[MaxGhenis]

What changes

North Dakota renter's refund income now counts SSI and TANF. N.D.C.C. 57-02-08.1(5)(c) defines income as income "from all sources ... including any county, state, or federal public assistance benefits", and leaves out "any amount excluded from income by federal or state law with the exception of income from social security benefits". No federal or North Dakota law excludes SSI or TANF from income. The IRS leaves need-based public assistance out of gross income under its general welfare exclusion, which "is not based on any specific statutory or regulatory authority" (CRS IF12326). If the exclusion clause took in that doctrine, the public assistance clause would have almost nothing to count. N.D.C.C. 1-02-07 and 1-02-38(2) call for a reading that gives both clauses effect: the inclusion counts public assistance, and the exclusion removes what a statute or regulation leaves out of income, including by repealing the provision that counted it, as #10697 reads it, but not a doctrine with no statutory basis such as the general welfare exclusion. The model already counted general assistance on this reading.

The exception for social security says which excluded amounts come back. The Legislative Assembly added it in House Bill 1174 (2019), the Act that created a state income tax deduction for social security. In its fiscal note, the Tax Commissioner's office wrote that section 1 of that Act keeps the full benefit in the homestead income test, so the exception says nothing about benefits no statute excludes.

• SSI raises no conflict: the application (SFN 24777, line 7) asks for "S.S.I.", and the Tax Commissioner's 2026 homestead credit application lists "SSI benefits" as income. The IRS leaves SSI out of gross income only under the general welfare exclusion, so the Commissioner's forms read the exclusion clause as this PR does. Add North Dakota renter's refund #8314's review dropped SSI in May 2026, reading 42 U.S.C. 1382 as a federal exclusion; that section sets SSI eligibility and excludes nothing from income.
• TANF conflicts with item 11 of the Tax Commissioner's guideline, which lists TANF among items that are not income, in the same list as food stamps and federal fuel assistance. The statute governs (the repo's conflicting-sources rule). The guideline's May 2013 edition counted TANF as income under item 10. This PR logs the reading as candidate us-nd:renters-refund-tanf in source_conflicts/candidates.yaml.
• SNAP and LIHEAP stay out. 7 U.S.C. 2017(b) and 42 U.S.C. 8624(f)(1) exclude them from income "for any purpose" under state law, and the clause excepts only social security.

Known limit

PolicyEngine pays TANF to an SPM unit, and add() gives each member an equal share that counts in the member's tax unit. So a tax unit that shares an SPM unit with a TANF family (a grandparent living with a daughter and grandchild, say) counts a per-person share of the family's TANF. Wisconsin's homestead income (wi_homestead_income), Rhode Island's property tax relief income (ri_property_tax_household_income) and Michigan's household resources (mi_household_resources) add TANF into tax-unit income the same way.

Tests

• YAML, with expected values from the statute. nd_renters_refund_income.yaml: case 3 now counts $6,000 of SSI ($18,000); a new case counts $3,600 of TANF and leaves out $2,400 of SNAP ($13,600); another counts a dependent's $9,000 of SSI ($27,000). nd_renters_refund.yaml case 8 carries SSI through to the refund ($360 rather than $600).
• Two Leave federally excluded income out of North Dakota renter's refund income #10697 cases now enter the parents' wages as employment_income_before_lsr. The YAML runner leaves employment_income inputs where they are, and TANF reads wages before any labor supply response, so a family whose case enters wages as employment_income draws ND TANF and the case breaks. (Python Simulation moves the input itself; 152 state test files already enter wages this way.)
• core/test_medical_expense_premium_conservation.py gives its households no SSI or TANF. Its ND oracle tests the medical base, and without that input a two-person household in it with no income draws $8,553 of computed TANF.
• Property-based tests (test_nd_renters_refund_public_assistance.py, Hypothesis). Dropping ssi and tanf from income_sources fails both tests, and adding snap or nd_liheap fails the SNAP and LIHEAP check.

Invariants

These hold for every input; the property tests check them over random SPM units.

1. A dollar of a tax unit's SSI, general assistance or TANF share raises its renter's refund income by at most a dollar and never lowers it, and by exactly a dollar when income without the benefits exceeds zero (only the zero floor can absorb any).
2. SNAP and North Dakota LIHEAP never change renter's refund income or the refund. The test also checks that both inputs reach the model, so the guard can fail.
3. Public assistance never raises the refund (the refund falls as income rises).
4. TANF counts once in total across the tax units of an SPM unit, in proportion to members. The test pins this attribution, the known limit above.

Impact (2026)

We estimate no change in the 2026 refund on the default Populace US 2024 release (populace-us-2024-spm-20260909), comparing base ac21cd383b (#10697) with head 73fd800090. Both branches have moved since: #10697 has merged with a new treatment of losses and possession income, and main has changed SSI (#10141, #10332). A rerun on main (a009a1d) against this head is in progress.

	Base	This PR	Change
Renter's refund	$6,631,322	$6,631,322	$0
Recipients (weighted tax units)	12,625	12,625	0
Refund income, tax units that can reach the refund	$3,204.8M	$3,207.8M	+$3.06M
Refund income, all ND tax units	$44,562.4M	$44,592.4M	+$30.0M

The sample holds 850 North Dakota tax unit records. Of those, 69 have a head or spouse who meets the age or disability test and pays rent, and 25 get the refund. One of the 69 (314 weighted units) has SSI: its income rises by $9,750, and with $13,916 of rent its refund stays at the $600 cap. None of the 69 has general assistance or a TANF share, so the per-person split in the known limit reaches no refund in this sample. A household below the cap does lose: in nd_renters_refund.yaml case 8, $9,000 of SSI cuts the refund from $600 to $360.

The comparison script checks that income rises by a unit's SSI and TANF share wherever base income exceeds zero (largest gap $0.0007), stays within that amount at the zero floor, and that no refund rises. All three hold.

Test cost

The new test file joins CI's remaining Python group. Locally it takes 8.2 s in pytest and 35 s with interpreter start and system build, and peaks at 1.34 GB RSS. The medical conservation test's five cases take 38 s.

axiom: TheAxiomFoundation/rulespec-us#2042 queued (extended with the public assistance reading and companion tests 11 to 13; blocked on corpus capture of chapter 57-02 and a billed encoder run)

🤖 Generated with Claude Code

[DTrim99]

PR Review: Count SSI and TANF in North Dakota renter's refund income

Summary

The fix is right. I read the statute and the Tax Commissioner's forms before the code.

• The statute. N.D.C.C. 57-02-08.1(5)(c) (file page 16) reads: "'Income' means income for the most recent complete taxable year from all sources, including the income of any dependent of the applicant, and including any county, state, or federal public assistance benefits, social security, or other retirement benefits, but excluding any federal rent subsidy, any amount excluded from income by federal or state law with the exception of income from social security benefits, and medical expenses paid during the year by the applicant or the applicant's dependent which is not compensated by insurance or other means."
  • It counts income from all sources, dependents' income, county, state or federal public assistance, social security, and other retirement benefits.
  • It leaves out a federal rent subsidy, amounts that federal or state law excludes from income (social security excepted), and uncompensated medical expenses.
• SSI: the statute and the forms agree.
  • SFN 24777 (10-2025), line 7: "Applicant's and spouse's income from other sources (S.S.I., net rental income, net income from business, capital gains, unemployment compensation, etc.)".
  • The 2026 homestead credit application (SFN 24757, p. 2) lists "2. SSI benefits" as income.
• TANF: the guideline disagrees, and the statute governs. Item 11 of the guideline (p. 2) lists "TANF (Temporary Assistance for Needy Families)" among "Items which are not income". I reached the PR's reading on my own:
  • TANF is a federal and state public assistance benefit, which the definition names.
  • To leave it out, the exclusion clause needs an amount "excluded from income by federal or state law". I found no federal statute or regulation, and no North Dakota statute, that excludes TANF.
  • The IRS's general welfare exclusion "is not based on any specific statutory or regulatory authority" (CRS IF12326; I checked the quote).
  • The agency doesn't apply that doctrine consistently. Its forms count SSI, which has the same federal tax status. Item 12 of the same guideline reads "By law ... not considered income" as specific federal statutes: the stipends from Foster Grandparents, RSVP, Senior Companion and VISTA.
  • So the text settles the question, and the statute-first rule applies rather than the tie-break to the agency. The PR records the choice as candidate us-nd:renters-refund-tanf with follows: statute, which is what CONTRIBUTING.md step 1 asks.
• SNAP and LIHEAP stay out. Both federal laws exclude them from income "for any purpose" under state law: 7 U.S.C. 2017(b) and 42 U.S.C. 8624(f)(1). The clause excepts only social security.
• The refund schedule in (2)(b) (20% of rent, less 4% of income, at most $600, at least $5) and the $70,000 limit in (1)(c)(2) (file page 15) match the parameters.

The diff adds ssi and tanf to the 2025 entry of income_sources and adds tests and the candidate. All 8 commits stay inside those 7 files. Every expectation checks out by hand, CI is 36/36 green at the head, and the branch merges cleanly with current main. I found no critical or should-address issues. I have five suggestions; two concern pre-existing gaps.

My basis is reading the statute and forms (PDF text), the code at the head and the CI logs, plus hand traces. I ran no Python locally.

Critical (must fix)

None.

Should address

None.

Suggestions

1. Say in the parameter file that the cited guideline disagrees on TANF.

   • Where: income_sources.yaml:7-13 (the comment) and the guideline reference at line 25.
   • The file's second reference is the guideline's page 2. That page lists TANF as "not income" and doesn't mention SSI. The comment explains the statute's reading but never says that this reference reads the other way, so a reader who opens it finds the opposite of the tanf line.
   • Fix: add one sentence, e.g. "Item 11 of the Tax Commissioner's guideline lists TANF among items that are not income; candidate us-nd:renters-refund-tanf records the choice." That wording doesn't match the Tie source-conflict markers to the register shared with Axiom #10287 ratchet's patterns. A wording with "guidance says" or "conflicts with" would need a KNOWN_UNMARKED line.

2. Make the Hypothesis tests deterministic.

   • Where: test_nd_renters_refund_public_assistance.py:197 and :288.
   • Both use @settings(max_examples=12, deadline=None) with no derandomize=True, so every CI run draws new households, and a red run may not reproduce on a rerun. 115 of the 139 test files on main that use @settings pass derandomize=True.
   • The @example rows already pin the floor and the shared SPM unit, so derandomizing loses nothing. The sibling test_nd_renters_refund_income_exclusions.py has the same gap, at lines 391, 449 and 513.

3. Follow-up, pre-existing: "other retirement benefits". The PR claims only SSI and TANF, so this doesn't block it.

   • The statute counts "social security, or other retirement benefits". railroad_benefits (Railroad Retirement Board annuities) is in neither market_income nor social_security, so North Dakota income leaves it out entirely.
     • At least the taxable part should count.
     • Tier 1 is taxed like social security (IRC 86), which may bring it under the social security exception too. That is a question for the follow-up.
     • Many income lists on main name it: SSI unearned income, AZ and AL LIHEAP, AZ CCAP and others.
   • survivor_benefits ("Survivor benefits other than Social Security survivor benefits") is left out wholesale, and income Case 3 pins that.
     • The variable mixes two kinds of payment. Survivor annuities from a pension are retirement benefits taxed under IRC 72.
     • Workers' compensation and VA survivor payments are excluded by federal law (IRC 104(a)(1); 38 U.S.C. 5301), and SFN 24777 and the guideline treat them as not income.
   • An issue would cover both.

4. Source captures for the candidate.

   • CONTRIBUTING.md step 1 says the ruling PR also adds, "if the Axiom corpus lacks a source's text, a capture under source_conflicts/sources/".
   • rulespec-us#2042 says axiom-corpus has no 57-02 row, and there is no source_conflicts/sources/us-nd/ on main or on this branch.
   • The three North Dakota renter's refund candidates that Leave federally excluded income out of North Dakota renter's refund income #10697 merged have the same gap, so this PR follows precedent. One capture PR would let step 2 register all four. It would hold 57-02-08.1 (p. 16), guideline p. 2, SFN 24777 p. 1 and SFN 24757 p. 2.

5. Refresh the Impact section of the PR body.

   • It compares base ac21cd383b with head 73fd800090, the head before the rebase.
   • It also says a rerun against a009a1d358 "is in progress". The current head is a636752bf1 on that base.
   • Post the finished rerun or drop the line. No code or test depends on it.

Verified correct

Hand checks. Basis: arithmetic from the statute and the parameters at the head.

#	Case	Arithmetic	Expected	Result
1	nd_renters_refund_income.yaml Case 3 (2025)	UC 8,000 + general assistance 4,000 + SSI 6,000 = 18,000. Child support (3,000), financial assistance (4,000) and survivor benefits (5,000) stay out, as before. The old value without SSI was 12,000.	18,000	✓
2	TANF and SNAP case (2026)	Wages 10,000 + TANF 3,600 = 13,600, with SNAP 2,400 left out and nothing excluded	13,600	✓
3	Dependent's SSI case (2026)	Social security 18,000 + the grandchild's SSI 9,000 = 27,000. The grandparent's own computed SSI is 0, since 1,500 a month of countable income is above the $994 benefit rate; CI confirms computed TANF is 0.	27,000	✓
4	nd_renters_refund.yaml Case 8, income	6,000 + 9,000	15,000	✓
5	Case 8, refund	0.20 × 4,800 = 960; 0.04 × 15,000 = 600; 960 − 600 = 360, within [5, 600]. Eligible: 70 ≥ 65, 15,000 ≤ 70,000, 360 > 0. Without SSI: 960 − 240 = 720, capped at 600, which matches the body's "$360 rather than $600".	360	✓
6	#10697 dependent tax-exempt interest case, wages now as employment_income_before_lsr	30,000 + (500 + 2,000) − 2,000	30,500	✓
7	#10697 adoption assistance case, same input change	45,000 − min(5,000, 45,000)	40,000	✓
8	Property @example, shared SPM unit	Renter: 3,000 + 500 + 3,600 × 1/3 = 4,700. Family: 2,400 + 3,600 × 2/3 = 4,800. The TANF shares 1,200 + 2,400 = 3,600.	as stated	✓
9	Property @example, floor	Unit 1: max(5,000 + 6,000 + 1,000 − 30,000, 0) = 0 against max(5,000 − 30,000, 0) = 0, so it adds 0, within [0, 7,000]. Unit 2: max(9,000 − 4,000, 0) = 5,000 against 0, so it adds 5,000 ≤ 9,000, with the floor absorbing 4,000.	invariant 1	✓
10	Second property test @example	Both tax units have wages above the floor; the shares 1,200 and 2,400 sum to 3,600	invariant 4	✓
11	Medical conservation test, ND oracle	With ssi 0 and tanf 0, income is max(0, wages − paid-cost base), unchanged	unchanged	✓

Refund parameters. rent_rate is 0.2, income_rate 0.04, cap 600 from 2025, minimum 5 and income_limit 70,000. They match (2)(b) and (1)(c)(2).

Mutation. Basis: hand traces.

• Dropping ssi fails income Case 3 (12,000), refund Case 8 (6,000 of income and a $600 refund), the dependent case (18,000) and both property tests.
• Dropping tanf fails the TANF case (10,000) and both property tests.
• Adding snap fails the TANF case (16,000).
• Adding nd_liheap fails the first property test, whose guard checks that the input reaches the model.
• Every new list entry is pinned.

Entity and period.

• ssi is a Person, MONTH variable. add() sums it to the tax unit for the year; Cases 3 and 8 pass in CI with yearly inputs.
• tanf is an SPMUnit variable. In policyengine-core's group-to-group branch of for_each_variable, each member gets tanf / nb_persons and the tax unit sums its members' shares. That matches the PR's known limit and tanf_person.
• No formula changes. The PR adds no state-gated int or enum key, so the NH Add New Hampshire state supplementation for 2026 #10156 pattern doesn't apply.

Known limit: the TANF split.

• The statute counts the applicant's and dependents' income. So when a TANF family that isn't the renter's dependent shares an SPM unit with an aged renter, the renter's tax unit counts part of a benefit it shouldn't.
• The model doesn't attribute TANF to an assistance unit. This PR takes the same approach as WI homestead, RI property tax relief and MI household resources (through tanf) and MO and MT (through tanf_person).
• The PR discloses it, and the second property test pins it. I don't count it as a finding.

No double counting. social_security is "not including SSI", and market_income holds no SSI, TANF or general assistance.

Callers and cycles. nd_renters_refund feeds only the state property tax credit and refund lists. Neither ND TANF nor SSI reads it, and the Microsimulation job is green, so there is no cycle.

The #10697 test edits.

• employment_income equals employment_income_before_lsr with no behavioral response, so the expected values don't move.
• The stated reason fits the CI results: TANF reads pre-response wages, and the YAML runner doesn't move employment_income.

Other tests. The other callers of nd_renters_refund_income either input the income or set up households that computed SSI and TANF don't reach. These are the eligibility file, the aggregate property-tax files, the premium input contract (which already zeroes ssi), the medical deletion test and the SPM parity test. The deletion test has a 65-year-old with $30,000 of wages; the parity test's heads are not aged and so get no refund.

References (CLAUDE.md "Parameter and variable references").

• Statute #page=16 shows (5)(c). Guideline #page=2 shows items 10 to 12 (see suggestion 1). SFN 24777 #page=1 shows line 7.
• Each entry is one source with a single-page href and no page number in its title.

Source conflicts.

• The candidate has the fields step 1 requires: id, program, question, years, six sources, readings and pe_us. It carries a state and no marker, as step 1 requires.
• The new comment lines in income_sources.yaml don't match the Tie source-conflict markers to the register shared with Axiom #10287 ratchet's patterns, so no scanned file's count changes.
• The id is unique in the merged tree, and main hasn't changed the pin or the register.

Test design (CLAUDE.md "Test design and CI cost").

• The property tests cover what fixed YAML cases can't: random amounts and SPM units that hold two tax units.
• The PR body reports local time and memory and the CI group. CI ran the two tests in 1.95 s and 1.79 s in remaining-2.
• The YAML cases sit in the variable-named files.

Housekeeping.

• changelog.d/nd-renters-refund-public-assistance.fixed.md uses a valid type.
• No partner test file changes.
• Commits 2 to 8 refine the candidate's wording, add the property tests (with the LIHEAP guard and the shared SPM example) and tighten docstrings. Nothing is out of scope.

Axiom line assessment

axiom: TheAxiomFoundation/rulespec-us#2042 queued is valid.

• The issue is OPEN and labelled pe-parity, last updated 2026-11-11 22:16 UTC. The only later commit rewords the candidate note and the docstrings.
• It has the module path (us-nd/statutes/57/57-02-08.1/5/c.yaml) and the corpus citation, marked missing. It records two blockers: axiom-corpus has no 57-02 row, and no billed run is approved.
• It also has the verbatim (5)(c), the required outputs and a review_finding text.
• Companion tests 11 to 13 match this PR's TANF case (13,600), dependent's SSI case (27,000) and the 2025 UC, general assistance and SSI case (18,000).
• Its implementation section matches the head: ssi and tanf in income_sources, and the per-person TANF split.

CI status

36 of 36 checks pass at a636752bf1 (run 43120500539). I checked again just before writing: none pending.

• The new property tests passed in Rest (Python shard 6), and the medical conservation tests in shard 3.
• The ND YAML files ran in the state shards.

Branch status

8 ahead and 77 behind main. git merge-tree --write-tree with main is clean.

• Since the merge base a009a1d358, main hasn't touched the ND renter's refund files, ND TANF, ssi, tanf, market_income or the medical conservation test.
• Main did change source_conflicts/candidates.yaml. It merges cleanly, the new id stays unique, and main's test_source_conflicts.py is identical to this branch's.
• Main's SSI changes (Deem income from a disabled parent who is not eligible for SSI #10630, Combine an SSI couple's income by each month's facility status #10820, Restore the part-year facility-stay test cases now that SSI follows each month's facility status #10856) don't reach these tests: the SSI cases input SSI, and the rest have incomes far above SSI levels.

Verdict: APPROVE

[DTrim99]

Approving. N.D.C.C. 57-02-08.1(5)(c) counts "any county, state, or federal public assistance benefits" and drops only amounts excluded by federal or state law. No statute or regulation excludes SSI or TANF, and the Tax Commissioner's own forms count SSI. All 11 hand checks pass, every new list entry is pinned, CI is green at a636752bf1, and the branch merges cleanly with main.

One small addition to the five optional suggestions in the report above: the new us-nd:renters-refund-tanf entry is state: ruled. The conflict is statute against agency guideline, which the project's statute-first rule (2026-10-19, refined 2026-10-20) covers. Naming that rule in the note, as other statute-first entries do, would show where the ruling comes from.