Count gambling winnings in federal gross income and fix the wagering losses deduction

[MaxGhenis]

Summary

Fixes #9637. Federal gross income left out gambling winnings while itemizers still deducted gambling losses, so gambling lowered tax. This PR:

1. counts gambling_winnings in gov.irs.gross_income.sources (26 U.S.C. 61(a); Schedule 1 line 8b);
2. makes wagering_losses_deduction the smaller of the allowed share of losses and the winnings, with a new parameter gov.irs.deductions.itemized.wagering_losses.rate: 1 through 2025 and 0.9 from 2026 (P.L. 119-21 § 70114);
3. counts only the head's and spouse's gambling in that deduction, as irs_gross_income already does for income (a dependent's gambling is on the dependent's own return);
4. moves the deduction from deductions_if_itemizing into the Schedule A list gov.irs.deductions.itemized_deductions (Schedule A line 16), and into the pre-2018 section 68(c) exclusions;
5. aligns the 16 states whose results the federal change would otherwise make wrong.

The issue left point 4 open ("decide separately"). It is needed here: once winnings are in AGI, every state that starts from federal itemized deductions would tax the winnings and deny the losses. With the deduction in the Schedule A total, those states carry it the way their returns do, the 2026 section 68 limitation counts it, and the itemization fallback comparison sees it.

The law

• Winnings. 26 U.S.C. 61(a): gross income is all income from whatever source derived. 2024 Schedule 1, line 8b "Gambling"; 2024 Form 1040 instructions, line 8b: "Enter any gambling winnings. Gambling winnings include lotteries, raffles…".
• Losses. 26 U.S.C. 165(d) allows wagering losses only to the extent of wagering gains. 2024 Schedule A instructions, line 16: gambling losses, "but only to the extent of gambling winnings reported on Schedule 1 (Form 1040), line 8b". That wording is also why a dependent's gambling does not count: it is not on the filer's return.
• 90% from 2026. P.L. 119-21 § 70114 rewrites 165(d)(1): the deduction "(A) shall be equal to 90 percent of the amount of such losses during such taxable year, and (B) shall be allowed only to the extent of the gains", for taxable years beginning after December 31, 2025. So the deduction is min(0.9 × losses, winnings).
• Section 68. Before 2018, 68(c) kept wagering losses out of the base of the 80% ceiling; the list limitation.excluded_deductions now names them. From 2026 the rewritten section 68 has no such exclusion, and the 2/37 reduction counts them.

Households, main and this branch

Single filer aged 40 in Texas with $60,000 of wages (the issue's cases), 2024, from evidence/state_flow_{base,branch}.json:

Winnings / losses	main: AGI, itemizes, income tax	Branch: AGI, itemizes, income tax	By hand (no state and local tax deduction)
none	$60,000, no, $5,216	$60,000, no, $5,216	$5,216
$10,000 / $0	$60,000, no, $5,216	$70,000, no, $7,241	$7,241
$10,000 / $8,000	$60,000, no, $5,216	$70,000, no, $7,241	$7,241: the $8,000 is below the $14,600 standard deduction
$30,000 / $25,000	$60,000, yes, $3,859	$90,000, yes, $9,123	$9,353 before PE's estimated sales tax deduction (YAML case)
$30,000 / $30,000	$60,000, yes, $3,259	$90,000, yes, $8,023	break-even gambling costs the standard deduction

On main a $5,000 net gain lowered tax by $1,357 and break-even gambling by $1,957. For 2026 the same $30,000 / $25,000 filer deducts $22,500 (90% of losses): main $3,839, branch $9,323.

States

Every state whose income tax reads federal AGI, gross income, taxable income or the federal itemized totals was reviewed (44 jurisdictions and Puerto Rico; eight research runs, ~/reviews/us-hub/gambling-9637/research/). I re-read each source cited for a change below. Changes:

State	Change	Authority
AL	Only filer and spouse winnings join Alabama gross income; new al_gambling_loss_deduction (losses up to winnings, no 90% share, outside the 2% limit)	2025 Form 40 booklet pp. 7, 21; Ala. Admin. Code r. 810-3-17-.01; ALDOR lists the 165(d) change as "Tied to Federal: No"
CA	New ca_wagering_losses_deduction replaces the federal amount: all losses up to winnings (California's Code is the IRC as of January 1, 2025), outside the high-income limitation	2025 Form 540 booklet pp. 62, 67–68; FTB conformity
DC	Wagering losses are outside the itemized deduction phase-out	D.C. Code § 47-1803.03(b-4)(2)(B)(iii)
DE	Itemized deductions carry the federal wagering losses deduction, allocated using only head and spouse losses	2025 PIT-RES instructions p. 12, line 16
HI	Itemized deductions include gambling losses, outside the overall limitation; hi_reduced_itemized_deductions returns the full total when the worksheet says "not limited" (it returned 0)	2025 N-11 instructions p. 19 and worksheet p. 34
IA	Before 2023, only filer and spouse winnings join Iowa gross income; from 2023 Iowa starts from federal taxable income	2021 IA 1040 line 13 instructions; 2022 instructions
KY	Wagering losses are disallowed for 2018 only; allowed again from 2019	Kentucky DOR gambling FAQ; 2019 Kentucky Acts ch. 151, §§37 and 83
KS	Itemized deductions leave out wagering losses (closed list)	K.S.A. 79-32,120(a)
MI	New mi_wagering_losses_deduction from 2021: the federal itemized amount, for federal itemizers	Treasury notice on 2021 PA 168; 2025 MI-1040 booklet p. 16
MN	Wagering losses are outside the itemized deduction limitation (also in the HF1938 contributed reform's copy of the formula)	Minn. Stat. 290.0122, subd. 2(d)(3) and subd. 8
MS	Itemized deductions leave out the federal amount: Mississippi income in PE has no gambling winnings, casino winnings bear a separate 3% tax, and Mississippi gaming losses come off Schedule A	2025 instructions pp. 16, 25
MT	New person-level mt_wagering_losses_deduction: attribute each spouse’s own losses first, then allocate the residual over uncovered winnings	2023 Form 2 instructions p. 39; 2025 instructions p. 13
NY	New ny_wagering_losses_deduction (2017 federal rules: all losses up to winnings), in the itemized total and outside the Pease worksheet's limited base	Tax Law § 615; 2025 IT-196 instructions p. 16, line 29
PA	New person-level pa_gambling_loss_offset: Pennsylvania taxes winnings net of each spouse's own wagering costs. PA-1000 rebate income no longer adds winnings on top of AGI, and nets the losses	72 P.S. § 7303(a)(7); 2025 PA-40 Schedule T; PA-1000 booklet p. 40
VA	Gambling losses enter the limited-deduction worksheet total and are protected from its reduction (the protected list held raw gambling_losses)	2025 Schedule A line 16a and worksheet lines 1–2
WI	Gambling losses do not count for the itemized deduction credit	Publication 600 pp. 4, 6, 9–10

Right after the federal change with no state edit: AZ, CO, CT, GA, ID, IL, IN, LA, MD, ME, MO, ND, NE, NM, OH, OK, OR, RI, SC, TN, UT, VT. Arkansas keeps its own rule (#9636): its gross income lists already count winnings and ar_gambling_loss_deduction has no 90% share. New Jersey's own gross income list already counts winnings. North Carolina's deduction (G.S. 105-153.5(a)(2)e, from 2025) is a separate PR stacked on this one.

Main-vs-branch differential

scripts/state_flow.py runs one single filer per state for 2021, 2024, 2025 and 2026, three wage levels, five winnings/losses pairs, with and without $25,000 of charitable gifts (1,530 households a year), on main (fe0c3a6) and on this branch.

• With no gambling, state_income_tax and income_tax are identical on main and the branch for every state, year and household (largest difference $0.00).
• On main, gambling moved state tax only through the federal itemization choice, in both wrong directions: for the $60,000 filer with $30,000 / $25,000 in 2024 it cut Colorado tax by $524, Iowa by $723 and South Carolina by $779, and raised DC tax by $949 and Virginia by $489.
• Examples from the branch's ordinary-source household grid for that filer in 2024: Illinois +$1,485 and Kansas +$1,674 (winnings taxed, no loss deduction); Michigan +$212.50 (4.25% of the $5,000 net, for a federal itemizer); Pennsylvania +$153.50 (3.07% of $5,000); Minnesota +$1,331.10 (6.8% of $30,000 less the $25,000 of losses above the standard deduction); Arkansas and New Jersey unchanged from main; Mississippi $0 (unmodeled).

"Gambling never lowers state tax" is not a law of every state, so it is not a test. The differential shows where it fails on the branch, each for a reason in the law or in an existing approximation:

• gifts above the AGI limit: more AGI lets more charity through (federal and every state that follows);
• Alabama deducts federal income tax, which rises when break-even gambling replaces the federal standard deduction;
• Georgia gives federal itemizers a $300 credit (O.C.G.A. § 48-7-29.23);
• PE estimates deductible state income tax from AGI, so winnings raise the federal and Hawaii state-tax deductions even when the state taxes only the net.

Invariants

For every winnings × losses × other-deductions grid point (test_gambling_winnings_and_losses_properties.py):

• the deduction equals min(share × losses, winnings), with the share restated from the statute (1 through 2025, 0.9 from 2026), is non-negative, never exceeds the winnings or the share of losses, and never falls as either rises;
• AGI rises by exactly the winnings;
• the deduction is in itemized deductions once, and an itemizer's taxable income is AGI less those deductions;
• holding other deductions fixed, income tax never falls as winnings rise and never rises as losses rise;
• holding other deductions fixed, gambling never lowers taxable income or income tax (including equal winnings and losses, the issue's case), and raises taxable income by at most the winnings, by exactly winnings less the deduction for a filer who itemizes either way;
• a dependent's winnings and losses change nothing on the filer's return.

Intended exceptions to "never lowers": deductions that grow with AGI (charitable gifts above the AGI limit, and the state income tax PE estimates from AGI). The tests fix the charitable and state and local tax deductions so the invariant is exact; the state differential above shows the exceptions.

test_itemized_deduction_limitation_properties.py now generates wagering losses as one of the 68(c) exclusions, so its worksheet restatement covers them for 2015–2017 and the 2026 base. test_ar_gambling_properties.py (unchanged) still checks that Arkansas's deduction equals the federal one through 2025.

Tests

Current head: 6c438507a582c9b44913b02bd371b58d75da963f, a merge commit bringing in main de3815f4ccec7c698f4c6c6a5e024acd79f3579b. It merges cleanly with current main.

Focused verification on this exact tree (84db243769): 221 passed across 30 files (187 YAML cases in 28 files; 34 Python tests in 2 files), one file per process with thread caps of 4.

• 29 files ran in the merge build's sequential runner (Python 3.14.7, local macOS arm64, under shared machine load).
• test_gambling_winnings_and_losses_properties.py crashed in that runner (exit -11 after 183 s) while the build was being stopped at its 6-hour limit. The hub re-ran it in the foreground on the same tree: 21 passed (log: hub-rerun-test_gambling_winnings_and_losses_properties.log in the hub's evidence for this merge).
• test_itemized_deduction_limitation_properties.py failed once in an interrupted first attempt and then passed (13) in the runner.

Merge resolution: main's #10068 added California's and Minnesota's non-disaster casualty and theft loss deductions in the same itemized-deduction formulas; both changes survive (each state's casualty deduction replaces the federal amount, and the permitted wagering deduction stays protected from the itemized-deduction reduction). All main YAML cases in the conflicted files keep their inputs and outputs. No expectation changed in the merge. The Alabama dependent-income case in al_agi.yaml is main's, not this PR's: main's #9943 changed its formula and expectation from $48,500 to $40,000 (parent-only AGI = $40,000 wages; dependent income and deductions no longer counted), and this branch now carries main's version. Partner contract tests were not edited.

File	Passed
test_itemized_deduction_limitation_properties.py	13
itemized_taxable_income_deductions_reduction.yaml	16
taxable_income_deductions_if_itemizing.yaml	3
wagering_losses_deduction.yaml	16
irs_gross_income.yaml	9
al_agi.yaml	12
al_gambling_loss_deduction.yaml	3
ca_itemized_deductions.yaml	8
ca_itemized_deductions_pre_limitation.yaml	13
dc_itemized_deductions.yaml	4
de_itemized_deductions_indv.yaml	11
de_itemized_deductions_unit.yaml	1
hi_itemized_deductions.yaml	5
hi_reduced_itemized_deductions.yaml	4
ia_gross_income.yaml	4
ks_itemized_deductions.yaml	4
ky_itemized_deductions_unit.yaml	5
mi_subtractions.yaml	8
mn_itemized_deductions.yaml	12
ms_itemized_deductions_unit.yaml	1
mt_wagering_losses_deduction.yaml	5
ny_itemized_deductions_phase_out.yaml	8
ny_itemized_deductions_phase_out_integration.yaml	2
ny_wagering_losses_deduction.yaml	3
pa_total_taxable_income.yaml	4
pa_property_tax_or_rent_rebate_income.yaml	10
va_itemized_deductions.yaml	8
wi_itemized_deduction_credit.yaml	3
mn_walz_hf1938.yaml	5
test_gambling_winnings_and_losses_properties.py	21 (hub re-run)

Not in this PR

Review findings that remain open for the US + core hub; the owner contested findings 1, 2, 5 and 7 as defects reproduced on main with non-gambling inputs. Main has since fixed finding 1 (#10136); findings 2, 5 and 7 remain pre-existing on main and are outside this PR. The original r1 response is preserved in hub-evidence/review-response-r1.md.

• 2026 section 68 in states that add up raw Schedule A components (AZ, DC, DE, KY, MD, ME, MT, NE, NM, OK, OR): the 2/37 reduction does not reach their itemized deductions for any component.

• Winnings by source. One gambling_winnings input cannot separate exempt state lottery prizes (CA; ID and VA prizes under $600; OR; NJ prizes of $10,000 or less; PA noncash prizes; MI pre-1989 installments), Massachusetts's licensed-wagering losses, Mississippi's three categories, West Virginia's in-state losses or Puerto Rico's exempt lotteries.

• Own income lists. Alabama's pooling is fixed on main by Leave tax unit dependents' income out of Alabama AGI #9943 (merged into this branch; al_agi.py uses tax_unit_non_dep_add, so al_gambling_winnings is now a harmless second exclusion); this PR excludes dependent winnings; Arkansas still pools a dependent's gambling (and other income) onto the head; New Jersey taxes gross rather than net winnings; New Mexico modified gross income and Oklahoma's credit gross income omit winnings; Michigan household resources lack the $300 exclusion; Washington's WFTC phase-out uses AGI where the statute uses earned income.

• Federal section 68 threshold (r1 finding 1): fixed on main. Main's Net the 2026 section 68 taxable-income leg of the QBI and other non-itemized deductions #10136, merged into this branch, makes itemized_taxable_income_deductions_reduction read taxable_income_before_itemized_deductions, which nets out QBID, tips, overtime, the senior deduction and car-loan interest; r1's example ($650,600 AGI, $20,000 QBID) now gives $0.

• AMT (r1 finding 2): AMTI still retains the regular section 68 reduction. This reproduces on main without gambling and awaits separate hub resolution.

• Hawaii threshold gate (r1 finding 9): left to Update Hawaii income tax for Act 35 (2025 IRC conformity) and fix its itemized deduction limit #9940. Whichever PR lands second must coordinate the wagering source and excluded-deduction list. Model the claim of right repayment deduction and credit (26 U.S.C. 1341) and North Carolina's rules that depend on it #10038 also overlaps this PR’s deduction lists and formulas.

• Stale comment (follow-up). taxable_income_before_itemized_deductions.py:29-36 still describes wagering losses as listed apart from the other itemized deductions and filters them out; they are now inside itemized_taxable_income_deductions, so the filter is a no-op. Harmless; to be cleaned up in a follow-up.

• Montana 2021-2023 joint allocation (open methodology choice). Each spouse keeps their own losses first (sourced in the 2023 Form 2 instructions); any remaining joint allowance is spread in proportion to each spouse's uncovered winnings, which the instructions do not state as a formula. Awaiting Max's decision (d1460).

Effects beyond income tax

Winnings now enter federal AGI and MAGI everywhere, so every AGI-based rule sees them: the EITC AGI test, the CTC phase-out, the premium tax credit and Medicaid MAGI, taxable Social Security, the 2025+ SALT-cap phase-down, IRMAA and state AGI-based credits. These follow the law. One gap: Medicaid spreads lottery and gambling awards of $80,000 or more over several months (42 U.S.C. 1396a(e)(14)(K)), which annual MAGI cannot show (listed in the follow-ups).

Section 165(d)(2) also counts any otherwise allowable deduction incurred in wagering as a wagering loss (added by Pub. L. 115-97 §11050 for 2018-2025 and made permanent by Pub. L. 119-21 §70114), and the 90% limit from 2026 applies to it. For the casual gamblers modelled here it adds nothing, since they have no such deductions.

Axiom

axiom: TheAxiomFoundation/rulespec-us#1658 TheAxiomFoundation/rulespec-us#1659 TheAxiomFoundation/rulespec-us#1660 TheAxiomFoundation/rulespec-us#1661 TheAxiomFoundation/rulespec-us#1662 TheAxiomFoundation/rulespec-us#1663 TheAxiomFoundation/rulespec-us#1664 TheAxiomFoundation/rulespec-us#1665 TheAxiomFoundation/rulespec-us#1666 TheAxiomFoundation/rulespec-us#1667 TheAxiomFoundation/rulespec-us#1668 TheAxiomFoundation/rulespec-us#1669 TheAxiomFoundation/rulespec-us#1670 TheAxiomFoundation/rulespec-us#1671 TheAxiomFoundation/rulespec-us#1672 TheAxiomFoundation/rulespec-us#1738 TheAxiomFoundation/rulespec-us#1739 queued

All seventeen parity drafts are filed: rulespec-us #1658-#1672 (federal 165(d) and fourteen states) and #1738 (Kentucky TY2018 exclusion) and #1739 (Iowa pre-2023 gross winnings), each under the hub's standing d1142 parity go.

Hub microsimulation impact (current head)

A real microsimulation on the default dataset, main at the merge base de3815f4cc against this PR's head 6c438507a5, for 2025 and 2026. The runs went one at a time under the shared heavy-job lock. Sums are weighted.

Output	2025 change	2026 change	2025 records changed (weighted)	2026 records changed (weighted)
gambling_winnings	+0	+0	0 (0)	0 (0)
wagering_losses_deduction	+0	+0	0 (0)	0 (0)
adjusted_gross_income	+0	+0	0 (0)	0 (0)
itemized_taxable_income_deductions	+0	+0	0 (0)	0 (0)
hi_reduced_itemized_deductions	+4.87bn	+4.83bn	1,233 (631,456)	1,221 (632,538)
state_income_tax	-2.19M	-2.24M	2 (118)	1 (118)
income_tax	+0	+0	0 (0)	0 (0)
household_net_income	+2.19M	+2.24M	2 (118)	1 (118)
household_benefits	+0	+0	0 (0)	0 (0)

The gambling variables do not move: the default dataset has no gambling inputs. Only hi_reduced_itemized_deductions moves among the itemized outputs (about 1,230 records). State income tax changes for 2 records in 2025 and 1 in 2026. The comparison files are national totals, so they don't show each record's state; the 2026 record matches the one Hawaii record in the pre-merge run, and federal tax, AGI and federal itemized deductions don't move in either year.

Comparison files are in fixes/10070-impact/ on the hub host.

[MaxGhenis]

Context from the original session (cbe2e2c2), for the US + core hub, which now owns this PR. I have not pushed to this branch since the hub adopted it.

Evidence at f7f555818f (the r1-fix commit the hub built on; all in ~/reviews/us-hub/gambling-9637/evidence/):

• YAML: 72 federal and 606 state cases passed in nine bounded batches (yaml_r3.log), including the new KY 2018, pre-2023 IA, AL and DE dependent, MT allocation and section 68 cases.
• Python: 61 passed (new property file, section 68 property file, Arkansas property file), 23.9 s of tests, 42 s wall, 3.35 GB peak RSS (pytest_r2.log). Isolated per-file cost: cost_r1.log.
• 51-state differential re-run at f7f555818f (state_flow_branch.json; scripts in ../scripts/): against the pre-fix branch only Iowa 2021 moved (winnings now in pre-2023 Iowa gross income). With no gambling, state and federal tax equal main everywhere.
• Microsimulation, 2026, main fe0c3a651c vs f7f555818f, populace_us_2024.h5@populace-us-2024-spm-20260909 (impact-g9637-2026.json): gambling inputs are zero on every record, so AGI, income tax and every touched state variable are unchanged, except Hawaii. The worksheet "not limited" correction raises hi_itemized_deductions on 7 records (+$83.1 million weighted) and lowers hi_income_tax on 1 record (−$2.24 million weighted; the same −$2.24 million in state_income_tax, +$2.24 million in household_net_income). Not re-run at 4bfb615ab7.

Review r1 (review-r1-out.md) and my response (review-response-r1.md): fixed AL and pre-2023 IA dependents, DE allocation, KY 2018, MT allocation, input docs, test placement, reference titles. Findings 1, 2, 5 and 7 (section 68 with QBID, AMT keeping section 68, states summing raw Schedule A components, WA WFTC base) reproduce on main with no gambling (preexisting_main.log), so I answered them as separate work; write-ups for follow-ups in ../followups.md (A–F, including the #9940 Hawaii and #10038 overlaps).

Axiom: rulespec-us#1658–#1672 were filed from drafts written before the r1 fixes. Not covered by them: Kentucky's 2018 exclusion (gov.states.ky.tax.income.deductions.itemized.wagering_losses_allowed), pre-2023 Iowa winnings (ia_gambling_winnings, IA 1040 line 13), and the head-and-spouse-only Alabama winnings (al_gambling_winnings), Delaware allocation and Montana allocation in their filed issues. My decision d1271 duplicated d1142 and is annotated as such.

Stack: North Carolina's wagering deduction is #10110 (draft), on top of this PR and #10028.

[MaxGhenis]

Heads-up on an overlap with #9977 (draft, §163(d) limit), from the session that wrote #9695.

Both PRs edit gov.irs.deductions.itemized.limitation.excluded_deductions and tests/core/test_itemized_deduction_limitation_properties.py:

• This PR adds wagering_losses_deduction to the list. That is right once the deduction moves into itemized_deductions.
• Limit investment interest deductions under §163(d) and Form 4952 #9977 makes interest_deduction carry investment_interest_expense_deduction (Form 4952 line 8) instead of interest paid. The list then has to name that variable instead of investment_interest_expense. It doesn't yet; the tested patch is in Limit investment interest deductions under §163(d) and Form 4952 #9977 (comment).

Whichever lands second will conflict on the list and the property test, and the resolved list needs all four entries:

    - medical_expense_deduction
    - investment_interest_expense_deduction
    - casualty_loss_deduction
    - wagering_losses_deduction

If this PR lands first, nothing changes here: investment_interest_expense stays correct until #9977 changes interest_deduction.

This PR also covers two of the follow-ups noted in #9695, wagering losses in the 2026 2/37 base and the 90% limit, so I'm not pursuing those separately.

[MaxGhenis]

Coordination with #10136, which fixes the 2026 section 68 taxable-income leg (#9170). Both PRs edit two files, and the overlaps are textual only:

• tests/.../itemizing/itemized_taxable_income_deductions_reduction.yaml: both append cases at the end. Keep both sets.
• tests/core/test_itemized_deduction_limitation_properties.py: both edit the docstring after item 7, and Net the 2026 section 68 taxable-income leg of the QBI and other non-itemized deductions #10136 also changes _simulate's signature and adds 2026 tests. Keep both edits.

Semantics after both land: #10136's new taxable_income_before_itemized_deductions (section 68(a)(2)'s taxable income, increased by itemized deductions) subtracts every deductions_if_itemizing entry except itemized_taxable_income_deductions and wagering_losses_deduction. Wagering losses are itemized deductions, so leg (2) adds them back. Once #10070 moves wagering_losses_deduction into itemized_deductions, that exclusion is a harmless no-op. Leg (1) then picks up the losses through total_itemized_taxable_income_deductions, which is #10070's fix. #10136's property test raises wagering_losses_deduction and checks that leg (2) does not move, which holds under either layout.