Skip to content

Add a school meal take-up input - #10834

Merged
MaxGhenis merged 4 commits into
mainfrom
add-school-meals-take-up
Nov 16, 2026
Merged

MaxGhenis merged 4 commits into
mainfrom
add-school-meals-take-up

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Add takes_up_school_meals_if_eligible so users and datasets can decline computed school-meal benefits. The input defaults to true, preserving existing amounts when callers omit it. Callers can supply false to model nonparticipation.

When a household tells the local educational agency that it doesn't want free or reduced-price benefits, its children "must have their benefits discontinued as soon as possible" under 7 CFR 245.6(c)(6)(iii).

School-meal totals and USDA participation

In the prior 2025-rules analysis, we estimated $31.670 billion in free school meals and $1.680 billion in reduced-price meals, or $33.350 billion combined.

USDA publishes both costs and participation through FY2025 in its October 9, 2026 tables. USDA reports these federal costs:

FY2025 spending Billions of dollars
National School Lunch Program cash payments 17.236
School Breakfast Program cash payments 6.069
School-food commodities 1.660
Combined, excluding Special Milk cash 24.965

Our prior combined estimate exceeds that amount by $8.385 billion, or 33.6%. USDA reports average daily participation of 30.0 million for lunch and 15.7 million for breakfast.

USDA averages participation across nine school months, excluding summer. The comparison spans annual modeled benefits and federal fiscal-year spending; USDA includes paid-meal reimbursements and commodities. USDA excludes federal support for state administration and breakfast start-up costs.

The gap does not identify a take-up rate or measure this PR's impact. Microcosm should supply participation assumptions through the new input. This PR retains full take-up defaults and leaves the modeled number of school days unchanged.

Invariants and validation

For households whose school-meal amounts the model computes from current-law rates:

  • Omitted take-up and explicit true reproduce main's amounts for every household.
  • False yields zero computed net, free and reduced-price benefits.
  • Take-up leaves school-meal eligibility, per-meal tiers and potential daily subsidies unchanged.
  • Under main's rates, benefits remain nonnegative, and participation never lowers them.
  • Free and reduced-price amounts sum to the computed net subsidy.

All 44 YAML cases passed:

File under policyengine_us/tests/policy/baseline/gov/usda/school_meals/ Passed
school_meal_net_subsidy.yaml 11
free_school_meals.yaml 17
reduced_price_school_meals.yaml 16

policyengine_us/tests/core/test_school_meal_takeup_invariants.py passed with six generated examples. It checks Core's per-unit masks and default preservation across mixed group sizes, school ages, states, meal tiers and participation settings.

Pull request run 43920337549 at f33d52067e passed all 37 jobs on GitHub's merge with main 8e1837790b.

For computed benefits, setting take-up to false removes school-meal contributions to spm_unit_benefits, household benefits and CBO means-tested transfers. It also removes the school-meal routes to ACP/EBB categorical eligibility and the Pell simplified-formula benefit list.

Methodology choices

  • Share one annual SPM-unit input across all children and both meals because free and reduced-price amounts draw from the same net subsidy. Separate per-child or breakfast/lunch participation would require a broader change.
  • Gate the net subsidy and both reported benefit amounts. The separate free/reduced gates also suppress computed benefits when callers supply a positive net-subsidy input.
  • Preserve Core's input overrides: supplied final benefit amounts override formulas. The take-up input controls computed benefits.
  • Use the input to model nonapplication and children who do not eat the meals as well as formal declines. The cited regulation addresses formal declines.
  • Retain true defaults to preserve existing benefit amounts. Microcosm owns participation imputation and any rate below full take-up.
  • Compare omitted take-up with explicit true for equality. Allow one cent for rounding against the test's own restatement of the per-meal arithmetic.
  • Leave MSP to the open #9495, which adds an MSP take-up gate. This PR changes school meals alone.

Related work

#10842 adds CSFP take-up separately. #10766 overlaps the free/reduced benefit files and their YAML cases. Merged #10838 splits free and reduced-price values by each child's meal tiers in the same two files, and the take-up gates apply to its formulas.

Microsimulation impact

We estimate that this branch changes none of the outputs below in 2026 on the default dataset. The new takes_up_school_meals_if_eligible input defaults to true, and the default dataset holds no column for it, so every SPM unit keeps the school-meal amounts it gets on main. When a caller or dataset sets the input to false for an SPM unit, the model computes $0 of free and reduced-price school meals for that unit. We compared main at a29c73d587 with this branch at f33d52067e.

Output Main total Change Records changed (weighted)
school_meal_net_subsidy $37.07bn $0 0 (0)
free_school_meals $35.58bn $0 0 (0)
reduced_price_school_meals $1.48bn $0 0 (0)
spm_unit_benefits $2,056.7bn $0 0 (0)
cbo_household_means_tested_transfers $1,096.6bn $0 0 (0)
state_income_tax $576.5bn $0 0 (0)
income_tax $2,594.9bn $0 0 (0)
household_benefits $2,060.2bn $0 0 (0)
household_net_income $17,246.6bn $0 0 (0)

axiom: TheAxiomFoundation/rulespec-us#2296 queued (7 CFR 245.6(c)(6)(iii); the corpus lacks 7 CFR part 245, so the signed encoder can't run yet)

🤖 Generated with Claude Code

Preserve participation by default and gate annual net, free and reduced-price school meal benefits when users or datasets opt out.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Convert decoded tiers to Python strings before scalar parameter lookups.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MaxGhenis and others added 2 commits November 14, 2026 22:36
Preserve the CPI-U projection cases and school meal take-up cases.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Distinguish paid-tier zeros from declined benefits and describe the participation input beyond formal declines. Remove the stale commit reference from the invariant test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 15, 2026 17:29
@MaxGhenis MaxGhenis closed this Nov 15, 2026
@MaxGhenis MaxGhenis reopened this Nov 15, 2026
@MaxGhenis MaxGhenis closed this Nov 15, 2026
@MaxGhenis MaxGhenis reopened this Nov 15, 2026
@MaxGhenis MaxGhenis closed this Nov 16, 2026
@MaxGhenis MaxGhenis reopened this Nov 16, 2026
@MaxGhenis MaxGhenis closed this Nov 16, 2026
@MaxGhenis MaxGhenis reopened this Nov 16, 2026
@MaxGhenis
MaxGhenis merged commit c0e3d57 into main Nov 16, 2026
90 of 123 checks passed
@MaxGhenis
MaxGhenis deleted the add-school-meals-take-up branch November 16, 2026 18:52
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

US + core hub merge audit, us#10834 at f33d52067e369f1424379fcff2ca2c933ef4ebe8 (squash): Add a school meal take-up input.

  • Reviews:
    • The PR has no GitHub reviews. Two independent review rounds ran outside GitHub, with a fix round between them.
    • The first round, at 93d4e5c629, read 7 CFR 245.6(c)(6)(iii) through the eCFR API, USDA's FY2025 cost and participation tables and the prior us-cliffs-2017 totals. It ran 59 households on the head and on the merge base 935b24ff4f, in 2016, 2023, 2025 and 2026. With take-up omitted or true, every school meal amount, spm_unit_benefits and ACP eligibility matched main. With take-up false, the net, free and reduced-price amounts fell to $0 in every tier.
    • For example, a Florida child at 50% of the poverty guideline gets (2.37 − 0.39 + 4.43 − 0.42) × 180 = $1,078.20 of free meals in 2025, and $0 after a decline. In that round the 44 YAML cases and the property test passed on the head, and the three new net-subsidy cases failed on main's code. A mutant that dropped the reduced-price gate failed one case.
    • It found the code correct and asked for an impact estimate or an n/a reason in place of the impact section's placeholder. It also found that two participating cases carried a comment copied from the decline case and that a test comment pinned a main commit. The cited regulation covers formal declines, while the input's documentation described participation in general.
    • The fix round merged main at a29c73d587 in 99809e6c79, which kept main's CPI-U projection cases beside the take-up cases. In f33d52067e it corrected the copied comments, dropped the pinned commit and documented that the input also stands for households that never apply and children who skip the meals. It cut the process narration from the description and named the outputs a false input moves. It changed no formula or expectation.
    • The second round, at f33d52067e, found no defect in the code. It confirmed that the merge of main dropped no case or error margin. It found no conflict with main at 0ae3d78441, where Limit school meal categorical eligibility to each qualifying child #10838 had added per-child formulas to free_school_meals.py and reduced_price_school_meals.py, and read Core 3.36's masking.
    • It recomputed every changed expectation by hand, including the projection cases, and checked the reimbursement rates against the 2024 and 2026 Federal Register notices. The USDA tables supported the description's figures. The eCFR page answered with an access challenge, so it didn't re-read the regulation.
    • Its one remaining finding asked for CI at the head against current main, since every job had skipped on the draft. Run 43741646474 then passed all 37 jobs with this head merged into main at f28e759188, after Limit school meal categorical eligibility to each qualifying child #10838. "Full Suite - Baseline (ssa-usda)" ran the three school meal YAML files, and the Rest Python shards ran the property file.
    • The second round set aside the first round's n/a request, since the description gets the measured impact at merge. Its one description correction asked for CI's result after the local results, and the validation section now carries it. I also quoted the operative words of 245.6(c)(6)(iii), named merged Limit school meal categorical eligibility to each qualifying child #10838 and Add MSP take-up gate and any-month Kansas CSFP categorical eligibility #9495's open state, and pointed the axiom: line at a new parity issue.
  • Methodology choices for maintainers:
    • One yearly SPM-unit input covers every K-12 child in the unit and both breakfast and lunch, since the free and reduced-price amounts split one net subsidy. A household that takes up lunch alone, or for one child of two, would need per-child or per-meal inputs.
    • A false input stands for a formal decline under 245.6(c)(6)(iii), and also for households that never apply and children who don't eat the meals. The regulation covers the decline. Nonapplication and skipped meals come from participation data, and under 245.6(b) and (d) a local educational agency can approve a child with no application.
    • The input defaults to true, so every household keeps main's amounts until a dataset sets it.
    • The gate masks computed values. A supplied school_meal_net_subsidy stays as given while the free and reduced-price amounts computed from it fall to $0, so for that unit they no longer sum to the net subsidy. A supplied final benefit amount overrides the gate.
    • With take-up false, school meals stop qualifying the household for ACP and EBB and leave the Pell simplified-formula benefit list, since those routes count a positive amount. SNAP, TANF and WIC carry inputs for reported receipt (receives_snap, receives_tanf, receives_wic), and school meals carry none.
  • Gates checked live at merge:
  • Impact: we estimate that this branch changes none of the nine measured outputs in 2026 on the default dataset, which holds no takes_up_school_meals_if_eligible column. The new input defaults to true, so every SPM unit keeps the school meal amounts it gets on main. Main a29c73d587 against f33d52067e.
  • Next:
  • Axiom: Encode households declining free or reduced-price school meals (7 CFR 245.6(c)(6)(iii)) (PE-US #10834) TheAxiomFoundation/rulespec-us#2296 queued, a new pe-parity issue for 7 CFR 245.6(c)(6)(iii), with five companion rows from this PR's YAML. The signed encoder can't run yet: neither the corpus nor the release rulespec-us pins holds 7 CFR part 245.

🤖 Generated with Claude Code

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