019 #10954 corrected Oregon's EITC match rates for 2014 to 2019. This PR's Kansas and Oregon cases run for 2024 to 2026. Main changed no file under usda/school_meals, usda/csfp, states/ca or states/tx. The branch merges into main at fd0a748698 without conflict;
    • the merge pins the head.
  • Impact: we estimate that this branch lowers free and reduced-price school meal subsidies by $303.1m in 2026 on the default dataset, for 160 sampled SPM units (242,418 weighted): subsidies fall for 126 and rise for 34. Household net income falls by the same amount, and school_meal_countable_income, the income the model tests against the meal limits for every SPM unit, rises by $1,321.2bn across 10,985 changed SPM units (28,307,346 weighted). Main 0ae3d78441 against 72c30a762a.
  • Next:
  • Axiom: Encode school meal countable income: counted cash sources, each person's business loss floors and excluded educational aid (7 CFR 245.6(a)(5); 91 FR 17933; USDA Eligibility Manual; 20 U.S.C. 1087uu) (PE-US #10873) TheAxiomFoundation/rulespec-us#2293 queued, a new pe-parity issue for 7 CFR 245.6(a)(5), USDA's income definition at 91 FR 17933, the Eligibility Manual's income rules and 20 USC 1087uu, with companion rows from this PR's YAML. The signed encoder can't run yet: the corpus holds no 7 CFR part 245, FR Doc. 2026-06842, Eligibility Manual or 20 USC 1087uu.

🤖 Generated with Claude Code

@MaxGhenis
MaxGhenis deleted the school-meal-countable-income branch November 16, 2026 16:21
MaxGhenis added a commit that referenced this pull request Nov 16, 2026
Resolve the conflict with #10873 in commodity_supplemental_food_program.yaml. Each of the six cases keeps this branch's take-up flag and main's csfp_countable_income and spm_unit_fpg inputs, with its name, period, age and expected payment.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MaxGhenis added a commit that referenced this pull request Nov 16, 2026
After #10873, CSFP eligibility divides csfp_countable_income by spm_unit_fpg, and an input of school_meal_fpg_ratio no longer reaches it. Two take-up cases and the property test still set that ratio. The property test's third anchor, a 70-year-old with a ratio of 100, then qualifies on computed income, which fails the test's check that the anchor stays ineligible.

Set csfp_countable_income and spm_unit_fpg in place of the ratio. In the property test, each person's income equals the drawn ratio times a 10,000 guideline, so each person stays on the same side of the limit. The two YAML cases set 15,000 against 10,000, the 2025 limit of 1.5, and keep their expected outputs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant