Apply the 6012(a)(1) federal filing requirement from 2026

[MaxGhenis]

#10135, which this PR was stacked on, has merged; this PR's diff is now its own changes against main.

What is wrong

tax_unit_is_required_to_file applies the 2018–2025 rule in every year: a return is due when gross income is more than standard_deduction. That rule is 26 U.S.C. 6012(f), and it covers only "a taxable year beginning after December 31, 2017, and before January 1, 2026". Pub. L. 119-21 does not amend section 6012; its sections 70102 and 70103 made the larger standard deduction (63(c)(7)) and the zero exemption amount (151(d)(5)) permanent. So from 2026 the rule is 6012(a)(1), with an exemption amount of zero:

(A) Every individual having for the taxable year gross income which equals or exceeds the exemption amount, except that a return shall not be required of an individual— (i) who is not married ... and for the taxable year has gross income of less than the sum of the exemption amount plus the basic standard deduction applicable to such an individual, ... (iv) who is entitled to make a joint return and whose gross income, when combined with the gross income of his spouse, is, for the taxable year, less than the sum of twice the exemption amount plus the basic standard deduction applicable to a joint return ...

(B) The amount specified in clause (i), (ii), or (iii) of subparagraph (A) shall be increased by the amount of 1 additional standard deduction ... in the case of an individual entitled to such deduction by reason of section 63(f)(1)(A) (relating to individuals age 65 or more), and the amount specified in clause (iv) ... for each additional standard deduction to which the individual or his spouse is entitled by reason of section 63(f)(1).

For a tax unit nobody can claim as a dependent, the model got 2026 wrong in two ways:

1. Blindness. The model's threshold includes the additional standard deduction for blindness (section 63(f)(2)). Section 6012(a)(1)(B) raises the threshold for age only. A blind single filer under 65 with $17,000 of wages in 2026 must file ($17,000 is not less than $16,100); the model said no ($17,000 is under $16,100 + $2,050).
2. The exact dollar. (A) excuses gross income "less than" the threshold, so gross income equal to it requires a return. The model used "more than".

Chart A of the 2017 Form 1040 instructions, the last year under (a)(1), reads the same way: it lists only age, and a return is due if gross income "was at least" the amount.

Fix

• New parameter gov.irs.income.filing_requirement.standard_deduction_rule.applies: false before 2018, true for 2018 to 2025, false from 2026. A reform can keep 6012(f) by setting it to true.
• 2018 to 2025 (switch on): no change. Gross income more than the standard deduction requires a return, as 6012(f) says.
• From 2026 (switch off), for a tax unit nobody can claim: a return is required when gross income is at least the exemption amount for each spouse (zero) plus basic_standard_deduction plus one additional standard deduction for each spouse aged 65 or older (aged_head, aged_spouse, and the existing aged_or_blind.amount for the filing status). The WFTCA contributed reform's bonus_guaranteed_deduction, which its parameter describes as "an increase to the basic standard deduction", counts too; it is zero unless that reform is on. A married individual filing separately must file with any gross income, as before.
• A tax unit whose head or spouse can be claimed keeps the standard deduction test in every year. From 2026 that test is 6012(a)(1)(C)(i)(II), and Chart B applies it. Whether the (A) threshold also binds these filers is decision d1395, so this PR does not change them.

No new inputs. Existing parameters supply every amount.

Readings and open points

• d1403 (new, queued for Max): 2018 to 2025 for filers nobody can claim. Section 6012(f)(1) excuses gross income that "does not exceed the standard deduction applicable to such individual ... under section 63", which includes the blindness addition. Chart A of the 2025 instructions lists only age and says "at least". The model follows the statute, as it did before; this PR does not change 2018 to 2025. YAML cases 20 and 22 sit on the two points where the readings differ and name d1403.
• d1395 (open): filers who can be claimed. Unchanged. YAML cases 28 and 29 pin the model's reading from 2026: a blind claimable filer keeps the blindness addition, and a joint return whose spouse can be claimed is not required to file at exactly the joint amount. Read literally, the last sentence of (A) removes clause (iv) when "any other taxpayer is entitled to an exemption for such spouse under section 151(c)". But under section 152(b)(2) a spouse who files a joint return is not a dependent, apart from the refund-only exception in Publication 501, which applies only when neither spouse must file. That point is noted for d1395 on Encode IRC 6012(a)(1) for taxable years after 2025 and compute the 63(c)(5) unearned income test in 6012(f)(2)(D) [pe-parity] TheAxiomFoundation/rulespec-us#1696.
• Zero gross income on a separate return (added to d1403). With a zero exemption amount, the opening clause of (A) ("gross income which equals or exceeds the exemption amount") is met by anyone, even with no gross income. The National Taxpayer Advocate read the 2018–2025 rule the same way, and says the IRS set $5 as relief. The model has always required a return only for positive gross income, and this PR keeps that (YAML Case 19). Three readings (literal: any; IRS: $5; model: more than $0) are queued with d1403.
• Same household. Clause (iv) applies only if the spouses share a household at the end of the year. The model has no such input and assumes they do, as before.
• Before 2018 the switch is also off, as the law requires. The model has no standard deduction amounts before 2018, so those years are not tested.

Tests

tax_unit_is_required_to_file.yaml: 29 cases. 2026: single, head of household and surviving spouse filers, under and over 65, blind and not, a dollar under and at each threshold; joint returns with no, one and two spouses aged 65, and both blind; separate returns at $0 and $1. 2025: the same thresholds under 6012(f), blindness included and the exact dollar excused. Two cases pin the claimable-filer reading (d1395). 9 of the 29 fail on a9e58977ff (#10135's head): every 2026 case whose gross income equals the 6012(a)(1) threshold (Cases 2, 3, 6, 7, 9, 11, 13, 15 and 17; three of them with a blind filer). Two cases in contrib/congress/wftca.yaml pin the WFTCA bonus in the 2026 threshold.

tests/core/test_filing_requirement_invariants.py (new). The reference rules are computed from the published amounts (Rev. Proc. 2025-32 sec. 4.14 for 2026, Chart A 2025 for 2025), not from the model's parameters.

• test_threshold_grid: one simulation over every filing status and every combination of age 65 and blindness for the head and spouse, at $0, $1, and a dollar under, at and over each rule's threshold. The 2026 result equals 6012(a)(1) and the 2025 result equals 6012(f), on every row.
• test_filing_requirement_properties: Hypothesis draws batches of tax units (any filing status, age, blindness and claimability; wages, interest, dividends, self-employment profit or loss; gross income often exactly at a threshold), each with a blind-flipped and an aged variant, plus three fixed anchor units in every example: a blind filer nobody can claim at the 2026 threshold, a claimable filer over the unearned income test, and a joint unit headed by a younger claimable adult (round 1's counterexample).

Invariants (each executed):

1. Differential: for a tax unit nobody can claim, the 2026 result equals 6012(a)(1) and the 2025 result equals 6012(f), computed from the published amounts (grid and random draws).
2. For a tax unit nobody can claim, blindness never changes the 2026 result. For every other tax unit and year, more blindness never adds a requirement.
3. Reaching 65 never adds a requirement, for any tax unit or year. Every test unit supplies its roles (tax_unit_role_input), so a variant cannot change who heads the unit.
4. Inclusion: in 2026 the model never excuses a tax unit nobody can claim that 6012(f), with the 2026 amounts, would require to file.
5. Claimable tax units keep one rule in 2025 and 2026: a return is required exactly when gross income is over the unit's standard deduction (any gross income on a separate return), or a claimable adult's interest and dividends are over $1,350 plus that adult's own additions for age and blindness.

tests/core/test_filing_requirement_dependent_invariants.py (from #10135). Its property 7, "being claimable never removes the requirement", now has one intended exception from 2026: gross income exactly equal to the threshold of a sighted tax unit nobody can claim. (a)(1)(A) requires that unit to file ("less than"), while the claimable-filer test is "in excess of" (d1395). The property skips exactly those units, and test_claimability_can_end_the_requirement_at_the_threshold pins the case. simulate() now passes keyword options to Simulation.

Mutation check (at 634f2ee275; the later commit changes only comments and the test units' roles; each mutant of the formula against the new YAML file, #10135's 48 cases, wftca.yaml and both Python files):

Mutant	New YAML (29)	#10135 YAML (48)	WFTCA YAML (4)	Python (7 tests)
#10135's formula (a9e58977ff): 6012(f) in every year	9 fail	pass	1 fail	3 fail
WFTCA bonus left out of the (a)(1) threshold	pass	pass	1 fail	pass
blindness counted in the (a)(1) threshold	3 fail	pass	pass	2 fail
> instead of >=	9 fail	pass	1 fail	3 fail
(a)(1) applied to claimable units too	2 fail	pass	pass	2 fail
switch ignored: (a)(1) in every year	3 fail	pass	pass	2 fail
spouse's age ignored	1 fail	pass	pass	1 fail
no positive gross income guard (separate return at $0)	1 fail	pass	pass	1 fail
separate return given the basic threshold	1 fail	pass	pass	2 fail
no age addition	5 fail	pass	pass	1 fail
claimable check on the head only, not the spouse	1 fail	pass	pass	pass

Every mutant fails at least one test. Hypothesis shrinking was off for this check, so the Python counts are a lower bound.

CI cost. The new Python file runs in the core group of the rest Python tests and uses only the default tax-benefit system. Locally (macOS, shared host at load 150-250, not Linux CI): 6.1 s inside pytest (0.55 s and 0.44 s of calls) and 32.6 s wall time including the model import, with a peak of 1.03 GB, the same as #10135's invariants file alone (1.04 GB). An earlier version checked the switch with a reformed tax-benefit system; that cost a second parameter tree (+0.68 GB peak) in a group that already peaks at 13.3 GB on the 16 GB runner (run 38210034933, rest-python-core-resources.txt), so properties 4 and 5 compare with references instead.

Also passing: tax_unit_is_required_to_file_dependents.yaml (48), tax_unit_is_filer.yaml (both), eitc_filer_gate.yaml, section_911_state_eitc.yaml, medicaid_person_is_required_to_file.yaml, basic_standard_deduction.yaml (124 cases together with the new file), and contrib/congress/wftca.yaml (4).

Hub microsimulation impact (current head)

A real microsimulation on the default dataset, main at the merge base f4539bc050 against this PR's head a9ed655465, 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)
tax_unit_is_required_to_file	+0	+43,205	0 (0)	33 (43,205)
tax_unit_is_filer	+0	+0	0 (0)	3 (0)
eitc	+0	+0	0 (0)	0 (0)
aca_ptc	+0	+0	0 (0)	0 (0)
is_medicaid_eligible	+0	+0	0 (0)	0 (0)
state_income_tax	+0	+0	0 (0)	0 (0)
income_tax	+0	+0	0 (0)	0 (0)
household_net_income	+0	+0	0 (0)	0 (0)
household_benefits	+0	+0	0 (0)	0 (0)

2025 is unchanged. In 2026, 33 records (43,205 weighted tax units) become required to file and 3 records (total weight 0.004) become filers; no other output changes. That matches the owner's 2026 run in "Impact (2026)" below, which compared against #10135's earlier head, so #10022's § 911 add-back in the merge base doesn't change the result on this dataset.

Comparison files are in fixes/10137-impact/ on the hub host.

Impact (2026)

A real microsimulation on the default dataset (populace_us_2024.h5@populace-us-2024-spm-20260909), one run per side under the shared heavy-job lock: #10135's earlier head 1a5dd0586a against this PR's earlier head 8ff66c7ea5, both on main 065d0e9560. Sums are weighted.

Output	Change	Records changed
tax_unit_is_required_to_file (tax units)	+43,205 (129.81M to 129.85M)	33, all from not required to required
tax_unit_is_filer (tax units)	+0.004	3, total weight 0.004
eitc	$0	0
income_tax	$0	0
state_income_tax	$0	0
aca_ptc	+$26	1, weight 0.002 (+$11,214 for that record)
household_net_income, household_benefits	$0	0
is_medicaid_eligible (people)	0	0

Of the 33 records that become required to file, 30 were already filers in the model. None of the 33 changes in EITC, income tax, state income tax, net income or benefits. 2025 is unchanged: the switch keeps 6012(f) there, and that path is the old formula.

The numbers hold at this PR's current head. Since the run, the branch was rebased onto #10135's a9e58977ff, whose round-1 change touches only the unearned income test, which applies only to claimable adults, and the WFTCA term was added, which is zero without that reform. A tax unit nobody can claim gets the same result at both heads on each side, and this PR does not change a claimable unit, so every record's difference is the same. Comparison files: ~/reviews/pe-us-6012a1-2026-10-11/impact/ on the hub host.

Independent review

Round 1 (GPT-6.1 Sol via Subfleet, at 634f2ee275): REQUEST_CHANGES, two minor findings. The reviewer recomputed the 2026 thresholds against Rev. Proc. 2025-32 and confirmed the switch dates, the age-only additions, the equality comparison, and that no path sends a claimable unit to the new test.

1. The age property could fail for a reason unrelated to age. With roles left to age order, raising both spouses to 65 can make a younger claimable spouse the head, which limits the joint standard deduction. Reproduced (2026: $32,200 to $4,650), then fixed: every test unit supplies its roles, and the counterexample is a fixed anchor. Dropping the roles makes the property fail on it.
2. The positive gross income guard departs from the literal text. Kept, because it is the model's existing behaviour and the IRS itself relieved the literal reading with $5. The formula now documents it with the Taxpayer Advocate source, and the reading is queued with d1403.

Round 2 (at 2a9771fcb9): APPROVE. Both findings resolved (the second as a documented deferral to d1403; the reviewer notes the Taxpayer Advocate post supports the literal and $5 readings, not the model's > $0, which this PR only preserves). One inherited nit, not from this PR: the 26/6012#c reference duplicates the section 6012 entry; left for a later reference cleanup so the reviewed head stays the head. Reports: ~/reviews/pe-us-6012a1-2026-10-11/review-r1.md and review-r2.md on the hub host.

Merge with main

Merged main f4539bc050400372ac2a4a8634b7df70c9a960bf without rebasing. The formula and variable-named YAML conflicted with #10022; main's newer #10135 also conflicted in the dependent YAML and invariant file. Preserved all 29 PR filing cases and all 8 main cases, all 49 dependent cases (including farm income), and both the separate-filer and 2026 equality exceptions in the claimability invariant. No original case names collide and no original expectations change.

This PR supplies the 2026 age-only threshold and equality comparison. #10022 supplies the gross section 911 add-back: Form 2555 line 43 when leaves apply, otherwise the nonnegative legacy exclusion; stacking deductions do not reduce it. The merged formula is:

G = head/spouse irs_gross_income + section_911_gross_exclusion
T = exemption * head_spouse_count + basic_standard_deduction
    + bonus_guaranteed_deduction + age_additions
2018–2025 or either adult claimable:
    income_test = G > (exemption if separate else standard_deduction + exemption)
otherwise (including 2026 nonclaimable filers):
    income_test = (G > 0) and G >= (exemption if separate else T)
required = income_test or existing claimable-adult unearned_income_test

METHODOLOGY_CHOICE: no (for this merge). 26 U.S.C. 6012(c) applies “For purposes of this section” and computes gross income “without regard to the exclusion provided for in section 911”, so its add-back also applies to the (a)(1) threshold after (f)'s 2018–2025 period. The previously documented d1395/d1403 choices remain.

Added one combined case: a blind single filer aged 40 in 2026 has $10,000 wages and a $6,100 gross housing exclusion. Allocable deductions of $6,100 leave line 2c at max(0, $6,100 - $6,100) = $0, but filing gross income is $10,000 + $6,100 = $16,100. That equals the $16,100 + $0 age threshold, requiring a return; blindness's $2,050 increases the standard deduction to $18,150 but not this threshold. All 37 old filing cases retain their expectations.

All requested files passed, one file per process with OMP/OpenBLAS/MKL/vecLib/NumExpr capped at four threads and OMP_WAIT_POLICY=PASSIVE: tax_unit_is_required_to_file.yaml 38; tax_unit_is_required_to_file_dependents.yaml 49; cdcc_filing_status_eligible.yaml 6; clean_vehicle_credit_prior_year_filing_status.yaml 2; wftca.yaml 4; test_filing_requirement_dependent_invariants.py 5; test_filing_requirement_invariants.py 2. Total: 106 passing cases/tests. make format (format and lint) and whitespace checks also passed.

The conflict resolution adds no behavior beyond this PR's stated rule plus #10022's gross-exclusion rule. No additional merge-specific methodology requires an impact rerun; the existing impact section still describes its stated older baseline rather than combined current-main totals.

axiom: TheAxiomFoundation/rulespec-us#1696 queued (extended in this comment with companion cases C1–C16 for filers nobody can claim, from the same sources as the YAML cases here)

🤖 Generated with Claude Code

[MaxGhenis]

Hand-off to the US + core hub

Head: 2a9771fcb9d83a12b1f4ea94bc28d1cf84bba1ce. Stacked on #10135 at a9e58977ff: merge #10135 first, and rebase this branch if #10135's head moves again (this PR's own commits are the three after a9e58977ff).

Independent review (GPT-6.1 Sol via Subfleet): round 1 REQUEST_CHANGES at 634f2ee275 (two minor findings: one test-design fix, one documented deferral to d1403); round 2 APPROVE at 2a9771fcb9, the current head. Reports: ~/reviews/pe-us-6012a1-2026-10-11/review-r1.md and review-r2.md.

Passed locally at the head:

• The new tax_unit_is_required_to_file.yaml (29), Apply the unearned income filing test only to filers who can be claimed as dependents #10135's dependent cases (48), and the filer, EITC gate, section 911, Medicaid filing and basic standard deduction files: 124 cases.
• contrib/congress/wftca.yaml: 4 cases.
• Both property-test files: 7 tests.
• The 23 other YAML files with 2026+ blind cases (369 cases), and the 5 files with 2026+ inputs equal to a new threshold (61 cases). These include two partner files (partners/analytics_coverage/.../ssi/federal.yaml and .../school_meals/federal.yaml), both unchanged and passing, so no partner gate is needed.
• Mutation check: all 11 mutants fail at least one test (table in the body; artifacts in ~/reviews/pe-us-6012a1-2026-10-11/mutation/).

Pending: CI. Run 38238135593 at 2a9771fcb9 was queued behind the org's Actions queue at hand-off.

Impact: in the body. A full 2026 microsim pair: +43,205 weighted tax units required to file (33 records); no change to EITC, income tax, state tax, net income or benefits. Files: ~/reviews/pe-us-6012a1-2026-10-11/impact/. The body explains why the numbers hold at the current head.

Decisions (none blocks the merge; both defaults are no change):

• d1403 (new): the 2018–2025 statute vs Chart A, plus the $0 separate-filer reading.
• d1395 (Apply the unearned income filing test only to filers who can be claimed as dependents #10135's): claimable filers, including the 2026 clause (iv) carve-out note added from this PR.

Axiom: TheAxiomFoundation/rulespec-us#1696 queued, extended with companion cases C1–C16.

Follow-ups owed, not blocking:

• The 26/6012#c reference duplicates the section 6012 entry. It is inherited from main; fix it in the next reference cleanup.
• The clause (iv) same-household condition is not modeled. This is an existing limitation; cohabitating_spouses defaults to false and is not a year-end test.

🤖 Generated with Claude Code

[MaxGhenis]

US + core hub merge audit, us#10137 at a9ed655465cd43553c8e1e00a832288404bf47d0 (squash): Apply the 6012(a)(1) federal filing requirement from 2026.

• Review: GPT-6.1 Sol delta r1 APPROVE at a9ed655 (reviews-gen/us-10137/review-r1.md): main merge with Count the gross section 911 exclusion toward the filing threshold #10022 verified, 6012(c) add-back on both thresholds, all cases kept; owner's earlier independent review approved 2a9771f.
• Gates checked live at merge:
  • gh pr checks exits 0;
  • the latest Pull request run at the head concluded success;
  • MERGEABLE;
  • not a draft;
  • no reviewer's latest review requests changes;
  • the semantic-overlap guard;
  • head pinned.
• Impact (hub full-file microsimulation, merge base f4539bc050 against this head, in the description): 2025: no record changes in the measured outputs | 2026: tax_unit_is_required_to_file +43,205 (33 records); tax_unit_is_filer +0 (3 records).
• Axiom: Encode IRC 6012(a)(1) for taxable years after 2025 and compute the 63(c)(5) unearned income test in 6012(f)(2)(D) [pe-parity] TheAxiomFoundation/rulespec-us#1696 queued (extended in this comment with companion cases C1–C16 for filers nobody can claim, from the same sources as the YAML cases here)