[PavelMakarchuk]

Summary

Fixes the Arkansas, Iowa, Mississippi and New Jersey rows of #9814. #9943 fixed the Alabama row.

irs_gross_income leaves a tax unit dependent's income off the filer's federal return. These four states still put it on the filer's return:

• ar_agi_indiv, ia_net_income and ms_agi moved each dependent's amount to a filer's column with move_dependent_amounts_to_filer.
• ar_agi_joint moved it to the head.
• nj_agi summed nj_total_income over every member.

Each state taxes a dependent on their own return. Only a married couple combines income on one return. This PR leaves a dependent's income off the filers' return in all four states. It follows #9943:

• AR, IA, MS: the person-level column is zero for a dependent. Nothing reads a dependent's column on its own. The column describes the filers' return, and readers sum it over the tax unit.
• NJ: nj_agi and the Worksheet D total income (nj_retirement_exclusion_fraction) sum nj_total_income over the head and spouse with tax_unit_non_dep_add. nj_gross_income and nj_total_income still give each person's own amount, dependents included. A dependent's own return could use them.

Law

Arkansas

• Ark. Code § 26-51-801:
  • (a): "Every person owning property or doing business in the State of Arkansas shall file a return", subject to the gross income minimums in (a)(1) to (10).
  • (b): spouses who live together each file "unless the income of each is included in a single joint return".
  • (c): a guardian files for a taxpayer who cannot file.
  • No subsection puts a dependent's income on another person's return.
• 2025 AR1000F/AR1000NR instructions, PDF p. 10, "Who must file": the gross income minimums apply to each person. Nothing exempts a dependent.
• PDF p. 12: the primary's income goes in column A and the spouse's in column B. For other filing statuses, "write all income in column A only". No line takes a dependent's income. The instructions do not mention Form 8814 or a child's income.

Iowa

• Iowa Code § 422.7: "net income" means "the adjusted gross income before the net operating loss deduction as properly computed for federal income tax purposes", with Iowa adjustments. Federal AGI leaves out a dependent's income (Leave dependents' investment income out of the filer's NIIT base #9784), so Iowa net income does too.
• § 422.13(1)(a): a person "claimed as a dependent on another person's return" must file a return with net income of $5,000 or more. Subsection (2) combines only "the combined net income of a husband and wife".
• 2022 IA 1040 instructions:
  • PDF p. 4, filing requirement (c), is the same rule.
  • PDF p. 8 has a "Note to dependents filing their own returns".
  • The 2021 booklet has the same text.
  • Neither booklet mentions Form 8814.

Mississippi

• Miss. Code § 27-7-5(1) imposes the tax "upon the entire net income of every resident individual".
• 35 Miss. Admin. Code Pt. III, 10.01.104 ("Minors"):
  • A minor is "taxable on his wages, on the income he receives from the property he owns".
  • Over the exemption, the minor "must file a return".
  • On that return the minor "is entitled to his own deductions and exemptions, like any other taxpayer".
• 2024 Form 80-100 instructions:
  • PDF p. 4 (printed p. 3): "Minor resident taxpayers" must file when their gross income exceeds the personal exemption plus the standard deduction.
  • PDF p. 5: a combined return puts one spouse's income in Column A and the other's in Column B.
  • The instructions do not mention Form 8814.

New Jersey

• N.J.S.A. 54A:8-3.1:
  • (a)(1)(c): an unmarried individual with gross income over $10,000 files a return.
  • (b): only resident spouses file a joint New Jersey return.
  • (f): a fiduciary files the return of a person unable to file "by reason of minority".
• 2025 NJ-1040 instructions:
  • PDF p. 5 (printed p. 3), "Do You Have to File": the $10,000 and $20,000 thresholds have no exception for dependents.
  • PDF p. 40, Worksheet L: household income adds "the total income and tax-exempt interest of your dependents", with estimates "for any dependents who do not file a New Jersey tax return". So dependents file their own returns, and their income is not in the filer's line 27.
  • The instructions do not mention Form 8814.

Where a dependent's income could be on a parent's return. That happens only through the federal Form 8814 election, which reaches federal AGI first. The model does not implement Form 8814, so no state case is modelled. In Iowa, an 8814 amount would flow into net income through federal AGI. None of the four states' instructions has a separate line for a child's income.

Before and after

The reproduction from #9814 on main at e53b426082 and on this branch. It uses a single parent with $40,000 of wages and one dependent child aged 16. Each cell gives the state income variable and the state income tax. A bold cell differs from the row with no child income.

State	Child income	main	This PR
AR (ar_agi_indiv)	none	40,000 / 957	40,000 / 957
	$8,000 interest	48,000 / 1,269	40,000 / 957
	$5,000 wages	45,000 / 1,152	40,000 / 957
IA (ia_net_income)	none	40,000 / 252.15	40,000 / 252.15
	$8,000 interest	48,000 / 252.15	40,000 / 252.15
	$5,000 wages	45,000 / 252.15	40,000 / 252.15
IA, TY2022	none	40,000 / 1,332.05	40,000 / 1,332.05
	$8,000 interest	48,000 / 1,830.64	40,000 / 1,332.05
	$5,000 wages	45,000 / 1,643.14	40,000 / 1,332.05
MS (ms_agi)	none	40,000 / 752.40	40,000 / 752.40
	$8,000 interest	48,000 / 1,104.40	40,000 / 752.40
	$5,000 wages	45,000 / 972.40	40,000 / 752.40
NJ (nj_agi)	none	40,000 / −80.68	40,000 / −80.68
	$8,000 interest	48,000 / 59.32	40,000 / −80.68
	$5,000 wages	45,000 / 6.82	40,000 / −80.68

TY2025 unless the row says otherwise. Delaware and Montana also use move_dependent_amounts_to_filer. Their bases are already free of a dependent's income in this reproduction, and this PR does not change them.

What changes for whom

The changes apply only to tax units whose dependents have income.

• Arkansas, Mississippi, Iowa before 2023, and New Jersey:
  • A dependent's income no longer raises the filers' state AGI or taxable income.
  • It no longer changes the separate-versus-joint choice (AR, MS) or the joint deduction allocation (MS).
  • It no longer changes the tax.
• Iowa from 2023: the tax starts from federal taxable income, so the tax does not change. ia_net_income, Iowa's state_agi, now leaves out the dependent's income.
• New Jersey:
  • A dependent's income no longer lowers the filers' pension exclusion percentage (Worksheet D uses line 27 total income).
  • It no longer raises the 2% medical floor.
  • The other retirement income exclusion (nj_other_retirement_income_exclusion, Worksheet D, line 28b) now uses only the filers' amounts. Its $3,000 earnings test adds NJ-1040 lines 15, 18, 21 and 22, the filers' own wages and business income, so a dependent's wages no longer take the exclusion away. Its maximum exclusion uses the total income of the age-eligible head or spouse only ("Joint filers: If only one spouse is 62 or older, only the income of that spouse can be excluded", 2025 NJ-1040 instructions, Worksheet D), not an age-eligible dependent's. New test: a 65-year-old head of household with $30,000 of interest and a $10,000 pension, and a dependent child with $5,000 of wages, gets $30,000 (main: $0).
  • A dependent's self-employed health insurance deduction (C.54A:3-5) now stays on the dependent's own return, with the income it is figured against. nj_medical_expense_deduction used to add dependents_self_employed_health_insurance_ald "because nj_agi counts every member's income". The exclusion of those premiums from the filers' medical base (C.54A:3-3(c)(3)) stays.
• Not in this PR:
  • New Jersey ANCHOR (nj_anchor, nj_anchor_eligible) and the property tax relief and StayNJ income (nj_property_tax_relief_income) still sum nj_gross_income over every member. Those programs have their own income definitions, which I did not research here.
  • Oklahoma's ok_gross_income in the sales tax relief credit. AL, AR, MS and NJ add tax unit dependents' income to the filer's state income (Iowa net income too), although federal AGI leaves it out #9814 lists it as not covered.
  • A dependent's own state return (Compute a tax-unit dependent's own income tax return #9618).
  • The dependents_self_employed_* ALD variables no longer have a state user. They are kept.

Overlap with open PRs

None of these PRs changes the formulas this PR changes. The textual overlaps:

• Leave dependents' capital income out of state lines on the filers' return #9894 (MaxGhenis, dependents' capital income in state lines):
  • Lists AR, MS, NJ and IA before 2023 as "Not in this PR".
  • Edits ia_reportable_social_security.py, which this PR does not touch.
  • Rewrites the same docstring paragraph of tests/core/test_state_tax_head_spouse_swap_invariance.py, for Montana. Whichever PR merges second needs a small docstring merge. With both PRs, Arkansas, Iowa, Mississippi and Montana all leave the dependents' amounts off, and only Delaware still moves them.
• Leave dependents' own items out of DC, Delaware and Ohio subtractions #9939 (DC, DE, OH subtractions) edits ia_reportable_social_security.py. There is no file overlap.
• Apply Iowa, Kansas, Louisiana and Maryland rules for a filer who can be claimed as a dependent #10045 (Iowa claimable-filer rules) sets ia_net_income as an input in its tests. It does not read a dependent's column, so it does not conflict.
• Count only filer-paid premiums in Schedule A medical expenses #10270 (MaxGhenis, filer-paid premiums in Schedule A) also edits nj_medical_expense_deduction.py and tests/core/test_medical_expense_premium_conservation.py. It keeps dependents_self_employed_health_insurance_ald "since New Jersey counts the dependent's income". This PR removes that premise, so the second PR to merge needs a rebase there.
• Offset New Jersey claim of right repayments within their income category (N.J.S.A. 54A:5-2) #10198 (NJ claim of right) edits nj_gross_income.py. This PR leaves that file alone and changes only the tax-unit sums.
• Apply Arizona, Arkansas and California rules for a filer who can be claimed as a dependent #10053 (AR claimable filer) and Apply New Jersey, North Dakota, Ohio, Oklahoma and Pennsylvania rules for a filer who can be claimed as a dependent #10043 (NJ claimable filer) set ar_agi_joint as an input or test NJ credits. There is no conflict.

Tests

New YAML cases. Each uses a parent with $40,000 of wages and a dependent child aged 16 who has $8,000 of interest or $5,000 of wages. State income equals the case with no child income.

• ar_agi_indiv.yaml:
  • ar_agi_indiv and ar_agi_joint [40,000, 0] for both child incomes.
  • AR tax 957.
• ms_agi.yaml:
  • ms_agi [40,000, 0] for both child incomes.
  • MS tax 752.40, which is 4.4% of (40,000 − 3,400 − 8,000 − 1,500 − 10,000).
• ia_net_income.yaml:
  • 2022 interest case: ia_net_income [40,000, 0] and IA tax 1,332.05.
  • 2025 wages case: ia_net_income [40,000, 0] and state_agi 40,000.
• nj_agi.yaml:
  • Both child incomes: nj_agi 40,000, nj_taxable_income 37,500 (40,000 − 1,000 − 1,500) and NJ tax −80.68.
  • A joint couple aged 65 with $90,000 of pension income and a child with $15,000 of wages: the exclusion percentage is 100%, the exclusion is 90,000 and nj_agi is 0. Counting the child's wages would have put total income at 105,000 and the percentage at 50%.

Existing tests that encoded the old behavior. Each now expects the dependent's amount to stay off the filers' return:

• ar_agi_indiv.yaml, three cases. These are the "goes to the spouse with the greater AGI" and "equal AGI split" cases.
  • [1,200, 0, 0] becomes [900, 0, 0], twice.
  • [1,150, 1,150, 0] becomes [1,000, 1,000, 0].
• ar_agi_joint.yaml: [1,200, −100, 0] becomes [900, −100, 0].
• ar_gambling_winnings_and_losses.yaml: a dependent's gambling winnings case, ar_agi_joint [51,000, 0] becomes [50,000, 0].
• ms_agi.yaml, four cases:
  • [9,000, 13,000, 0] becomes [9,000, 10,000, 0], twice.
  • [11,500, 11,500, 0] becomes [10,000, 10,000, 0].
  • In the dependent's self-employment case, the parent's ms_agi becomes 40,000 (was 47,793.52).
• ms_health_savings_account_adjustment.yaml: ms_agi [43,500, 28,500, 0] becomes [38,500, 28,500, 0].
• ms/tax/income/integration.yaml, first case. The child (age 10, $3,000 of wages) was in the head's column. The values are recomputed by hand in the test:
  • ms_agi [20,000, 18,000].
  • The joint allocation gives the head 10,000 of the 19,600 of deductions and exemptions, so the deductions are 2,346.94 / 2,253.06 and the exemptions 7,653.06 / 7,346.94.
  • Taxable income is [10,000, 8,400] combined and [10,200, 8,200] separate.
  • Tax is 396 combined and 398 separate, so the tax is 396 (was 530).
• ia_net_income.yaml, four cases:
  • [9,400, 0, 0] becomes [8,000, 0, 0].
  • [8,300, 8,300, 0] becomes [8,000, 8,000, 0].
  • [8,000, 30,600, 0] becomes [8,000, 30,000, 0], twice.
• nj_medical_expense_deduction.yaml, four dependent cases:
  • Dependent SE health insurance: nj_agi 50,000 becomes 40,000, and the deduction 1,000 becomes 0.
  • Separately deducted premiums: nj_agi 40,000 becomes 30,000, and the deduction 1,000 becomes 0. The 500 of other expenses is under the 600 floor.
  • One dependent's deduction against another dependent's premiums: 2,000 becomes 1,000.
  • Dependent long-term care: nj_agi 104,000 becomes 100,000, and the deduction 3,140 becomes 220.
• tests/core/test_medical_expense_premium_conservation.py: the NJ property no longer adds the dependents' SE deductions to the expected deduction.
• tests/core/test_state_tax_head_spouse_swap_invariance.py:
  • The differential now expects each filer's own amount, and zero for a dependent, in Arkansas, Iowa and Mississippi.
  • Delaware keeps the numpy allocation oracle.
  • The docstring is updated.

Local runs on this branch:

• policyengine-core test:
  • states/ar: 1,052 passed.
  • states/ms: 499 passed.
  • states/nj: 779 passed.
  • states/ia: 792 passed (this includes its Hypothesis file).
• pytest:
  • The head and spouse swap invariance file: 8 passed.
  • Related core and unit property files: 116 passed. These are the medical premium conservation, MS HSA, dependent ALD, premium input contract, AR medical, state estate income, AR gambling and AR casualty files.
• policyengine-core test on states/tax/income/state_agi.yaml and the AR, IA, MS and NJ contrib tests: 55 passed.
• ruff format and ruff check on the changed Python: clean.

Fixes #9814

🤖 Generated with Claude Code

[DTrim99]

PR Review: Leave tax unit dependents' income out of Arkansas, Mississippi, New Jersey and Iowa income

Summary

The legal premise holds in all four states. I checked the statute and the claimed years' instructions before the code. A dependent with income files their own return. Only spouses combine income on one return. None of the four states' booklets has a Form 8814-style line or election for a child's income. The model doesn't implement Form 8814, so a dependent's income never reaches the filers' federal AGI. Zeroing a dependent's column therefore drops nothing that legally stays on the parents' return.

The AR, IA and NJ changes are correct, and so are the MS wage, interest and adjustment changes. I hand-checked 13 expectations and they all match.

One Mississippi regression needs fixing. MS counts capital gains through loss_limited_net_capital_gains_person, which limits and allocates the loss over every tax unit member, dependents included. On main, move_dependent_amounts_to_filer put the dependent's share back on a filer. The PR zeroes that share instead. So a dependent's capital gains now change the parent's Mississippi AGI, and the parent's own capital loss can escape the $3,000 limit. I confirmed this with a local probe at the PR head and at the merge base.

There are also three should-address items:

• the PR body has no axiom line;
• the new Worksheet D total-income guard has no test;
• three variables list the same source document more than once.

Critical (must fix)

C1. Mississippi: a dependent's capital gains change the parent's AGI, and the parent's capital loss can exceed the federal limit.

• Where: policyengine_us/variables/gov/states/ms/tax/income/ms_agi.py:38-39 (return where(is_dependent, 0, net_income)). It combines with policyengine_us/parameters/gov/states/ms/tax/income/income_sources.yaml:8 (loss_limited_net_capital_gains_person).
• Mechanism: loss_limited_net_capital_gains_person divides the person's gains by the sum over every member. When that sum is a net loss, it shares out loss_limited_net_capital_gains, which is also computed over every member. When the sum is a net gain, each person keeps their own unlimited amount. The per-person shares add up to the unit's figure. On main, MS moved the dependent's positive net income back onto a filer's column. With the dependent's share zeroed, the filers' columns no longer add to any loss-limited total.
• Law: the 2024 Form 80-100 instructions, line 40, PDF p. 11, say: "Mississippi generally follows IRS rules concerning computation of capital gains and losses. Capital loss deductions are subject to the same limitations as federal." The federal limit is $3,000 under 26 U.S.C. 1211(b). The parent's own return carries only the parent's gains and losses.
• Reproduction: a probe run with policyengine-core test on a three-case YAML. TY2025, MS, single parent aged 40 with $40,000 of wages and a $10,000 long-term capital loss. The child is 16 and a dependent.

Child's income	Law (parent's own return)	main (merge base)	This PR
none	ms_agi 37,000 (40,000 − 3,000)	37,000	37,000
$5,000 LTCG	37,000	37,000	34,000
$20,000 LTCG	37,000	50,000	30,000

• $5,000 case: the unit's net is −5,000, limited to −3,000. The parent's share is −3,000 × (−10,000 / −5,000) = −6,000. main added back the child's +3,000 share and got the right answer. The PR drops it.
• $20,000 case: the unit's net is +10,000, so nothing is limited and the parent keeps the full −10,000. That is $7,000 more loss than §1211(b) allows. main was wrong the other way: it added the child's $20,000 gain.
• Tax effect: with HoH exemptions, MS tax is 488.40 (5,000 case) and 312.40 (20,000 case), against 620.40 under the law. That is 4.4% of 27,100, 17,100 and 24,100 less the $10,000 zero bracket. The PR claims "A dependent's income no longer raises the filers' state AGI". Here a dependent's income lowers it.
• Fix:
  • For MS, use a filer-only, per-person loss-limited capital gain. Count each head's or spouse's own gains, limit the filers' net loss, and share it between head and spouse. Dependents get nothing.
  • Open sibling Leave dependents' capital income out of state lines on the filers' return #9894 adds exactly this as filer_loss_limited_net_capital_gains_person. Either add it here, or sequence this PR after Leave dependents' capital income out of state lines on the filers' return #9894 and switch MS income_sources.yaml to it.
  • Add the two child-gain cases above to ms_agi.yaml, each expecting ms_agi: [37_000, 0].
• Other states: AR, IA and NJ use each person's own capital amounts with no cross-member allocation, so they are not affected.

Should address

S1. No axiom line.

• The PR body has no axiom: line. CLAUDE.md "Axiom Parity" requires one for any policy fix. This changes the income base in four states, and the author is staff, so axiom: needed would not do either.
• Add axiom: <legal id> encoded-correct | <rulespec PR> encoded | <rulespec issue> queued. A queued issue must be OPEN, labelled pe-parity and dispatch-ready. Use rulespec-us#1467 as the model: module path, corpus citation, verbatim law (Ark. Code 26-51-801; 35 Miss. Admin. Code Pt. III, 10.01.104; N.J.S.A. 54A:8-3.1; Iowa Code 422.7 and 422.13(1)(a)), required outputs, and companion tests from the same external sources.

S2. Unpinned guard: Worksheet D total income now counts only age-eligible head and spouse.

• nj_other_retirement_income_exclusion.py:37 changes age_eligible * nj_total_income to eligible_person * nj_total_income (head or spouse only). The PR body explains why: "not an age-eligible dependent's".
• The new YAML case's dependent is 16, so age_eligible and eligible_person agree for them. No existing case has a dependent aged 62 or older. Reverting line 37 fails no test.
• The legal direction is right. The 2025 Worksheet D (PDF p. 23) says: "Joint filers: If only one spouse is 62 or older, only the income of that spouse can be excluded." A dependent is not on the filers' return at all.
• Fix: add a case with an age-eligible dependent who has income. Example:
  • Head aged 65 with $10,000 of private pension and $10,000 of interest. Dependent parent aged 70 with $20,000 of interest.
  • Expected: line 1 = min($20,000, $75,000), less the $10,000 pension exclusion, gives an unclaimed exclusion of $10,000.
  • The old rule counted $40,000 and gave $30,000.

S3. References repeat the same source document (CLAUDE.md reference rules).

• CLAUDE.md says: "One reference entry per source document; URLs that differ only by #page= are one source". For a multi-page cite in a variable tuple, use one href with a # PDF pages … comment above it.
• ar_agi_indiv.py:18,25 adds 2025 AR1000F instructions #page=10 next to the existing #page=12. Merge them into one href #page=10 with # PDF pages 10, 12.
• ia_net_income.py:19,31,32 adds 2022 IA 1040 instructions #page=4 and #page=8 to the bare media/2721/download?inline already in the tuple. Merge them into one href #page=4 with # PDF pages 4, 8.
• nj_agi.py:22,24 adds 2025 NJ-1040 instructions #page=5 and #page=40 as two entries. Merge them into one with # PDF pages 5, 40.
• The page numbers themselves are right. I checked each against the PDFs: AR p. 10 "Who must file" and p. 12 columns; IA p. 4 filing requirement (c) and p. 8 "Note to dependents"; MS p. 4 "Minor resident taxpayers"; NJ p. 5 filing thresholds and p. 40 Worksheet L.

Suggestions

1. Worksheet D anchor.
   • nj_other_retirement_income_exclusion.py:12 (existing) and the PR body cite 1040i.pdf#page=21. That is the printed page. Worksheet D and the "Joint filers" sentence are on PDF p. 23 of the 2025 booklet; Offset New Jersey claim of right repayments within their income category (N.J.S.A. 54A:5-2) #10198 changes the anchor to 23.
   • The current/1040i.pdf URLs (here and the new ones in nj_agi.py) are replaced every year, so the page anchors will drift. Prefer the dated archive copy.
2. Overlap section.
   • Offset New Jersey claim of right repayments within their income category (N.J.S.A. 54A:5-2) #10198 also edits nj_other_retirement_income_exclusion.py, on the same earnings = add(tax_unit, period, ["earned_income"]) line this PR changes. It subtracts nj_claim_of_right_wage_offset. The body says only that Offset New Jersey claim of right repayments within their income category (N.J.S.A. 54A:5-2) #10198 edits nj_gross_income.py.
   • Note it, so that whichever PR merges second keeps both changes: filers-only earnings, less a filers-only claim-of-right offset.
3. Follow-up issues for the declared deferrals. NJ ANCHOR, property tax relief and StayNJ income still sum nj_gross_income over every member. File them as a tracked issue so they aren't lost. (Iowa's ia_reportable_social_security still sums dependents' benefits before 2023; Leave dependents' capital income out of state lines on the filers' return #9894 covers it.)

Verified correct

• Law, researched first.
  • Arkansas:
    • Ark. Code 26-51-801(a)-(b) (each person over the gross income minimum files; spouses combine only "in a single joint return").
    • 2025 instructions PDF p. 10 (filing minimums by the filer's status) and PDF p. 12 ("write the primary's income in column A and the spouse's income in column B… all other filing statuses, write all income in column A only").
    • Lines 10, 11 and 22 have no child-income or 8814 item, and the AR regulation index has no minors' income rule.
    • The line 27 note "The $2,470 Standard Deduction does not apply to taxpayer's dependent(s)" assumes dependents file their own returns.
  • Mississippi:
    • 35 Miss. Admin. Code Pt. III, 10.01.104: "A minor is taxable on his wages, on the income he receives from the property he owns… On his return, the minor is entitled to his own deductions and exemptions". Expenses for income "includable in the gross income of the child (and not of the parents)" are treated as paid by the child.
    • 2024 instructions PDF p. 4: "Minor resident taxpayers – you have gross income in excess of the personal exemption plus the standard deduction".
  • New Jersey:
    • N.J.S.A. 54A:8-3.1(a) ($10,000 threshold), (b) (joint returns only for husband and wife) and (f) (a fiduciary files for a minor).
    • 2025 NJ-1040 instructions PDF p. 5 (thresholds, no dependent exception) and PDF p. 40, Worksheet L (household income adds "the total income… of your dependents. Include estimated income for any dependents who do not file a New Jersey tax return"). No 8814 text.
  • Iowa:
    • Iowa Code 422.7 ("net income" = federal AGI before NOL, with adjustments).
    • Iowa Code 422.13(1)(a) (a person "claimed as a dependent on another person's return" files with net income of $5,000 or more) and (2) (only "the combined net income of a husband and wife").
    • 2022 IA 1040 instructions PDF p. 4 (requirement (c)) and PDF p. 8 ("Note to dependents filing their own returns"). The 2021 and 2022 booklets have no 8814 text.
• No parent-side income dropped.
  • The model has no Form 8814 election: eitc_relevant_investment_income and net_investment_income document that a child's income reaches the parent only through 8814.
  • gov.irs.ald.filer_amounts_recorded_on_dependents lists filer amounts recorded on a dependent (adoption assistance, the savings bond exclusion). None of them is in the AR, IA, MS or NJ person-level income or adjustment lists, so zeroing a dependent's column loses no filer amount.
  • The one cross-member allocation in these lists is MS capital gains (C1).
• Readers are consistent.
  • Every reader of ar_agi_indiv/ar_agi_joint, ia_net_income and ms_agi either reads a filer's column or sums over the unit. That covers AR taxable income, itemized allocation, medical, misc and casualty floors and the low-income table; ia_prorate_fraction, ia_cdcc, ia_modified_income, state_agi; and ms_prorate_fraction and MS taxable income.
  • NJ's filing-threshold zeroing (nj_income_tax, nj_income_tax_before_refundable_credits), property tax deduction eligibility, the 529 income limit and the 2% medical floor now all read filers-only nj_agi, which matches the filers' line 29.
• Vectorization and entities:
  • where(is_dependent, 0, net_income) on Person.
  • tax_unit_non_dep_add on TaxUnit; the helper raises for other entities.
  • Bool × float masks in nj_other_retirement_income_exclusion.
  • No new int or enum keys, so the NH Add New Hampshire state supplementation for 2026 #10156 pattern doesn't apply.
• Mutation coverage: each new mask fails a test if removed, except S2. That covers the AR indiv and joint masks ([900, 0, 0], [900, −100, 0], gambling [50,000, 0]), the IA mask ([8,000, 0, 0]), the MS mask ([9,000, 10,000, 0]), nj_agi (40,000 vs 48,000), the NJ exclusion fraction (90,000 pension case), the NJ earnings test (the 30,000 case) and the NJ medical change (four YAML cases plus the conservation property).
• Hand-checked expectations (13):
  1. AR ar_agi_indiv [900, 0, 0]: 1,000 − 100; spouse max(100 − 200, 0) = 0; dependent 0.
  2. AR equal-AGI case [1,000, 1,000, 0].
  3. AR ar_agi_joint [900, −100, 0]: the spouse's −100 is not floored on a joint return.
  4. AR tax 957 (TY2025):
     • Taxable income: 40,000 − 2,470 standard deduction = 37,530.
     • Table midpoint 37,550 × 3.9% = 1,464.45, less the 419.96 reduction = 1,044.49, which rounds to 1,044.
     • Less 3 × $29 personal and dependent credits (HoH 2, dependent 1) = 957.
  5. IA ia_net_income [8,000, 0, 0] (10,000 − 2,000) and [8,000, 8,000, 0].
  6. MS tax 752.40: 40,000 − 3,400 − 8,000 − 1,500 = 27,100; 4.4% × 17,100.
  7. MS integration case (TY2021):
     • Separate returns: 10,200 and 8,200 taxable. Tax 240 (30 + 200 + 10) and 158 (30 + 128), 398 in all.
     • Combined return: 19,600 of deductions and exemptions, 10,000 of it to the head, so 10,000 and 8,400 taxable. Tax 230 + 166 = 396.
     • Both columns sit in the 4% band, so 396 = 60 + 4% × (18,400 − 10,000) is the minimum.
     • Allocations: 4,600 × 10,000 / 19,600 = 2,346.94 and 15,000 × 10,000 / 19,600 = 7,653.06.
  8. MS HSA ms_agi [38,500, 28,500, 0].
  9. NJ taxable income 37,500 = 40,000 − 1,000 − 1,500.
  10. NJ tax −80.68:
      • HoH uses the joint schedule: 1.4% × 20,000 + 1.75% × 17,500 = 586.25.
      • Federal EITC: 4,328 − 15.98% × (40,000 − 23,350) = 1,667.33. NJ EITC is 40% of that, 666.93.
      • 586.25 − 666.93 = −80.68.
  11. NJ pension case: filers' total income 90,000 ≤ 100,000, so the fraction is 1. Exclusion 90,000; unused other-retirement exclusion 0; nj_agi 0. With the child counted, 105,000 would give 50%.
  12. NJ other retirement exclusion 30,000:
      • HoH maximum 75,000; line 1 = min(40,000, 75,000) − 10,000 = 30,000.
      • The filer's earnings are 0 ≤ 3,000. With the child's 5,000 counted, the exclusion would be 0.
  13. NJ medical, dependent LTC case: max(0, 2,220 − 2% × 100,000) = 220. Cross-dependent case: nj_agi input 0, so 1,000.
• Housekeeping:
  • Changelog changelog.d/fix-9814-dependent-income-ar-ms-nj-ia.fixed.md is top-level with type fixed.
  • New cases are in variable-named YAML files.
  • Python edits only adjust existing property tests, with no new model builds.
  • No partner contract test files are touched.
• Source-conflict ratchet: no added or removed line matches CONFLICT_COMMENT_RE, and no touched file is in KNOWN_UNMARKED. Other main-side guards (Integrate modeled state unemployment insurance into income flows #9558, Count military SBP annuities in state benefit income lists #10317, Count each Social Security benefit once in Pennsylvania CCW countable income #10408) are not touched.

Axiom line assessment

Missing (S1). The PR body has no axiom: line and no rulespec-us link. A staff policy fix needs encoded-correct, encoded or a valid queued line; axiom: needed is not acceptable for staff.

CI status

All 36 checks pass at head 1db71d8539 (rechecked at the end of the review). The PR's own YAML and the two edited Python property files ran green in CI. C1 is a gap in coverage, not a CI failure. Its probe case isn't in the suite.

Branch status

• 2 ahead and 18 behind main. git merge-tree --write-tree PolicyEngine/main 1db71d8539 is clean. main has not changed any of the PR's 21 files since the merge base e53b426082.
• Sibling notes (branch notes only):
  • Leave dependents' capital income out of state lines on the filers' return #9894: a textual conflict only in the docstring of tests/core/test_state_tax_head_spouse_swap_invariance.py, as the body predicts. The rest auto-merges. Leave dependents' capital income out of state lines on the filers' return #9894 also adds filer_loss_limited_net_capital_gains_person, the variable the C1 fix needs.
  • Count only filer-paid premiums in Schedule A medical expenses #10270: conflicts in nj_medical_expense_deduction.py, its YAML and tests/core/test_medical_expense_premium_conservation.py. More conflicts come from Count only filer-paid premiums in Schedule A medical expenses #10270 being 495 behind main. The semantic clash is that Count only filer-paid premiums in Schedule A medical expenses #10270 keeps the dependents' SE ALD "since New Jersey counts the dependent's income", which this PR removes.
  • Offset New Jersey claim of right repayments within their income category (N.J.S.A. 54A:5-2) #10198: a textual conflict on the earnings = line of nj_other_retirement_income_exclusion.py (Suggestion 2). Its other conflicts come from being 495 behind main.
  • Leave dependents' own items out of DC, Delaware and Ohio subtractions #9939: no shared files. Its merge-tree conflicts come from being 785 behind main.
  • Apply New Jersey, North Dakota, Ohio, Oklahoma and Pennsylvania rules for a filer who can be claimed as a dependent #10043, Apply Iowa, Kansas, Louisiana and Maryland rules for a filer who can be claimed as a dependent #10045, Apply Arizona, Arkansas and California rules for a filer who can be claimed as a dependent #10053: clean merges with no shared files. Their tests set ar_agi_joint/ia_net_income as inputs or read NJ credits, so there is no semantic conflict.

Verdict: REQUEST_CHANGES

[DTrim99]

Requesting changes per the review above: 1 critical (Mississippi AGI zeroes the dependent's share of a capital-gain amount that is loss-limited across all members, so a child's gains change the parent's MS AGI and the parent's loss escapes the $3,000 limit), 3 should (no axiom line; unpinned NJ exclusion change; duplicate same-document references), 3 suggestions.

[hua7450]

Approving at 17e49be31d. I checked the follow-up commit against David's review.

Fixed

• Mississippi capital losses (David's critical). income_sources.yaml now uses filer_loss_limited_net_capital_gains_person. It limits only the head's and spouse's own net loss, and a dependent has none. I checked the five new ms_agi cases by hand:
  • Parent's $10,000 loss with no child gain, a $5,000 child gain or a $20,000 child gain: ms_agi 37,000 each time. The tax is 4.4% × (37,000 − 3,400 − 8,000 − 1,500 − 10,000) = 620.40.
  • Child's $10,000 loss against the parent's $5,000 gain: 45,000.
  • Spouses' $10,000 net loss limited to 3,000 and shared 8:2: [27,600, 19,400, 0].
• Other cross-member allocations in the MS income list. I found none that matter. taxable_social_security goes only to the head and spouse, and the retirement exemption subtracts it back out.
• Worksheet D guard (S2). It's now tested: the age-eligible dependent parent with $20,000 of interest gives an exclusion of 10,000, not 30,000.
• Duplicate references (S3). Merged into one href each for AR, IA, MS and NJ, with the Worksheet D anchor moved to PDF p. 23.

Nits (optional)

1. The PR body still has no axiom: line (David's S1).
2. The overlap section should say that 9894 adds the same filer_loss_limited_net_capital_gains_person.py. The section should also note that 10198 edits the earnings = line in nj_other_retirement_income_exclusion.py, so whichever PR merges second keeps both changes.
3. The Tests section of the body doesn't list the new MS capital gain cases or the Worksheet D case.