Add the section 911(a)(1) exclusion to NIIT modified AGI (26 U.S.C. 1411(d))

[MaxGhenis]

Summary

niit_magi returned adjusted gross income plus the estate and trust code H change and never added the amount excluded under section 911. 26 U.S.C. 1411(d) defines modified adjusted gross income as AGI "increased by the excess of— (1) the amount excluded from gross income under section 911(a)(1), over (2) the amount of any deductions (taken into account in computing adjusted gross income) or exclusions disallowed under section 911(d)(6) with respect to the amounts described in paragraph (1)" (repeated in Treas. Reg. 1.1411-2(c)(1)).

Reproducer (2025, single, Texas, age 45): wages $150,000, qualified dividends $50,000, tax-unit foreign_earned_income_exclusion $100,000.

	Before	After
adjusted_gross_income	200,000	200,000
niit_magi	200,000	300,000
net_investment_income_tax	0	1,900 (3.8% × min(50,000, 300,000 − 200,000))

Earned income only, not housing

Section 1411(d)(1) cites section 911(a)(1), the foreign earned income. Section 911(a)(2), the housing cost amount, is a separate exclusion. The 2025 Form 8960 Line 13 MAGI Worksheet (instructions, PDF page 23) matches: line 2(a) is "your foreign earned income exclusion (from line 42 of Form 2555)", less line 2(b), "the deductions reported on line 44 of Form 2555 allocable to your foreign earned income exclusion". Neither the housing exclusion (Form 2555 line 36) nor the housing deduction (line 50) is added back. The 2024 worksheet is the same.

The model can't separate them on its own. foreign_earned_income_exclusion is line 2c of the Form 1040 Foreign Earned Income Tax Worksheet (2025 instructions, p. 37): Form 2555 lines 45 and 50 (line 45 is lines 36 + 42 − 44), less line 2b, "itemized deductions or exclusions you couldn't claim because they are related to excluded income". So the add-back is a new tax-unit variable, niit_magi_section_911_addition:

• Default: max(0, foreign_earned_income_exclusion). For a filer with no housing exclusion, no housing deduction and nothing on worksheet line 2b, it equals the section 1411(d) amount. Housing amounts make the default too high. Itemized deductions on line 2b make it too low, since section 1411(d)(2) subtracts only deductions "taken into account in computing adjusted gross income". The variable's documentation says so.
• Input: for any other filer, a user enters the Form 8960 worksheet line 2(c) amount directly. YAML cases cover a housing exclusion and line 2b itemized deductions. This follows estate_income_niit_magi_adjustment (a formula default that a user can override).

niit_magi adds max(0, niit_magi_section_911_addition): section 1411(d) adds "the excess (if any)", so a negative entered amount adds nothing. foreign_earned_income_exclusion's documentation now points to the new variable.

Invariants

They hold for every input and are tested in policyengine_us/tests/test_niit_section_911_magi.py (Hypothesis plus a fixed grid, 2018, 2022, 2025 and 2026, all five filing statuses):

1. Nothing added changes nothing, bit for bit. With no exclusion, or an entered addition of zero or less, AGI, niit_magi, net_investment_income_tax, income_tax_before_refundable_credits and income_tax are np.array_equal to the pre-change formula. That formula runs as a reform on the same households, so this is a differential test of old against new code. A household without an exclusion also gets identical results whatever the others in its batch exclude.
2. MAGI is AGI plus the positive code H change plus the addition, and the addition is the exclusion, or the entered amount, floored at zero.
3. Monotone: niit_magi and net_investment_income_tax never fall as the excluded amount rises (exact; float rounding is monotone).
4. Statute differential: the tax is 3.8% × min(max(0, NII), max(0, MAGI − threshold)), with the section 1411(b) thresholds ($250,000 joint or surviving spouse, $125,000 separate, $200,000 otherwise, not indexed) hard-coded independently of the parameter files. Also 0 ≤ tax ≤ 3.8% × max(0, NII).

YAML (niit_magi.yaml, net_investment_income_tax.yaml): the reproducer; a case below the threshold (MAGI 190,000, tax 0); a case where the MAGI excess binds (1,140); a joint return pushed over the $250,000 threshold only by the add-back (1,520); a separate return ($125,000 threshold, 760); an entered addition without housing amounts (760 rather than the default's 1,900); an entered addition that restores line 2b itemized deductions (MAGI 300,000 where the default gives 295,000); the add-back alongside code H; and a negative entered amount. On the pre-change formula, 9 of the 10 new YAML cases fail (all but the negative-amount case) and so does the property test.

Local runs (Python 3.14, dev extras installed, -rs shows nothing skipped): the two NIIT YAML files (32 cases) pass, and the new property test file, the dependent and estate NIIT invariant tests and test_section_911_tax_stacking.py pass (49 tests, about 2 minutes). make format is clean. CI runs the full YAML shards.

Microsimulation impact

I ran the default dataset (populace-us-2024-spm-20260909, checksum-verified by the loader) twice in one process, this branch against the pre-change niit_magi as a reform, for 2025 and 2026. The dataset has no foreign_earned_income_exclusion column: none of its 18 tax-unit or 318 person columns matches foreign, 911, housing, possession or puerto (read from the cached h5 tables). The run confirms the variable and the new addition are 0 for all 79,729 tax units. niit_magi, net_investment_income_tax, income_tax_before_refundable_credits and income_tax are bit-identical:

Year	Outputs compared	NIIT total	Change
2025	niit_magi, net_investment_income_tax, income_tax_before_refundable_credits, income_tax ($2,417.929bn)	$68.231bn	$0.00; 0 of 79,729 units changed
2026	niit_magi, net_investment_income_tax	$73.200bn	$0.00; 0 of 79,729 units changed

The 2026 income tax comparison wasn't run: the process was killed in an app restart after the 2026 MAGI and NIIT lines. The only baseline reader of niit_magi is net_investment_income_tax, so no 2026 income tax change is expected, but that figure is not measured.

Run output (script: microsim_impact.py, default dataset, this branch vs the pre-change formula as a reform)

2025 tax units 79729 weighted 164,209,856
2025 tax units with a section 911 exclusion: 0 addition nonzero: 0
2025 niit_magi fixed total $18,020.351562bn before total $18,020.351562bn weighted change $0.00 units changed 0 bit-identical True
2025 net_investment_income_tax fixed total $68.230759bn before total $68.230759bn weighted change $0.00 units changed 0 bit-identical True
2025 income_tax_before_refundable_credits fixed total $2,527.251953bn before total $2,527.251953bn weighted change $0.00 units changed 0 bit-identical True
2025 income_tax fixed total $2,417.929443bn before total $2,417.929443bn weighted change $0.00 units changed 0 bit-identical True
2026 tax units 79729 weighted 165,382,512
2026 tax units with a section 911 exclusion: 0 addition nonzero: 0
2026 niit_magi fixed total $18,990.990234bn before total $18,990.990234bn weighted change $0.00 units changed 0 bit-identical True
2026 net_investment_income_tax fixed total $73.199623bn before total $73.199623bn weighted change $0.00 units changed 0 bit-identical True

The fix changes household calculations that enter a section 911 exclusion, and microsimulation once the data carries one.

Overlap check

• Deny the refundable CTC and the EITC to section 911 filers (26 U.S.C. 24(d)(3), 32(c)(1)(C), 6012(c)) #9750 (refundable CTC and EITC for section 911 filers) touches no file here.
• Use the married-filing-separately breakpoint on Form 6251 Part III line 18 #9780 and Tax 28% rate and unrecaptured 1250 gain at regular rates below the 24% bracket top, and use Schedule D Tax Worksheet line 21 on Form 6251 line 27 #9792 edit test_section_911_tax_stacking.py, so the new properties are in their own file.
• The Biden FY2025 NIIT reform and Alabama's federal income tax deduction read niit_magi and NIIT, and pick up the fix.

Other federal MAGI definitions that add back section 911 amounts (CTC §24(b)(1), student loan interest §221(b)(2)(C), §86, §25A, §25B, §§224, 225 and 163(h)(4)) are missing it too. That work is a separate follow-up with a reproducer for each. The premium tax credit's aca_magi already adds the exclusion. The model has no §219(g) IRA phase-out to fix.

Axiom

us:statutes/26/1411#niit_modified_adjusted_gross_income encodes the add-back, but the §911(d)(6) rule it imports is versioned only from 2026-01-01. On the pinned engine (89571cc2), these tests' 2025 cases fail with "no formula version at 2025-01-01"; the same cases dated 2026 pass with the expected values. TheAxiomFoundation/rulespec-us#1495 is dispatch-ready, with the verbatim law, a review_finding and these companion cases. The signed re-encode needs approval for a billed run.

axiom: us/statute/26/1411/d TheAxiomFoundation/rulespec-us#1495 queued

🤖 Generated with Claude Code

[MaxGhenis]

Handing this PR to the US + core hub.

• Head: f33e44297d5c4835c9533581bc7518aacb1101ca
• Independent review (Opus 5.5, subfleet run --task review --tier standard):
  • Round 1 at 4466c20a: REQUEST_CHANGES. The blocking finding was that the default's documentation ignored worksheet line 2b; six other findings were non-blocking or nits. All seven were fixed in 43bb42c0 and a34c6d4b.
  • Round 2 at a34c6d4b: APPROVE, with two wording nits, fixed in f33e4429.
  • Round 3 at f33e4429: APPROVE.
  • Verdicts are in ~/reviews/pe-us-niit911/ (review_r1.md, review_r2b.md, review_r3.md).
• Passed locally (Python 3.14 with dev extras, nothing skipped): both NIIT YAML files (32 cases), plus 49 tests across the new property file, the dependent and estate NIIT invariant files, and test_section_911_tax_stacking.py. make format is clean. I confirmed the new tests fail on the pre-change formula. The microsim run is in the PR body.
• Pending: CI. The Pull request workflow for f33e4429 was queued at 19:16Z, behind 14 queued runs repo-wide. It is not mergeable yet: checks are still pending, and the required approving review can only come from a maintainer. Under Max's merge rules, admin-merge with --match-head-commit f33e4429 once every check passes.
• Depends on: nothing.
• Decisions: d975 is approval of the OpenAI-billed signed re-encode for Version IRC §911(d)(6) from 1982 so §1411 NIIT MAGI evaluates for tax years 2013–2025 TheAxiomFoundation/rulespec-us#1495. That is the Axiom side of the axiom: line. It outlives this merge and does not block it.
• Follow-ups: the other federal MAGI definitions that should add back section 911 amounts (§§24(b)(1), 221, 86, 25A, 25B, 224, 225 and 163(h)(4)) are a separate task with a reproducer for each.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (909176a) to head (f33e442).
⚠️ Report is 217 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main     #9808   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            4         3    -1     
  Lines           76        35   -41     
  Branches         2         0    -2     
=========================================
- Hits            76        35   -41     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

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]

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

• Head: f33e44297d5c4835c9533581bc7518aacb1101ca, handed off by its build session at 19:20Z on 10/5.
• Independent review (Opus 5.5, Subfleet):
  • r1 at 4466c20a requested changes. All seven findings were fixed in 43bb42c0 and a34c6d4b.
  • r2 at a34c6d4b: APPROVE.
  • r3 at f33e4429: APPROVE. The hub re-checked the a34c6d4b..f33e4429 diff itself: two files, comment and docstring wording only, no formula, input or expected-value change.
• CI: three quick checks failed after running about an hour during the 10/5 GitHub Actions runner incident. The hub reran the failed jobs, and every check passed at this head before merging.
• Impact: the microsimulation section in the body shows no change on the default dataset, which has no section 911 exclusion column. The fix changes household calculations that enter one.
• Decisions: d975 is the Axiom re-encode for rulespec-us#1495 and does not block this merge.