Make state income taxes independent of which spouse is labelled head

[MaxGhenis]

Summary

Swapping a married couple's is_tax_unit_head and is_tax_unit_spouse labels while keeping every other input fixed can change state income tax. The review of #9617 first found this for 12,944 dataset units: federal outcomes were unchanged, while state tax moved by −$8.931M in 2025 and −$8.551M in 2026. This PR removes the identified label dependencies in nine states.

State	Change
Arkansas	ar_agi_indiv moves the dependents' net income to the filer with greater own state net income; exact ties split it equally.
Delaware	Applies the same dependent-income convention in de_agi_indiv; jointly chooses dependent personal-credit allocations and the permitted CDCC/EITC recipients to minimize combined separate tax. Review r1 makes candidate scoring match downstream float32 calculations. The CDCC tie election and inherited EITC comparison remain readings for Max, described below.
Iowa (through 2022)	Applies the dependent-income convention in ia_net_income, retaining signed own income.
Minnesota	Counts an incapacitated spouse under either label for CDCC, retains PE's convention of at most one qualifying spouse, and excludes a single filer as their own qualifying individual.
Missouri	Either spouse aged at least 60 who personally receives survivor benefits may be the property-tax-credit claimant. Age and benefits must belong to the same person.
Mississippi	Applies the dependent-income convention in ms_agi, preserving negative filer columns and clipping dependent losses before transfer. HSA, self-employed health-insurance and pension deductions retain known per-person attribution; amounts with no per-person attribution split equally between the spouses.
Montana	Applies the dependent-income convention in legacy mt_agi_indiv (through 2023). Both joint itemized-deduction formulas add both filers' deductible mortgage interest and investment-interest expenses, excluding dependents. The legacy dependent-care qualifying-individual formula recognizes an incapacitated spouse under either label and excludes a single filer as their own qualifying individual.
Oklahoma	Either spouse may be the disabled property-tax-credit claimant, subject to the claimant's ownership, maintenance and own-support conditions.
Oregon	Working-family care-credit employment eligibility recognizes either spouse's earnings and a disabled spouse's imputed earnings. The qualifying-individual calculation retains main's claimable-filer restrictions and recognizes either disabled spouse; a single filer cannot qualify as their own care recipient.

The shared move_dependent_amounts_to_filer helper in model_api.py contains the dependent-income convention. The Delaware helper keeps each eligible filer's own personal credit and aged credit in that person's column, while coordinating the other credit choices.

How the swap asymmetry was found

• The original random scan on main (f33a4b3b16) used six seeds and 40 married couples per state, with and without dependents, 27 adult income/deduction inputs and years 2019–2027. AR, IA through 2022, MS and MT changed by more than $0.005.
• The original edge scan used 25 crafted couple types in every state, 2021–2026, and found Oklahoma's $200 disabled-spouse case. Later review cases exposed the Minnesota, Missouri, Oregon and Delaware credit paths now covered here.
• Delaware's original dependent-wage example ($100,000/$30,000 adult wages and $20,000 dependent wages) had no effect: the dependent's federal AGI and de_pre_exclusions_agi were zero, and state tax was $6,137.12 under both labels. A dependent reaches this allocation through inputs such as de_additions. This historical result does not imply that Delaware's credit routing was invariant.
• Montana's original no-dependent example traced to joint itemized interest: one 2023 couple's deduction was $12,018.81 versus $22,023.81 under exchanged labels.

The law and methodology choices (for the hub to queue for Max)

The dependent-income convention is a new modelling reading, not a state rule requiring a dependent's income to appear on a parent's return. The reviewed instructions give each filer their own column: Arkansas 2025 instructions, p. 12, Delaware 2025 instructions, p. 5, Iowa 2021 instructions, p. 7, Mississippi instructions, p. 5, and Montana 2023 instructions, p. 12. A dependent's own return income belongs on that dependent's return. PE currently includes these amounts on the filers' return and does not compute dependents' own returns (#9618); this PR preserves that existing coverage boundary and changes the receiving column.

Choice	PR's pick and measured result	Alternative and status for Max
Dependent income: AR, DE, IA, MS and legacy MT	Greater own state net income before transfer receives the dependent amount; exact ties split equally. AR 2025, adult wages $80,000/$10,000 and dependent wages $20,000: $3,577.	Proportional allocation gives $3,315; excluding the dependent income from both filing paths gives $2,497. The greater-income/tie rule is a new modelling convention; continued inclusion of dependent income follows existing PE coverage. The analogy is the federal parent election in 26 USC 1(g)(5)(B) and Form 8814, whose comparison is taxable income. The PR uses each spouse's own state net income before transfer.
Mississippi deductions without per-person attribution	Split HSA, self-employed health-insurance and pension deductions equally. For wages $100,000/$5,000 and an $8,550 deduction, state tax is $3,041.50, for each provision.	Attributing all of it to the higher earner gives $2,853.40; all to the lower earner gives $3,073.40. Equal halves are a new imputation when ownership is unknown. Using known per-person amounts, scaled to the tax-unit total, follows existing PE attribution.
Delaware dependent personal-credit allocation	Choose dependent-credit increments jointly with the later credits, minimizing tax after nonrefundable EITC: $310.68 in the reviewed household.	Another dependent allocation produces $365. Allowing dependent $110 increments to move follows PIT-RES line 27a, p. 8 and existing PE optimization; coordinating the allocation with the later credits is new. It is separate from whether a tied CDCC recipient may be elected.
Delaware CDCC on equal taxable incomes	Permit either tied column and choose it jointly with the other credits: $1.50.	The review's alternative recipient gives $15. This is a new unresolved tie-election reading, conditional on allowing that election; 30 Del. C. §1114(b) and the 2025 instructions specify the spouse with lower taxable income without establishing the election when equal. rulespec-us#1607 remains blocked on that reading. No tie authority was found in the 2011, 2013 and 2022–2025 resident instructions, the 2025 DOR Tax Preference Report, the DOR TIM index/2005-02 memorandum, or the regulation search described below. This is a scoped negative search result.
Delaware nonrefundable EITC routing	Retain the form's higher-Line-23-taxable-income comparison; allow either tied column and minimize the combined result.	30 Del. C. §1117(b) instead says greater tax otherwise due. The taxable-income comparison follows existing PE/form convention; tax-minimizing routing on ties is new. The actual 2025 PIT-RSS Schedule II line 12 also selects the greater Line 33 tax balance, conflicting with the PIT-RES instruction prose, p. 10. The worked 2025 example below gives $53.02 versus $15.20 household tax using the official tables; these are hand calculations, not PE outputs.
Oklahoma disabled-spouse claimant	Permit either spouse to be the claimant: $200 with $9,000 household income and $400 property tax.	Head-only treatment gives $0 when the disabled spouse is not the tax-unit head. This is a new reading of the claimant rule; either spouse's age already counted in PE. The claimant must meet the ownership, maintenance and own-support conditions described below.
Minnesota qualifying spouse	At most one incapacitated spouse contributes to the CDCC count: the tested two-incapacitated-spouse household receives $960.	A hypothetical two-person-cap calculation gives $1,920. The one-qualifying-spouse convention follows existing PE federal count_cdcc_eligible, now applied symmetrically in Minnesota; it is not a new two-spouse entitlement.
Montana first-listed storage	The joint itemized deduction is stored on the first-listed filer's person row.	This existing PE output-storage convention has $0 tax effect because downstream readers sum the filers' amounts. The PR changes which people's deductible expenses are included, while retaining the storage convention.

The Delaware authority search covered the 2025, 2024, 2023, 2022, 2013 and 2011 resident instructions; the 2025 Tax Preference Report; the TIM index and TIM 2005-02; the administrative-code index; and targeted official-domain searches for regulations and technical guidance. None located a CDCC equality election or reconciled the EITC comparison. The 2011 booklet already contains the EITC conflict: taxable-income prose alongside a greater-tax worksheet comparison. The actual 2025 worksheet supports the statute's greater-remaining-tax comparison. The PR keeps the two implemented routing readings and explicitly queues them for Max.

A 2025 EITC-routing disagreement exists because the age credit can reverse the ordering. Consider full-year residents filing jointly federally and combined-separately in Delaware, with no dependents, no other income or adjustments, neither spouse claimable elsewhere, and otherwise eligible for childless federal EITC:

Official-table hand calculation	Spouse A, age 60	Spouse B, age 45
Wages	$12,000	$11,000
Standard deduction	$3,250	$3,250
Taxable income	$8,750	$7,750
Gross tax, 2025 DE table	$213	$174
Own plus aged personal credits	$220	$110
Tax remaining / Line 33	$0	$64

Higher taxable income selects A; greater remaining tax selects B. The 2025 IRS EIC table gives $244 federal EITC at $23,000 of married-joint income without qualifying children, yielding a $10.98 refundable Delaware alternative and $48.80 nonrefundable potential. Taxable-income routing cannot use nonrefundable EITC against A's zero balance, so household tax is $64 − $10.98 = $53.02. Greater-tax routing applies $48.80 against B, producing $15.20. These hand calculations use published tax-table amounts before whole-dollar return rounding; they are not model results. PE's continuous-rate gross taxes are $212.25/$173.25, leaving $0/$63.25, which also reverses the ordering; the precise table-based liabilities above should not be attributed to PE.

Proportional income shares are undefined only when the spouses' combined income is zero. With mixed signs and a nonzero total they are defined, but can be negative or exceed one; this makes them a further allocation choice for signed IA/MS incomes. Dropping dependent income should be considered with dependents' own-return modelling (#9618), so that income is not simply removed from the tax system.

For Oklahoma, 68 O.S. §2906 and Form 538-H, p. 3 require the claimant to own or jointly own and maintain the home, and furnish their own support for home furnishings and other material necessities. The PR's either-spouse reading assumes the qualifying disabled spouse meets those claimant conditions; joint ownership or disability alone does not establish all of them. The formula does not add new inputs to establish those conditions. The form's age question already refers to either spouse.

Invariants and tests

The intended invariants are unchanged tax-unit tax when labels are exchanged; unchanged per-person AR/DE/IA/MS/legacy-MT income columns; conservation of each formula's dependent amount across the filer columns; and unchanged allocation for units with no spouse. Mississippi preserves its existing clipping of dependent losses before transfer. The helper assumes one head and at most one spouse.

The property test test_state_tax_head_spouse_swap_invariance.py runs Hypothesis batches and seeded edge cases in AR, DE, IA, MN, MO, MS, MT, OK and OR for 2021–2026, an all-state seeded guard for 2026, and an independent numpy reference for AR/DE/IA/MS allocations. Review r1 adds the Delaware high-income rounding witness to the seeded population, alongside YAML regressions under both labels.

Five existing numerical YAML cases change, rather than the two named in the earlier body:

File and case	Old → new expected amounts
ia_net_income.yaml: IA net income unit test 2, now Spouses with equal net income split the dependent net income equally	Net income [8,600, 8,000, 0] → [8,300, 8,300, 0].
ms_agi.yaml: The dependent income is allocated to the head, now The dependent income goes to the spouse with the greater income, here not the head	AGI [12,000, 10,000, 0] → [9,000, 13,000, 0].
ms_health_savings_account_adjustment.yaml: A joint return without per-person amounts attributes the deduction once, to the head, now ...half to each spouse	HSA adjustment [3,000, 0] → [1,500, 1,500]; AGI [37,000, 30,000] → [38,500, 28,500].
Same HSA file: A dependent's income is not reduced by the filer's deduction, nor by the dependent's own account	HSA adjustment [3,000, 0, 0] → [1,500, 1,500, 0]; AGI [42,000, 30,000, 0] → [43,500, 28,500, 0].
Same HSA file: Without per-person amounts the head takes the deduction, even above the head's own income, now ...each spouse takes half, even above one spouse's own income	HSA adjustment [3,000, 0] → [1,500, 1,500]; AGI [−2,000, 50,000] → [−500, 48,500]. Final unit tax remains $941.60.

The Missouri claimant-only test is also recast explicitly: a spouse personally meeting both age and survivor-benefit requirements now qualifies, while a separate case retains false when age and benefits belong to different spouses. Montana's joint itemized YAML files preserve main's mortgage-cap cases and use deductible_mortgage_interest in the PR's filer-aggregation cases.

Historical validation of the original test version failed all eight tests on original main and passed all eight on the branch. Its local macOS arm64 measurement was 25 seconds test time, 45 seconds including imports and 2.9 GB peak RSS; those are historical figures, not a benchmark of the expanded final test. The file remains in the Rest/core CI group. Review-r1 pre-change property run: 8 passed in 63.81 seconds, 78.79 seconds wall time. Peak RSS was unavailable because the sandbox denied the timing tool's sysctl call. The added Delaware witness joins the existing yearly batches without adding model constructions. Review-r1 after-change measurement: 8 passed in 37.56 seconds, 62.31 seconds wall time; peak RSS remained unavailable. These are single local runs with variable timing, so the difference is not evidence of a reliable speed improvement. Rest/core CI placement is unchanged.

Review r1: responses

Finding	Change	Evidence
P2: Delaware high-income swap witness differed by $0.015625	Candidate scoring now reproduces stored float32 personal credits, credit addition and Line 33 subtraction, promotes stored columns for downstream summation/EITC, and rounds the final candidate tax to float32. It uses a strict comparison and retains the existing allocation and recipient rules. Added the exact 2025 witness in YAML under both labels and to the property test's seeded population.	Before: $131,091.96875 / $131,091.984375. After: $131,091.96875 / $131,091.96875; 9981-de-fail-before.log: 1 failed, 7 passed in 10.52 seconds, with expected $131,091.96875 versus actual $131,091.984375 (difference $0.015625 > $0.001 tolerance). After-change output: 9981-de-pass-after.log: 8 passed in 11.56 seconds, including both labelings at $131,091.96875 within the $0.001 YAML tolerance.
P2: CDCC tie election lacks established support; EITC form/statute comparison differs	Searched instructions from other years, regulations, technical information memoranda and worksheets. Kept the implemented readings and made the two issues explicit in the methodology table, with alternatives and measured effects.	No CDCC tie election was established by the reviewed sources; the actual 2025 PIT-RSS worksheet supports greater remaining tax and conflicts with the taxable-income prose. The hand example above demonstrates different recipients and liabilities. Search record: 9981-delaware-authority.md; code comments now cite the statute, instruction booklet and actual worksheet.
P3: body omitted final changes and methodology choices	Added MN/MO/OR and MS attribution to the summary; named all five existing numerical YAML changes and the Missouri test replacement; corrected signed proportional shares and Oklahoma's own-support condition; removed stale Virginia and Maryland fallback notes.	Body tables above; The updated summary, methodology and case tables record these corrections; the original axiom: line is preserved exactly and the attribution footer remains last.
P3: two YAML files had extra EOF blank lines	Removed the extra EOF blank lines in Minnesota CDCC and Montana care-qualifying-individual YAML.	UV_NO_SYNC=1 make format passed: two files reformatted and 7,316 unchanged.

Every file in the main-merge report's table, including the new Delaware regression coverage, was rerun individually in the foreground with OMP_NUM_THREADS, OPENBLAS_NUM_THREADS, MKL_NUM_THREADS, VECLIB_MAXIMUM_THREADS and NUMEXPR_NUM_THREADS set to 4 and OMP_WAIT_POLICY=PASSIVE.

All 209 tests passed across 26 files: 27 Python and 182 YAML, including the two new high-income Delaware label regressions. make format and git diff --check passed.

File	Passed
test_state_tax_head_spouse_swap_invariance.py	8
test_ms_health_savings_account_adjustment_invariants.py	2
test_mt_mortgage_interest_federal_limit_properties.py	6
test_claimable_filer_federal_invariants.py	7
test_claimable_filer_or_me_hi_invariants.py	4
ar_agi_indiv.yaml	5
de_agi_indiv.yaml	4
de_income_tax_before_refundable_credits_separate.yaml	8
de_personal_credit_indv.yaml	7
de_personal_credits.yaml	6
ia_net_income.yaml	6
mn_cdcc_dependent_count.yaml	5
mo_ptc_taxunit_eligible.yaml	24
ms_health_savings_account_adjustment.yaml	10
ms_agi.yaml	10
mt_child_dependent_care_expense_deduction_qualifying_individual.yaml	8
mt_itemized_deductions_joint.yaml	11
mt_itemized_deductions_for_federal_itemization_joint.yaml	10
mt_itemized_deductions_indiv.yaml	5
mt_itemized_deductions_for_federal_itemization_indiv.yaml	10
mt_agi_indiv.yaml	4
ok_ptc.yaml	9
or_wfhdc_employment_eligible.yaml	8
or_wfhdc_has_qualified_individual_eligible.yaml	15
or_cdcc_relevent_expenses.yaml	9
or_wfhdc_eligibility_category.yaml	8

Microsimulation impact

Rerun 2026-10-19 by the hub at this head df6550b10b against main at the merge base f9b55fc4dd, on the default dataset populace_us_2024.h5@populace-us-2024-spm-20260909, through lockrun.sh, with a by-state breakdown. State income tax, weighted, $ millions:

State	2025 change	2025 households changed (records / weighted)	2026 change	2026 households changed (records / weighted)
AR	+3.824	19 / 30,948	+4.340	18 / 27,264
MS	-0.773	4 / 10,069	-0.656	4 / 10,140
DE, IA, MT, OK and every other state	0.000	0 / 0	0.000	0 / 0
All states	+3.052	23 / 41,016	+3.683	22 / 37,404

• Arkansas moves both ways: +6.254 / -2.429 in 2025 and +6.581 / -2.241 in 2026.
• Mississippi moves both ways: +1.182 / -1.955 in 2025 and +1.133 / -1.790 in 2026.
• Montana income tax doesn't change at this head; mt_itemized_deductions_joint changes on 2 zero-weight records in 2025. The earlier run's base, f33a4b3b16, predates Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's merge into main on 2026-10-09, and Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004 also changed Montana's joint interest lines.
• Delaware, Iowa and Oklahoma don't change on this dataset.
• Federal income tax is unchanged. Household net income falls by the state tax change: $3.052M in 2025 and $3.683M in 2026.

The earlier figures (cd24bdd4af against f33a4b3b16: all states -7.057 in 2025 and -5.995 in 2026, mostly Montana) are superseded by this table. Comparison files: fixes/9981-impact/ on the hub host.

Merge with main and overlaps

The previous merge brought in main f9b55fc4dd05f1bed921a4dd06cb8c81f6ceb2c7 via 2df930fac41a9dbd51d84c3a0e70b0aab9776cf0. Review r1 started from the requested head 2df930fac41a9dbd51d84c3a0e70b0aab9776cf0; fetched gh/main remained f9b55fc4dd05f1bed921a4dd06cb8c81f6ceb2c7, already merged. No new merge or conflict resolutions were required. The review-r1 head is df6550b10bd4006bcc5e80fff22a7870d7f9c407.

• Montana: the two joint itemized formulas combine Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's federally deductible mortgage-interest correction with this PR's head/spouse-only aggregation. Main's individual formulas are unchanged. Investment-interest expenses remain main's raw expense variable, summed over head and spouse. Main's mortgage-cap YAML cases and this PR's dependent-exclusion and $500 investment-interest swap cases are retained.
• Delaware: both the personal-credit formula and the optimizer retain main's claimable-filer restrictions. A filer claimable elsewhere has no own personal credit, and eligible dependent increments are derived from federal exemptions after subtracting eligible filers. Two filing-exception cases cover both labels.
• Oregon: main's qualifying-individual helper retains claimable-filer restrictions and exceptions while recognizing either disabled spouse. Main's YAML cases and this PR's single-disabled-filer case are retained.
• Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's Montana mortgage-interest correction is included. The previous merge deferred Limit investment interest deductions under §163(d) and Form 4952 #9977's §163(d) investment-interest changes; whichever change lands second must preserve both intents. The historical microsimulation above predates these merge resolutions.
• Give each spouse their own above-the-line deductions in per-person AGI #9890/Net household members' Medicaid MAGI before the zero floor, and give each spouse their own deductions #9860 edit legacy mt_agi_indiv above the allocation block; Leave dependents' own items out of DC, Delaware and Ohio subtractions #9939/Leave dependents' capital income out of state lines on the filers' return #9894 change amounts that can reach filer state lines. Honor dataset-supplied tax-unit roles #9617 makes dataset-supplied roles reach these formulas and is the origin of this follow-up.

axiom: 68 O.S. §2906 TheAxiomFoundation/rulespec-us#1555 queued | Minn. Stat. 290.067 / 26 USC 21(b)(1)(C) TheAxiomFoundation/rulespec-us#1603 queued | RSMo 135.010(1) TheAxiomFoundation/rulespec-us#1604 queued | ORS 315.264 / 26 USC 21 TheAxiomFoundation/rulespec-us#1605 queued | 30 Del. C. §1110 / PIT-RES line27 TheAxiomFoundation/rulespec-us#1606 queued | 30 Del. C. §1114 TheAxiomFoundation/rulespec-us#1607 queued | 30 Del. C. §1117 TheAxiomFoundation/rulespec-us#1608 queued | MCA 15-30-2131(1)(a), (1)(c)(i)(C) TheAxiomFoundation/rulespec-us#1512 queued | MCA 15-30-2131 (Montana itemized deductions, 2021–2023) queued, blocked: the corpus has only the repeal notice | n/a for the dependent-income column rule: it places income the law keeps on dependents' own returns (#9618), so there is no provision to mirror | n/a for unknown-account ownership and first-listed output storage conventions

🤖 Generated with Claude Code