Keep dependents' above-the-line deductions off the filer's AGI

[MaxGhenis]

Follows #9646/#9647, whose "Not in this PR" list named the other above-the-line deductions, starting with self_employment_tax_ald.

What changes

irs_gross_income already leaves out every tax unit dependent's income. The above-the-line deductions did not. A dependent with Schedule C income added nothing to the parent's gross income, but half of the dependent's self-employment tax came off the parent's AGI. The same was true of these items of the dependent's:

• self-employed health insurance;
• self-employed retirement plan deductions;
• IRA contributions;
• early withdrawal penalty;
• educator expenses;
• alimony paid.

Item in gov.irs.ald.deductions	Level	Before	After
self_employment_tax_ald, self_employed_health_insurance_ald, self_employed_pension_contribution_ald	tax unit, adds its _person variable	every member	tax_unit_non_dep_sum of the _person variable
alimony_expense_ald	tax unit formula over members	every member	non-dependents' alimony_expense_ald_person (new)
early_withdrawal_penalty, educator_expense, traditional_ira_contributions	person	summed over every member by above_the_line_deductions' adds	head and spouse only
qualified_adoption_assistance_expense, us_bonds_for_higher_ed	person	every member	every member, deliberately: new parameter gov.irs.ald.filer_amounts_recorded_on_dependents
capped_qualified_tuition_expenses_ald	tax unit	sums every member's tuition	unchanged (a parent takes a dependent's tuition: Form 8917)
student_loan_interest_ald	person	already 0 for dependents (§221(c))	unchanged
loss_ald	tax unit	already masked (#9647)	unchanged
health_savings_account_ald, domestic_production_ald, specified_possession_income, puerto_rico_income	tax-unit inputs	as is	unchanged

above_the_line_deductions is now a formula: tax_unit_non_dep_add(tax_unit, period, gov.irs.ald.deductions, include_dependents=gov.irs.ald.filer_amounts_recorded_on_dependents). tax_unit_non_dep_add gained the include_dependents argument and still adds the variables in the order given.

Two amounts belong to the filer wherever they are recorded:

• Employer adoption assistance is the adopting employee's exclusion (§137(a)), but the Ohio adoption credit reads it on the adopted child (oh_adoption_credit_person).
• Education savings bond interest is excluded only for the bond's owner, who must have been at least 24 when the bond was issued, and the expenses may be a dependent's: "A bond bought by a parent and issued in the name of his or her child under age 24 does not qualify for the exclusion by the parent or child" (Form 8815, 2025, p. 3). By convention, an amount recorded on a dependent is a parent-owned bond's exclusion recorded on the student whose expenses it paid.

Masking either would drop the parent's exclusion.

The _person variables keep each person's own amount, and none is renamed. A dependent's own return reads them in dependent_taxable_ss_magi, which now also counts the dependent's own pre-2019 alimony.

Metadata. above_the_line_deductions and the three SE aggregates now have formulas instead of adds, so their adds metadata (exported by policyengine-api and read by policyengine.py) is empty. policyengine-core rejects a variable that declares both a formula and adds.

Readers

These sum the ALD list themselves and now use the same non-dependent sum:

• taxable_ss_magi;
• taxable_uc_agi;
• student_loan_interest_ald_magi;
• adjusted_gross_income_person, which also masks dependents' _person SE deductions, so a dependent's per-person AGI is 0 rather than negative;
• ma_part_b_agi.

These had to change with it:

• mo_adjusted_gross_income subtracted an all-member personal-ALD total from above_the_line_deductions. Without this change, a dependent's Schedule SE deduction would raise the head's Missouri AGI, to 40,706.48 in the test case. It now takes personal ALDs over non-dependents.
• medicaid_adjusted_gross_income_person gave a dependent's IRA contribution, early withdrawal penalty and alimony to the head. It now subtracts a dependent's own person-level deductions from the dependent's own AGI, as 42 CFR 435.603(e) takes each person's AGI. The head's Medicaid AGI rises accordingly; the sharing rule between head and spouse is unchanged. The SE deductions and alimony are attributed through person_share_of_tax_unit_amount, and only when they are in the active gov.irs.ald.deductions list. A tax-unit input or a reform that removes one therefore reaches the per-person AGI; before, the SE _person amounts were subtracted even when a reform removed them from the list.
• ma_gross_income_loss_adjustment added back every member's business losses, while the federal gross income it adjusts leaves dependents out. The property tests found this. It now uses tax_unit_non_dep_add.
• Alabama (al_agi) and New Jersey (nj_medical_expense_deduction) count every member's income, dependents included. They now also subtract the new dependents_self_employed_health_insurance_ald and dependents_self_employed_pension_contribution_ald, so both match main.
• Mississippi (ms_agi_adjustments) is a per-person adds. It projected the tax unit's SE health insurance and plan deductions onto every member, so each spouse subtracted both spouses' amounts. It now lists ms_self_employed_health_insurance_adjustment and ms_self_employed_retirement_adjustment, built with the new person_share_of_tax_unit_amount helper. Each person gets their own _person amount. When a tax-unit input or a reform of the aggregate makes it differ from the head's and spouse's own amounts, their amounts are scaled in proportion to sum to it. With non-negative amounts, as for these deductions, no share turns negative. If they have no own amount, the head takes all of it. Either way, the aggregate counts once. This also changes Mississippi for spouses without dependents (finding 9 of review r1).
• Minnesota M1PR (mn_homestead_credit_refund_household_income) starts from federal AGI, but added back every member's IRA contributions. Its retirement additions, contributions and compensation now count the head and spouse only.

Checked, consistent with no edit:

• mt_applicable_ald_deductions, mn_social_security_subtraction, mn_walz_hf1938;
• pa_property_tax_or_rent_rebate_income (AGI plus ALDs, so the net is unchanged);
• mn_marriage_credit, nd_mpc, eitc_earned_income;
• the per-person AGI readers: KY, MT, DE, VA, OH, DC, MN renters, UT renter relief (adult dependents already count with AGI 0), withholding.

mi_household_resources is left to #9790's stack.

Law

Statutes are from the pinned Axiom corpus (axiom-corpus 8f7d60aac), which matches OLRC. IRS PDFs were extracted with pdftotext -layout.

• Only spouses aggregate on one return. §6013(a): "A husband and wife may make a single return jointly of income taxes under subtitle A, even though one of the spouses has neither gross income nor deductions".
• Each deduction belongs to the individual who has it.
  • §164(f)(1): "In the case of an individual, ... there shall be allowed as a deduction for the taxable year an amount equal to one-half of the taxes imposed by section 1401".
  • §219(a): "In the case of an individual, there shall be allowed as a deduction an amount equal to the qualified retirement contributions of the individual for the taxable year."
  • §62(a)(6) allows §404 "In the case of an individual who is an employee within the meaning of section 401(c)(1)".
• Each person files their own Schedule SE. Schedule SE (Form 1040) 2025 is headed "Name of person with self-employment income". Line 13: "Deduction for one-half of self-employment tax. Multiply line 12 by 50% (0.50). Enter here and on Schedule 1 (Form 1040), line 15" (https://www.irs.gov/pub/irs-prior/f1040sse--2025.pdf).
• The one election that moves a child's income does not move the child's deductions. Instructions for Form 8814 (2025), p. 2, "Tax benefits you cannot take": "you cannot take certain deductions that your child could take on his or her own return such as: ... Penalty on early withdrawal of child's savings" (https://www.irs.gov/pub/irs-prior/i8814--2025.pdf). Dependents' losses reduce the filer's AGI and raise the filer's §461(l) limit (loss_ald, limited_capital_loss) #9646 quotes §1(g)(7)(B) in full: it moves gross income only.
• The worksheets read the filer's own Schedule 1. Line 6 of the 2025 Social Security Benefits Worksheet and line 3 of the Student Loan Interest Deduction Worksheet say "Enter the total of the amounts from Schedule 1, lines 11 through 20, and 23 and 25" (https://www.irs.gov/pub/irs-prior/i1040gi--2025.pdf#page=32, #page=99).
• The parent takes a dependent's tuition. Form 8917 (Rev. January 2020): "in order to claim the deduction for education expenses for a dependent, you must have paid the expenses in the current year and must claim the student as a dependent on your current year tax return".

Tests

dependent_above_the_line_deductions.yaml: 9 cases. Expected values are Schedule SE, Schedule 1, Form 1040 and Form 8917 arithmetic in cents, with each line commented. The first six fail on main. The last three pin amounts that stay on the parent's return, and pass on both.

Case	Output	main	This PR = forms
HoH parent wages 40,000; child aged 17, Schedule C 10,000	adjusted_gross_income	39,293.52	40,000
	child's Schedule SE line 12 / line 13	1,412.96 / 706.48	1,412.96 / 706.48 (on the child's return)
Joint: wages 60,000; spouse Schedule C 20,000; child Schedule C 8,000, SE health insurance 1,000, SEP 500	self_employment_tax_ald	1,978.14	1,412.96 (spouse only)
	adjusted_gross_income	76,521.86	78,587.04
HoH parent 50,000; student dependent aged 23: IRA 2,000, early withdrawal penalty 100, educator 300	adjusted_gross_income	47,600	50,000
Dependent mother pays pre-2019 alimony of 3,000	adjusted_gross_income	47,000	50,000
Grandparent: Social Security 24,000, pension 20,000; grandchild Schedule C 10,000	tax_unit_taxable_social_security	3,146.76	3,500 (worksheet lines 1–18)
Parent wages 91,000, student loan interest 2,500; dependent's IRA 2,000	student_loan_interest_ald	1,833.33	1,500 (worksheet lines 1–9)
TY2020: dependent student's tuition 3,000	capped_qualified_tuition_expenses_ald	3,000	3,000 (Form 8917 lines 1–6)
Adoption assistance 5,000 recorded on the adopted child	adjusted_gross_income	45,000	45,000
Bond interest 1,000 recorded on the student	adjusted_gross_income	51,000	51,000

Reader cases (TY2025):

Case	main	This PR
mo_adjusted_gross_income, parent and child as in the first case	[40,000, 0]	[40,000, 0]; reverting the MO change gives 40,706.48
ma_part_b_agi, child with SE health insurance 1,000	39,000	40,000
ma_gross_income, child's Schedule C loss −3,000	39,000	42,000
medicaid_adjusted_gross_income_person, student as in the third case	[47,600, 6,500]	[50,000, 4,100] (each person's own Form 1040 line 11)
medicaid_adjusted_gross_income_person, dependent mother: pension 4,000, alimony 3,000	[47,000, 4,000]	[50,000, 1,000]
al_agi, child with Schedule C 10,000, SE health insurance 1,000, SEP 500	48,500	48,500
nj_medical_expense_deduction, the same child	1,000	1,000
ms_agi, the same child (the child's income goes to the parent)	[46,293.52, 0]	[47,793.52, 0]; the child's 2,206.48 of adjustments subtracted once
ms_agi, joint without dependents: head with Schedule C 30,000 and SE health insurance 3,000	[24,880.57, 17,000]	[24,880.57, 20,000]
mn_homestead_credit_refund_household_income, dependent's IRA 2,000	45,720	45,720

contrib/taxsim/taxsim_v10.yaml, tax unit 16, now expects 252,502 (joint wages). The 500 of Schedule C income belongs to a dependent, whose deduction is on their own return, and TAXSIM's own record has no self-employment income.

tests/core/test_dependent_above_the_line_deductions_invariants.py also has five deterministic tests:

• a tax-unit alimony, SE health insurance and SEP input reaches Medicaid AGI and the Mississippi adjustments;
• a reform that drops alimony from the ALD list removes it from both federal and Medicaid AGI;
• a joint Mississippi couple's tax-unit SE health insurance deduction set to 0, and set to 500 below the spouse's own 1,000, gives non-negative shares;
• a reform that zeroes the aggregate's formula does the same.

In the last two, Mississippi AGI, Medicaid MAGI and Medicaid household income all equal those of a couple whose own amount is that deduction. The first two fail at 49d6476; the last two fail at 49c9a3a. The original ms_agi_adjustments.yaml inputs (tax-unit aggregates, 500) are unchanged from main and pass.

The property tests. It runs Hypothesis (5 batches of 10–40 tax units) and a seeded population of 200 units, for TY2020 and TY2025:

• units are single, head of household or joint, with and without 1–2 dependents aged 5–23;
• states are TX, MO, MA and MN;
• each batch runs twice, with the dependents' own-return deduction inputs as drawn and set to zero.

The invariants hold for every unit:

1. Dependents never reach the filer's return. The dependents' deduction inputs never change any of the filer's:
   • filing status, AGI and above_the_line_deductions;
   • SE and alimony aggregates;
   • Social Security and unemployment MAGIs, and taxable Social Security;
   • MA gross income and Part B AGI;
   • MN M1PR household income;
   • the head's and spouse's per-person AGI, student loan interest MAGI and deduction, Medicaid AGI and Missouri AGI.
2. Accounting identities. Members' adjusted_gross_income_person sum to AGI, and so do the head's and spouse's Medicaid AGIs.
3. Differential. The checks run against independent numpy sums:
   • above_the_line_deductions = loss_ald + tuition + the head's and spouse's person-level deductions + every member's adoption and bond amounts;
   • each aggregate = its head-and-spouse sum;
   • AGI = Σ irs_gross_income minus that reference;
   • for units without dependents, above_the_line_deductions equals the previous all-member sum;
   • otherwise it lies between 0 and that sum.
4. A dependent's own amounts survive.
   • self_employment_tax_ald_person = ½ × self_employment_tax for everyone.
   • alimony_expense_ald_person = alimony × (divorce before 2019).
   • A dependent's Medicaid AGI = their own gross income less their own deductions.

Mutation check (head 1a97062). Each changed source file was reverted to main one at a time:

Reverted to main	Seeded property tests (2020, 2025)	Touched YAML cases
self_employment_tax_ald	2 of 2 fail	4 fail
self_employed_health_insurance_ald	2 of 2 fail	2 fail
self_employed_pension_contribution_ald	2 of 2 fail	1 fail
alimony_expense_ald	2 of 2 fail	1 fail
above_the_line_deductions	2 of 2 fail	2 fail
adjusted_gross_income_person	2 of 2 fail	3 fail
student_loan_interest_ald_magi	2 of 2 fail	1 fail
taxable_ss_magi	2 of 2 fail	0 fail
taxable_uc_agi	2 of 2 fail	0 fail
ma_part_b_agi	2 of 2 fail	0 fail
ma_gross_income_loss_adjustment	2 of 2 fail	1 fail
mo_adjusted_gross_income	2 of 2 fail	1 fail
medicaid_adjusted_gross_income_person	2 of 2 fail	1 fail

The taxable_ss_magi, taxable_uc_agi and ma_part_b_agi YAML cases read the SE aggregates, which are already masked upstream. Only the person-level items in the property tests reach those three files on their own.

Local runs at 7bcaed0.

• Property and input/reform tests: 9 passed.
• test_ald_determinism.py: 9 passed. It now intercepts tax_unit_non_dep_add, so it checks that the tax-unit sums go through it in sorted, deduplicated order. Its float32 hash-seed probe is unchanged.
• YAML: 1,137 passed across:
  • gov/irs/income/taxable_income/adjusted_gross_income;
  • AL, MS and MA income, NJ deductions, MN M1PR, MO AGI;
  • hhs/medicaid;
  • contrib/taxsim/taxsim_v10.yaml.

Population effect

PE-US at this branch (7bcaed0) against main (e8f9487). Both runs use the default dataset (populace_us_2024.h5@populace-us-2024-spm-20260909) for TY2025 with policyengine-core 3.32.12. All figures are engine output, weighted.

• AGI rises in 391 tax units (1.12M weighted), by $1.074bn. No unit's AGI falls. Every unit whose AGI changes has a dependent with one of these deductions. The dependents' amounts in the data are:
  • Schedule SE deductions, $273M in 132 units;
  • IRA contributions, $120M in 140 units;
  • alimony, $657M in 3 units;
  • educator expenses, $24M in 121 units;
  • SE health insurance, $0.14M in 5 units.
• Federal income tax rises by $217M in 307 units (0.92M weighted). Refundable CTC falls by $27.6M in 26 units. ctc_social_security_tax, which adds self_employment_tax_ald (§24(d)(2)), falls by $273M. The EITC falls by $0.6M in 15 units.
• Taxable Social Security rises by $0.25M in 5 units.
• State income tax rises by $56.7M in 228 units and falls in 5; CA accounts for $49.2M of the rise. Twelve of those units have no dependent deduction amounts:
  • 11 are Mississippi's double-count fix (+$0.33M);
  • 1 is a Massachusetts unit (weight 0.003). The only Massachusetts change outside the deductions is the business-loss adjustment.
• Medicaid AGI: heads and spouses +$801M, dependents −$800M. Medicaid eligibility changes for 6 people with a combined weight of 0.002. 1,249 units without dependent amounts differ by at most $0.50 (+$6,667 weighted). This is float32 rounding: the per-person split now follows the parameter list's order and reconciles to the tax-unit amount.
• Peak memory: 12.3 GB on main and 15.5 GB on this branch.

Scripts and outputs: ~/reviews/pe-us-dependent-ald-2026-10-05/{scripts,microsim}.

These figures are from a run at 7bcaed0; the final head, dd9a054, changes only a docstring. An earlier run at 49d6476, before the per-person reconciliation, gave identical values for every tax output above. Only Medicaid AGI differed: head-and-spouse totals by at most $0.50, and one dependent's by less than $0.001. The dataset (populace-us-2024-spm-20260909) stores none of self_employment_tax_ald, self_employed_health_insurance_ald, self_employed_pension_contribution_ald, alimony_expense_ald or above_the_line_deductions.

Invariants (stated for review)

• For every tax unit, a dependent's own-return deduction inputs never change the filer's AGI or any tax-unit ALD sum.
• adjusted_gross_income = Σ irs_gross_income − above_the_line_deductions, and Σ adjusted_gross_income_person = adjusted_gross_income, without the taxable UBI reform.
• 0 ≤ above_the_line_deductions ≤ the previous all-member sum, with equality for units without dependents.

Not in this PR (routed to the US + core hub)

• ctc_social_security_tax still adds every member's employee payroll tax. That fix is Count only the filer's and spouse's payroll taxes in the refundable CTC and Additional Medicare Tax, and in the CRFB surtax base #9825, which lands after this PR.
• The CRFB AGI surtax's expanded base also goes in Count only the filer's and spouse's payroll taxes in the refundable CTC and Additional Medicare Tax, and in the CRFB surtax base #9825.
• educator_expense enters the list uncapped. Cap the educator expense deduction per eligible educator (26 U.S.C. 62(a)(2)(D)) #9833 adds the §62(a)(2)(D) cap. Whichever lands second updates PERSON_DEDUCTIONS in this PR's invariant test.
• adjusted_gross_income_person still splits person-level ALDs 50/50 between spouses. This is pre-existing and queued at the hub.
• Mississippi still projects the tax-unit health_savings_account_ald onto every member. This is pre-existing, with no person-level HSA variable.
• qualified_business_income_deduction sums every member's qbid_amount. This is pre-existing.
• Head and spouse inference ignores an input is_tax_unit_dependent. The adult-dependent YAML cases set is_tax_unit_spouse: false; Exclude input dependents from tax unit head and spouse inference #9630 fixes this.

axiom: TheAxiomFoundation/rulespec-us#1494 queued. rulespec-us us/statutes/26/62.yaml takes the return's deduction total as an input and defers the §62(a) composition. 164/f.yaml gives the per-person §164(f) deduction. The issue asks to sum the individual deductions over taxpayer_or_spouse_of_tax_unit, with companion tests from the same Schedule SE and Form 1040 arithmetic.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.12%. Comparing base (909176a) to head (1a97062).
⚠️ Report is 199 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##              main    #9801      +/-   ##
===========================================
- Coverage   100.00%   98.12%   -1.88%     
===========================================
  Files            4       13       +9     
  Lines           76      266     +190     
  Branches         2        7       +5     
===========================================
+ Hits            76      261     +185     
- Misses           0        3       +3     
- Partials         0        2       +2     

Flag	Coverage Δ
unittests	98.12% <100.00%> (-1.88%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[MaxGhenis]

Hand-off to the US + core hub. Head dd9a0542443adfed87b2f8d51c97b447e59efd2d. Only CI remains.

Review. Independent review, all rounds at ~/reviews/pe-us-dependent-ald-2026-10-05/review/:

Round	Head	Verdict	Report
r1	1a97062	REQUEST_CHANGES	review-r1.md
r2	49d6476	REQUEST_CHANGES	review-r2-full.md
r3	49c9a3a	REQUEST_CHANGES	review-r3-full.md
r4	7bcaed0	APPROVE	review-r4-full.md
r5	dd9a054	APPROVE	review-r5-full.md

• r1 was Opus 5.5; r2–r5 were GPT-6.1 Sol, because the Claude lanes were rate-limited.
• r5 is a delta review: dd9a054 changes only a docstring, per r4's P3 note.
• Every finding was fixed: the AL/MS/NJ/MN readers, the TAXSIM expectation, the adoption and bond amounts recorded on dependents, per-person alimony, and the Mississippi/Medicaid reconciliation to tax-unit inputs and reforms. No negative shares remain.

Passed.

• CI was fully green at 49d6476: 35/35, including the previously cancelled contrib shard and taxsim_v10. CI at dd9a054 is pending.

• Locally at 7bcaed0:

  • 9 property and input/reform tests;
  • 1,137 YAML cases (federal AGI, AL, MS, MA, NJ, MN M1PR, MO, Medicaid, TAXSIM v10);
  • test_ald_determinism.py 9/9.

• Mutation check: reverting each of the 13 original changed files fails at least one test.

• Microsim at 7bcaed0, TY2025, default dataset:

  • AGI +$1.074bn in 391 units, none falls;
  • federal income tax +$217M;
  • state income tax +$56.7M.

  Scripts and outputs are in ~/reviews/pe-us-dependent-ald-2026-10-05/{scripts,microsim}.

Merge order and dependencies.

• Count only the filer's and spouse's payroll taxes in the refundable CTC and Additional Medicare Tax, and in the CRFB surtax base #9825 (CTC payroll taxes) is approved on condition that it lands after this PR.
• Cap the educator expense deduction per eligible educator (26 U.S.C. 62(a)(2)(D)) #9833 (educator cap): whichever of the two lands second changes PERSON_DEDUCTIONS in test_dependent_above_the_line_deductions_invariants.py, from educator_expense to educator_expense_ald_person.
• Add back income excluded under sections 911, 931 and 933 in modified AGIs (student loan interest and 11 siblings) #9817 overlaps taxable_ss_magi, taxable_uc_agi and student_loan_interest_ald_magi; the second to land rebases.
• Count only the claimant and spouse in Michigan household resources (MI-1040CR lines 14-27, 30 and 31) #9796 (Michigan line 30): its session was unreachable at the end. If it builds the ALD list with tax_unit_non_dep_add, it should pass include_dependents=gov.irs.ald.filer_amounts_recorded_on_dependents.

Axiom parity. TheAxiomFoundation/rulespec-us#1494 is queued (pe-parity), with population numbers in a comment.

No Max decision needed. This is a bug fix that moves aggregates, which merges on the gates.

New follow-ups for the hub, both pre-existing and outside this PR:

• Mississippi still projects the tax-unit health_savings_account_ald onto every member's ms_agi_adjustments. It is a double count for couples, and there is no person-level HSA variable.
• qualified_business_income_deduction sums every member's qbid_amount, dependents included (r1 finding 9).

[MaxGhenis]

Merged by the US + core hub on its merge gates. Audit:

• Head: dd9a054244.

• Independent review: r1 (Opus 5.5) through r5 (GPT-6.1 Sol). r4 APPROVE at 7bcaed0; r5 approves the docstring-only delta to this head. Reports are in ~/reviews/pe-us-dependent-ald-2026-10-05/review/. Every r1–r3 finding is fixed: the AL, MS, NJ and MN M1PR readers, the TAXSIM expectation, adoption and bond attribution, per-person alimony, and the MS/Medicaid reconciliation.

• CI: every check passed at this head.

• Impact: TY2025, default dataset:

  • AGI +$1.074bn in 391 units, none falling;
  • federal income tax +$217M;
  • state income tax +$56.7M.

  This is a bug fix that brings dependents' deductions in line with the law (they belong on the dependent's own return), so it merges on the gates whatever its aggregate size.

• Order: Count only the filer's and spouse's payroll taxes in the refundable CTC and Additional Medicare Tax, and in the CRFB surtax base #9825 lands after this. Cap the educator expense deduction per eligible educator (26 U.S.C. 62(a)(2)(D)) #9833, Add back income excluded under sections 911, 931 and 933 in modified AGIs (student loan interest and 11 siblings) #9817 and Count only the claimant and spouse in Michigan household resources (MI-1040CR lines 14-27, 30 and 31) #9796 each rebase onto it.