[PavelMakarchuk]

Summary

The child and dependent care credit (CDCC) limits creditable expenses to the taxpayer's earned income, or to the lower earner's on a joint return. head_earned and spouse_earned read adjusted_earnings, which started from earned_income. That variable adds employment_income, the wages before pre-tax payroll deductions. A spouse who defers part of the pay into a traditional 401(k) or 403(b) therefore had a limit that was too high by the deferral.

Repro (TY2025, TX, from the issue): married couple, head $56,000 of wages, spouse $4,000 of wages with a $1,000 traditional 401(k) deferral, two children, $9,103 of childcare.

Variable	main	this PR	Form 2441
irs_employment_income (spouse)	3,000	3,000	3,000
adjusted_earnings (spouse)	4,000	3,000	3,000
cdcc_relevant_expenses	4,000	3,000	3,000
cdcc (20% rate, AGI $59,000)	800	600	600

adjusted_earnings now subtracts the wages that are not includible in gross income (employment_income − irs_employment_income), the same way filer_adjusted_earnings has done since #10030. earned_income does not change: about 18 formulas read it directly, and some benefit programs count wages before these deductions (the scope question that closed #4066). #10030 already moved the EITC and ACTC to taxable wages.

The law

• 26 U.S.C. 21(d)(1): expenses may not exceed the taxpayer's earned income or, for a married taxpayer, "the lesser of" the two spouses' earned income.
• Treas. Reg. 1.21-2(b)(3): "the term earned income has the same meaning as in section 32(c)(2) and the regulations thereunder."
• 26 U.S.C. 32(c)(2)(A)(i): employee compensation counts "only if such amounts are includible in gross income for the taxable year."
• 26 U.S.C. 402(e)(3): an elective deferral is not "treated as distributed or made available to the employee", so it is not in gross income. Cafeteria-plan premiums and payroll HSA contributions are also out of W-2 box 1. Designated Roth deferrals stay in (26 U.S.C. 402A(a)(1)).
• Form 2441 instructions (2025), lines 4 and 5: earned income starts from "The amount shown on Form 1040, 1040-SR, or 1040-NR, line 1z". Line 1a is W-2 box 1.
• 26 U.S.C. 129(b) limits the dependent care assistance exclusion to the same earned income, and 129(e)(2) gives it "the meaning given such term in section 32(c)(2)". Form 2441 lines 18 and 19 figure it "in the same way as earned income for figuring the credit".

What changes, and for whom

• Filers in which a head or spouse has traditional 401(k)/403(b) deferrals, pre-tax health premiums or payroll HSA contributions, and in which that person's earned income is the binding limit.
• In California, also:
  • Every child AMT filer (Schedule P line 7 now uses California's own minimum), in every year.
  • Filers with employer dependent care benefits. The California credit now goes through FTB 3506 Part IV in every year, so the benefits reduce the base through federal Form 2441 line 31, which also subtracts them from the expenses (line 30). Main used the federal credit outside 2021, which skips line 30.
  • Filers with a business loss, which now offsets California wages before the zero floor.
  • Years before 2021, which now use the California computation as well.
• New York 2026 results do not change from main.

Formula	Rule	Check
head_earned, spouse_earned, min_head_spouse_earned → cdcc_relevant_expenses	IRC 21(d)(1)	Reg. 1.21-2(b)(3) → 32(c)(2). Correct.
dependent_care_assistance_exclusion (also feeds cdcc_limit under 21(c))	IRC 129(b)	129(e)(2) → 32(c)(2). Correct.
amt_exemption, child cap before 2018	IRC 59(j)	See below. Taxable wages.
filer_adjusted_earnings	IRC 32(c)(2)	Now sums adjusted_earnings over non-dependents. Same values; the formula is no longer duplicated.
va_child_dependent_care_expense_deduction	Va. Code 58.1-322.03(3): expenses "upon which the federal credit is based under § 21"	Follows 21(d). Correct.
ma_dependent_care_credit (to 2022)	St. 2021, c. 24, § 29 (former c. 62, § 6(x)): expenses allowed for the section 21 credit	Follows 21(d). Correct.
mn_cdcc	Minn. Stat. 290.067, subd. 1a(d): "the provisions of section 21(d) of the Internal Revenue Code apply"	Correct.
az_dependent_care_expense_subtraction (from 2026)	A.R.S. 43-1022(34): section 21 expenses that exceed the federal section 21 credit	Follows 21(d). Correct.
md_dependent_care_subtraction	Md. Code, Tax-Gen. 10-208(e): expenses up to the § 21(c) dollar limit, without the § 21(a)(2) percentage	Reads cdcc_relevant_expenses, so it keeps the federal earned income limit, as on main.
sc_cdcc_potential	S.C. Code 12-6-3380: "computed as provided in Internal Revenue Code Section 21", at 7%	Follows 21(d). Correct.
co_low_income_cdcc	C.R.S. 39-22-119.5(4)(a): "the individual's earned income" (undefined; 39-22-103(11) gives it the IRC meaning)	Correct.
ca_cdcc_relevant_expenses, ca_dependent_care_assistance_exclusion	R&TC 17052.6(d)(2): earned income "limited to earned income subject to tax under this part"; FTB 3506 line 4: "other taxable employee compensation"	California wages. See below.
ca_amt_exemption, child	R&TC 17062 adopts IRC 55-59; Schedule P (540) line 8: "wages, tips, and other payments received"	Same reasoning as the federal cap, on California wages. Line 7 uses California's own minimum ($9,750 for 2025, not the federal $9,550). See below.
ar_federal_cdcc	A.C.A. 26-51-502 adopts IRC 21; AR2441 line 4 starts from AR1000F line 8 wages	Correct.
ny_cdcc_max (606(c), to 2025)	IT-216 (2025): "other taxable employee compensation", IT-201 line 1, federal amount	Correct.
ny_cdcc_qualifying_expenses (606(c-2), from 2026)	Tax Law 606(c-2)(2)(C)	Kept on gross wages. See below.
id_household_and_dependent_care_expense_deduction	Idaho Code 63-3022D: expenses "as further specified and limited by section 21 (d) and (e)"	Correct.
or_cdcc_relevant_expenses	ORS 315.264(1)(c)(A): "earned income taxable by Oregon"	Correct.
wi_childcare_expense_subtraction, wi_childcare_expense_credit_potential	Wis. Stat. 71.05(6)(b)43. and 71.07(9g): section 21 expenses; Wisconsin replaces only the 21(c) dollar limit	Follows 21(d). Correct.

States that take a percentage of the federal credit inherit the change through cdcc. The individual_eitc (Winship) reform also reads adjusted_earnings; it now uses the same taxable wages as the baseline EITC. cdcc_single_parent_work_requirement and the Minnesota HF 1938 reform read the CDCC earnings and follow the federal rule. The single-parent reform overrides federal cdcc_relevant_expenses, so it no longer reaches the California credit, which now uses ca_cdcc_relevant_expenses in every year (on main it reached California outside 2021).

AMT child cap (before 2018). #10030 kept this cap on gross earnings. IRC 59(j)(1)(A) uses earned income "as defined in section 911(d)(2)", which has no gross-income inclusion clause. But 911(d)(2)(A) counts "amounts received as compensation for personal services", and 402(e)(3) says an elective deferral is not made available to the employee. The 2017 Form 8615 instructions apply the same 911(d)(2) definition (through IRC 1(g)(4)) to the same children, and say earned income "is generally the total of the amounts reported on Form 1040, lines 7, 12, and 18". Line 7 is taxable wages. The 2017 Form 6251 instructions (worksheet line 8) use the same "personal services" sentence, but only the Form 8615 instructions point to Form 1040 line 7. So the cap now follows adjusted_earnings. It applies only before 2018 (IRC 55(d)(4)(A)(iii)), and California's Schedule P uses the same wording. If reviewers prefer to keep #10030's reading, amt_exemption and ca_amt_exemption can add back the excluded wages the way New York does below.

California. California excludes traditional 401(k)/403(b) deferrals and pre-tax health premiums, but it taxes payroll HSA contributions: the Schedule CA (540) instructions, line 1h, add the W-2 box 12 code W amount in column C (also FTB Pub. 1005; ca_hsa_addition). #10030 added ca_taxable_wages (W-2 box 16) for this. With the federal change alone, California's CDCC limit would have dropped the HSA amounts. This PR therefore adds:

• ca_adjusted_earnings (Person): adjusted_earnings measured on California wages (ca_taxable_wages), which keep the payroll HSA contribution. ca_taxable_wages subtracts a subset of the federal pre-tax list, so it is never below irs_employment_income. A business loss offsets these wages before the zero floor (earnings_before_zero_floor, now shared with adjusted_earnings). ca_head_earned and ca_spouse_earned sum it for the head or the spouse.
• ca_cdcc_min_head_spouse_earned: the lesser of the two, with the same deemed income for a student or disabled spouse (FTB 3506 line 5: $250 or $500 a month). The deeming logic of min_head_spouse_earned moved into a helper, lesser_head_spouse_earned, that both variables call. The federal results do not change.
• ca_cdcc_relevant_expenses caps expenses at ca_cdcc_min_head_spouse_earned. ca_dependent_care_assistance_exclusion raises the federal exclusion by the amount that California earned income exceeds federal earned income, up to the benefits. That equals the lesser of the benefits, the California cap and California earned income, because the federal cap is at least California's. ca_amt_exemption (Schedule P worksheet line 8) uses ca_head_earned.
• The final credit: child_dependent_care/input.yaml now selects ca_federal_capped_cdcc (FTB 3506 lines 3-10) in every year from 2012, instead of the federal cdcc outside 2021, so California's own expense, employer benefit and earned income limits reach ca_cdcc. R&TC 17052.6(a)(1) has applied this computation since 2000, and the model runs California tax for years before 2021 on backfilled parameters. ca_cdcc_rate is the federal cdcc_rate outside 2021: 17052.6(a)(1) uses IRC 21 "as applicable for federal income tax purposes for the taxable year", which overrides the R&TC 17024.5 specified date, so 2026 takes the P.L. 119-21 schedule. The cap at federal tax before credits less the foreign tax credit is the same as federal cdcc_credit_limit.
• ca_amt_exemption worksheet line 7 reads a new California parameter, amt.exemption.child.amount (Schedule P (540) instructions: 7,850, 8,300, 8,950, 9,450 and 9,750 for 2021-2025), instead of the federal IRC 59(j) amount.

No other California formula reads adjusted_earnings, head_earned, spouse_earned or min_head_spouse_earned. CalEITC and the YCTC already use ca_taxable_wages.

New York from 2026. Tax Law 606(c-2)(2)(C) (L. 2026, ch. 59, Part A, § 2) defines earned income as "the wages, salaries, tips and other employee compensation, and those items of gross income which are includible in the computation of net earnings from self-employment". It leaves out IRC 32(c)(2)(A)(i)'s "only if such amounts are includible in gross income". ny_cdcc_qualifying_expenses therefore measures each person's earned income on gross wages with earnings_before_zero_floor (employment_income as the wage base), so a business loss offsets the wages before the zero floor, and sums it for the head and the spouse. This equals adjusted_earnings on main, so the results stay as on main. DTF has not published the 2026 IT-216 instructions. Revisit this if they start the line from federal wages.

Known gaps, not changed here (the same on main; follow-ups for separate PRs):

• ca_federal_capped_cdcc and federal cdcc cap the base at federal tax before credits, but R&TC 17052.6(a)(2) applies California's percentage "without taking into account whether there is a federal tax liability".
• ca_cdcc calls p.rate.calc(agi) without right=True, so AGI of exactly $40,000, $70,000 or $100,000 drops a band.
• Federal cdcc_relevant_expenses skips Form 2441 line 30, and the federal exclusion is not limited to qualified expenses (lines 16 and 17).
• mn_cdcc does not reduce its dollar limit by the section 129 exclusion (290.067, subd. 1a(c)).
• hi_cdcc_min_head_spouse_earned reads gross earned_income directly and does not use this chain.
• or_cdcc_relevant_expenses uses the federal deemed income for a student or disabled spouse. ORS 315.264(5)(a) deems $1,000 a month ($2,000 for two or more qualifying individuals) for students only.
• ar_federal_cdcc does not leave out active-duty military pay, which Arkansas exempts (AR2441 line 4 starts from AR1000F line 8).

Tests

New cases, run against main (bf2ec882ba) and this branch:

Case	Variable	main	this PR
$4,000 wages, $1,000 traditional 401(k)	adjusted_earnings	4,000	3,000
$10,000 wages, $1,000 Roth 401(k), $500 pre-tax premiums, $1,000 payroll HSA	adjusted_earnings	10,000	8,500
$4,000 wages, $1,000 deferral, $10,000 self-employment income	adjusted_earnings	13,293.52	12,293.52
$2,000 wages, $1,000 deferral, −$3,000 self-employment (control)	adjusted_earnings	0	0
Issue repro (TY2025)	cdcc	800	600
Joint, $5,000 employer benefits, spouse $4,000 wages with $1,500 deferral	dependent_care_assistance_exclusion	4,000	2,500
Virginia, joint, spouse $4,000 wages with $1,500 deferral, $9,000 care	va_child_dependent_care_expense_deduction	4,000	2,500
New York 2026, same household	ny_cdcc_qualifying_expenses	4,000	4,000
California, joint, spouse $6,000 wages with $1,000 401(k) and $1,000 payroll HSA, $9,000 care	ca_cdcc_relevant_expenses	6,000	5,000
California, joint, $5,000 employer benefits, spouse $4,000 wages with $1,000 payroll HSA	ca_dependent_care_assistance_exclusion	4,000	4,000
California child filer, $10,000 wages, $1,000 401(k)	ca_amt_exemption	19,550	18,750
California child filer (2022), $5,000 wages	ca_amt_exemption	13,200	13,300
California, joint (2025), head $90,000, spouse $4,000 wages with $1,000 payroll HSA, two children, $9,000 care	ca_cdcc / ca_income_tax	272 / 503.38	272 / 503.38
Same household (2021), spouse $4,000 self-employment loss	ca_cdcc / ca_income_tax	0 / 1,299.18	0 / 1,299.18
Same household (2025), spouse $3,500 self-employment loss	ca_spouse_earned / ca_cdcc	500 / 34	500 / 34
California, $5,000 employer benefits, spouse $4,000 loss	ca_dependent_care_assistance_exclusion	0	0
California (2025), $2,000 employer benefits, $2,500 care, one child	ca_cdcc	86	43
California (2026), head of household, AGI $80,000	ca_cdcc_rate	0.35	0.32
New York 2026, joint, spouse $4,000 wages with $1,000 deferral and a $4,000 business loss	ny_cdcc_qualifying_expenses	0	0
Same, $3,500 business loss	ny_cdcc_qualifying_expenses	500	500

Each row was run against main (bf2ec882ba). Bold values differ from main, and each expected value has its arithmetic in a comment. The rows without bold values, including the self-employment loss control, match main because main counted gross wages. For the same reason the California exclusion of 4,000 matches main; that case fails on main at the federal exclusion (4,000, expected 3,000). In hua7450's reviews, the first California revision of this PR (0623f5606f) gave 204 / 571.38, 68 / 1,231.18, 1,000 / 0 and 1,000 for the four California household rows, and e3bb2ad416 gave 1,000 for both New York loss rows. The 2026 ca_cdcc_rate of 0.35 on main did not reach the credit, because main used the federal cdcc from 2022.

ca_cdcc_relevant_expenses.yaml set min_head_spouse_earned as an input in 7 places; those inputs are now ca_cdcc_min_head_spouse_earned, the variable the formula reads. Its section 129 case now also sets min_head_spouse_earned and checks the California exclusion (5,000). For the same reason, the 7 head_earned inputs in ca_amt_exemption.yaml are now ca_head_earned, and New York cases 1-6 set employment_income instead of head_earned. The 2021 and 2022 ca_cdcc cases no longer set cdcc, which feeds nothing.

Changed: eitc_earned_income.yaml case "AMT kiddie-tax exemption retains its pre-2018 gross earnings cap" (from #10030) is now "... caps at taxable earnings before 2018". It expects 15,500 = (10,000 − 2,000 deferral) + 7,500 instead of 17,500. Its 2018 companion case only loses "gross" from its name.

tests/core/test_eitc_elective_deferral_invariants.py asserted that adjusted_earnings and amt_exemption do not depend on payroll deductions. It now asserts that head_earned, spouse_earned and min_head_spouse_earned match a run with the same W-2 box 1 wages and no deductions, and keeps the check that gross earned_income does not depend on them. It no longer checks amt_exemption: at 2026 the child cap is off, so that assertion could not fail. The eitc_earned_income.yaml case above covers the pre-2018 cap.

Local results (PolicyEngine-Core test runner and pytest, Python 3.11):

• YAML, 3,177 cases in gov/irs/credits/cdcc, gov/irs/income, gov/irs/tax/federal_income/alternative_minimum_tax, gov/irs/credits/earned_income, gov/irs/credits/ctc, the VA, MA, MN, CO, CA, AR, NY, NYC, ID, OR, WI and HI credit folders, ca/tax/income/alternative_minimum_tax, reform/winship.yaml, reform/cdcc_single_parent_work_requirement.yaml, contrib/states/ny/a06774, contrib/states/mn/walz and contrib/taxsim: 3,176 passed. The one failure was the Exclude pretax payroll deductions from EITC and ACTC earnings #10030 AMT case above; after the update, eitc_earned_income.yaml and the AMT folder pass (135).
• tests/policy/baseline/partners: 623 passed.
• pytest, 25 files (the elective deferral, dependent care exclusion, claimable-filer and head/spouse swap invariants, federal credit limit order, KY CDCC federal match, NYC credit properties, reference URL and source-conflict code health): 478 passed, 1 failed. The failure, test_source_conflicts.py::test_unscannable_files_fail_closed, also fails on main in this environment.
• After the rebase on 4a1a3cdeb7: the CDCC, AMT, EITC earned income, NY credit and VA deduction folders (612) and the two invariant files (7) pass.
• After hua7450's review fixes (ec4438cb78): gov/states/ca/tax 295, gov/irs/credits/cdcc with adjusted_earnings.yaml and eitc_earned_income.yaml 155, tests/policy/baseline/partners 623, other CA-reading tests and the federal AMT folder 161, and the invariant and code-health files 260, all passed.
• After the California change: gov/states/ca/tax and gov/irs/credits/cdcc, 424 passed; tests/policy/baseline/partners, 623 passed; the two invariant files and the reference URL and source-conflict code-health files, 235 passed and the same 1 environment failure.
• After hua7450's second review (9959a00e7e, Python 3.14): gov/states/ca/tax 312, gov/states/ny/tax 441, gov/irs/credits 732, tests/policy/baseline/partners 623 (no partner file changed), adjusted_earnings.yaml with the federal AMT folder, reform/cdcc_single_parent_work_requirement.yaml, contrib/states/ny/a06774 and contrib/states/ca/prop3 168, and the invariant file with seven code-health files 260, all passed.
• ruff format --check and ruff check pass on the changed Python files.

Note: #10501 merged first; e3bb2ad416 merges main and keeps both changes in ca_amt_exemption. #10308 touches the federal AMT child exemption tests, but no file in this PR.

Fixes #4007

axiom: n/a: us/statutes/26/21.yaml in rulespec-us takes the taxpayer's and spouse's earned income as inputs (taxpayer_earned_income_for_cdcc, spouse_earned_income_for_cdcc); this PR changes only how PolicyEngine measures those inputs (Treas. Reg. 1.21-2(b)(3), IRC 32(c)(2)), not the section 21 rule

🤖 Generated with Claude Code

[PavelMakarchuk]

@hua7450 — could you review this one when you have a moment? CI is green. It fixes #4007 for the child and dependent care credit chain only (Treas. Reg. 1.21-2(b)(3) and IRC 32(c)(2): earned income is pay "includible in gross income"), and leaves earned_income itself unchanged. Two points need your judgment: the pre-2018 AMT child cap now uses taxable earnings (#10030 left that question open; the PR explains the § 911(d)(2) reading and how to keep the gross cap instead), and California keeps payroll HSA contributions in its CDCC earned income (R&TC 17052.6(d)(2)), which also corrects ca_dependent_care_assistance_exclusion and ca_amt_exemption.

🤖 Posted by Claude Code

[hua7450]

Requesting changes for two confirmed California calculation defects. The main federal taxable-wage correction and pre-2018 AMT wage treatment are supported by the reviewed regulation and IRS instructions.

Reviewed commit: 0623f5606f61ffcce2b38cc8c393e99a0ce8a4cc.

Review status: PARTIAL because the New York 2026 interpretation remains unresolved. The two California findings below are confirmed.

Must fix

C1 — Route the final California CDCC through the California earnings calculation (OPEN)

CRITICAL (Must Fix), confirmed. Location: policyengine_us/variables/gov/states/ca/tax/income/credits/ca_cdcc_relevant_expenses.py:54-56 and its final consumer, ca_cdcc.py:18-21; the unchanged parameters/gov/states/ca/tax/income/credits/child_dependent_care/input.yaml selects federal cdcc again from 2022. Consequently the newly corrected California expense calculation never reaches the final credit in TY2025. A joint California household with head wages $90,000, spouse wages $4,000 including a $1,000 payroll HSA contribution, two qualifying children aged 5 and 3, and $9,000 of qualifying care has $4,000 of California spouse earnings. The PR calculates that correctly, and ca_federal_capped_cdcc is $800, but final ca_cdcc uses the federal $3,000 earnings cap and returns $204 rather than $272. California income tax is $571.38 rather than $503.38, a $68 overcharge; captured merge-base formulas return $272 and $503.38. The legal calculation is min($9,000, $6,000, $90,000, $4,000) × 20% × 34% = $272 under FTB Form 3506, 2025, Side 1, with California taxable compensation under its line 4 instructions and the HSA addition under Schedule CA line 1h. This diff introduces the lower federal cap, so the existing final-consumer routing now defeats the intended California protection. Wire the California computation into the final state credit for the affected years, with a final ca_cdcc/tax assertion.

C2 — Add California HSA wages before flooring the person's net earnings (OPEN)

CRITICAL (Must Fix), confirmed. Location: policyengine_us/variables/gov/states/ca/tax/income/ca_spouse_earned.py:35-36, mirrored in ca_head_earned.py:35-36. spouse_earned has already floored federal earnings at zero; adding California HSA wages afterward revives earnings that a business loss should offset. For TY2021, take the same joint household and qualifying care as C1, but give the spouse a $4,000 self-employment loss. California wages are $4,000 and net California earned income is $0 = max($4,000 − $4,000, 0). The new formula instead calculates max($3,000 − $4,000, 0) + $1,000 = $1,000. In 2021 the final state credit does consume this state computation: the PR awards ca_cdcc = $68 rather than $0 and reduces California income tax to $1,231.18 rather than $1,299.18. Captured merge-base formulas return the correct zero credit. IRC 21(d)'s earned-income limit, incorporated by California R&TC 17052.6(d)(2), requires net earned income; the Treasury definition and the FTB line 4 computation count wages together with net self-employment earnings. Restore the person's California wages before applying the zero floor, then aggregate the appropriate head or spouse; test a loss that offsets all California wages. The new addition can also incorrectly raise the reported state dependent-care exclusion. The 2021 example uses the existing 2021 state-credit path and the unchanged earned-income condition: the cached regulation dates from 2007 and the 2021 credit amendments did not replace section 21(d). The cached FTB Pub. 1005 HSA discussion identifies California's nonconformity with federal HSA legislation effective in 2004. No claim is made that a separate 2021 manual was downloaded or inspected.

Should address

A1 — New California AMT test codifies a federal amount instead of the published state amount (OPEN)

SHOULD ADDRESS; OPEN. At policyengine_us/tests/policy/baseline/gov/states/ca/tax/income/alternative_minimum_tax/ca_amt_exemption.yaml:100-102, the new TY2025 case states that California's child minimum is $9,550 and expects $18,550 after stipulated California earnings of $9,000. The official 2025 Schedule P (540) instructions, Part II line22, Exemption Worksheet lines7–10 specify a $9,750 child minimum, so conditional on those stipulated earnings, the worksheet gives min(92,749, 9,750 + 9,000) = 18,750. The formula's use of the federal child parameter already existed; this PR adds an incorrect policy oracle that locks it in. Correct the expected amount/source treatment rather than describing this as an introduced $200 tax error. The new test also stipulates dependent-child and payroll-HSA inputs; lawful HSA eligibility for that combination was not established in this review, so the example is not offered as a demonstrated lawful household's tax liability.

New York 2026 evidence gap

Tax Law 606(c-2)(2)(C)-(D) omits the federal taxable-inclusion clause, while section 607(a) supplies comparable federal meanings. The available official evidence did not resolve how the new definition treats payroll deferrals and business losses. A final-output change was reproduced, but it is not counted as a confirmed legal defect. This interpretation needs resolution before claiming complete legal alignment.

Validation

380 targeted tests passed, covering the changed YAML cases and relevant federal, California, Virginia and New York calculations plus two Python invariant files. Ten diagnostic calculations (five current/control pairs) completed with zero setup/calculation errors. Controls used captured merge-base formulas in memory; no source files or partner tests were edited. Source review covered 24 sources, including 7 PDFs and 2 form pages visually inspected at 300 DPI.

Please address C1 and C2 and add final ca_cdcc/California income-tax regression assertions for both demonstrated cases.

[PavelMakarchuk]

@hua7450, thank you for the review of 0623f5606f. New head: ec4438cb78 (fixes), then e3bb2ad416 (merge of main). C1, C2 and A1 are fixed; the New York 2026 note is answered at the end, with no code change.

C1: the final California credit now uses the California computation

• policyengine_us/parameters/gov/states/ca/tax/income/credits/child_dependent_care/input.yaml:5-8: from 2021, ca_cdcc reads ca_federal_capped_cdcc (FTB 3506 Part III, lines 3-10) in every year. The 2022 entry that switched back to federal cdcc is gone. California income tax starts in 2021 in the model (rate brackets from 2021-01-01), so this covers every year the PR affects.
• policyengine_us/variables/gov/states/ca/tax/income/credits/ca_cdcc_rate.py:18-22: outside 2021 the line 7 decimal is the federal cdcc_rate (the 2025 FTB 3506 line 7 chart is the federal 35%-20% table). For 2022-2025 the old and new formulas give the same rate (the 2022 cases in ca_cdcc_rate.yaml pass unchanged). The old formula had no 2026 second phase-out, so without this change the 2026 rate would have moved.
• The cap in ca_federal_capped_cdcc (income tax before credits less the foreign tax credit) is the same as federal cdcc_credit_limit, whose preceding_credits is [foreign_tax_credit]. So the routing changes only the expense and earned income limits. It also lets the existing FTB 3506 Part IV logic (California's $5,000 exclusion cap and the line 32 Form 2441 amount) reach the credit after 2021.
• Tests, ca_cdcc.yaml:
  • The four 2022 cases now give the amount through ca_federal_capped_cdcc (they set cdcc before).
  • New case at ca_cdcc.yaml:87-123, your TY2025 household: cdcc 600, ca_federal_capped_cdcc 800, ca_cdcc 272 and ca_income_tax 503.38. The arithmetic is in the comments: line 6 min(9,000, 6,000, 90,000, 4,000) = 4,000; x .20 x .34 = 272. California tax: 94,000 CA AGI - 11,412 = 82,588; 221.58 + 607.40 + 1,202.40 = 2,031.38; less 1,256 exemption credits and 272 = 503.38.

C2: California wages are restored before the zero floor

• policyengine_us/variables/gov/irs/income/adjusted_earnings.py:4-22: the pre-floor computation moved into a helper, earnings_before_zero_floor(person, period, parameters, taxable_wages). adjusted_earnings (lines 56-60) passes irs_employment_income, so its values do not change.
• New policyengine_us/variables/gov/states/ca/tax/income/ca_adjusted_earnings.py:34-44 (Person): the same helper with California wages, max(ca_taxable_wages, irs_employment_income), then the zero floor. A business loss now offsets the California wages, including the payroll HSA contribution, before the floor.
• ca_head_earned.py:21-25 and ca_spouse_earned.py:21-25 now sum ca_adjusted_earnings for the head or the spouse, as head_earned and spouse_earned do for adjusted_earnings.
• The exclusion: ca_dependent_care_assistance_exclusion raises the federal exclusion by ca_cdcc_min_head_spouse_earned - min_head_spouse_earned, so the floor fix removes the false increase there too.
• Tests:
  • ca_cdcc.yaml:129-164, your TY2021 household with the $4,000 loss: ca_spouse_earned 0, ca_cdcc 0, ca_income_tax 1,299.18 (80,394 taxable; 2,357.18 - 1,058 exemption credits).
  • ca_cdcc.yaml:166-198, TY2025, a $3,500 loss that offsets all federal wages but only part of the California wages: ca_spouse_earned 500 (not 1,000), ca_cdcc 34.
  • ca_dependent_care_assistance_exclusion.yaml:103-131: $5,000 benefits and a spouse loss that offsets all California wages; the exclusion stays 0 (it was 1,000).
  • New tests/policy/baseline/gov/states/ca/tax/income/ca_adjusted_earnings.yaml: HSA and 401(k), a partial loss (500) and a loss larger than the wages (0).

A1: California's own child minimum

• New parameter policyengine_us/parameters/gov/states/ca/tax/income/amt/exemption/child/amount.yaml with the FTB values from the Schedule P (540) instructions, line 22, Exemption Worksheet line 7: 7,850 (2021), 8,300 (2022), 8,950 (2023), 9,450 (2024), 9,750 (2025). The federal IRC 59(j) amounts are 7,950, 8,200, 8,800, 9,250 and 9,550. Uprating follows exemption/amount.yaml (gov.states.ca.cpi).
• ca_amt_exemption.py:36-37: worksheet line 7 now reads this parameter instead of the federal one.
• Tests: the TY2025 case at ca_amt_exemption.yaml:99-103 expects 18,750 = min(9,750 + 9,000, 92,749). A new 2022 case at ca_amt_exemption.yaml:105-128 has no HSA or deferral inputs: min(8,300 + 5,000, 84,550) = 13,300 (the federal amount gives 13,200). On the HSA eligibility point: the 2025 case checks the worksheet arithmetic on stipulated inputs only, and the 2022 case does not depend on it.

New York 2026

Agreed: this is an open interpretation, not a confirmed defect, and I made no change. Tax Law 606(c-2)(2)(C) omits IRC 32(c)(2)(A)(i)'s gross-income clause, and section 607(a) can be read to import the federal meaning. ny_cdcc_qualifying_expenses adds the excluded wages back, so the 2026 results stay as on main, and the add-back sits after the zero floor (a known gap in the PR body). DTF has not published the 2026 IT-216 instructions. When it does, the line that starts the earned income computation (federal wages or gross wages) settles it, and I will open a follow-up then.

Checks

• On 0623f5606f, 9 of the new or changed cases fail with the values in your review (ca_cdcc 204 against 272, ca_spouse_earned 1,000 against 0 and 500, ca_cdcc_min_head_spouse_earned 1,000 against 0, ca_amt_exemption 18,550 and 13,200, and the three 2022 ca_cdcc cases that now feed ca_federal_capped_cdcc). On ec4438cb78 all pass.
• Local runs on ec4438cb78 (Python 3.14, macOS): gov/states/ca/tax 295 passed; gov/irs/credits/cdcc, adjusted_earnings.yaml and eitc_earned_income.yaml 155 passed; tests/policy/baseline/partners 623 passed (no partner value moves, so no partner file changed); state_income_tax_excludes_use_tax.yaml, contrib/states/ca/prop3, contrib/federal/abolish_payroll_tax.yaml, reform/cdcc_single_parent_work_requirement.yaml and the federal AMT folder 161 passed; test_eitc_elective_deferral_invariants.py and seven code-health files (reference URLs, source conflicts, parameter keys, uprating placement) 260 passed.

Not changed here (pre-existing, for follow-up)

• 2026 rate. California applies the IRC as of January 1, 2025 (R&TC 17024.5), so its 2026 line 7 table may stay at the pre-P.L. 119-21 35%-20% schedule (pre_obbba_cdcc models that for other states). Main multiplies the live federal credit, so the credit keeps the federal 2026 rate, as before.
• Federal tax cap. FTB 3506 lines 6-10 have no federal tax limit, but ca_federal_capped_cdcc (2021, on main) and federal cdcc (2022 on, on main) both cap the base at federal tax before credits. The cap is unchanged.
• Overlap. Phase out the California AMT exemption at 25% instead of the federal 2026 rate #10501 merged while I worked, so e3bb2ad416 merges main: ca_amt_exemption.py keeps main's 25% phase-out (line 5) and this branch's lines 7-8, and ca_amt_exemption.yaml keeps both sets of cases (CA CDCC and AMT folders, 88 cases, pass after the merge). Apply Arizona, Arkansas and California rules for a filer who can be claimed as a dependent #10053 and Keep claimable-filer rules in the repeal of state dependent exemptions #10134 append cases to ca_cdcc.yaml and ca_cdcc_relevant_expenses.yaml; the second PR to merge needs a small rebase.

I re-requested your review.

[hua7450]

Thanks, Pavel. I re-reviewed at e3bb2ad. C1, C2 and A1 are fixed:

• C1. ca_cdcc reads ca_federal_capped_cdcc from 2021. The TY2025 household now gives ca_cdcc 272 and ca_income_tax 503.38.
• C2. ca_adjusted_earnings nets a loss against California wages before the floor. The TY2021 household gives ca_cdcc 0 and ca_income_tax 1,299.18.
• A1. The new California child amounts match the 2021 to 2025 Schedule P (540) instructions, line 7.

The rest of your reply checks out:

• The California credit cap equals cdcc_credit_limit.
• ca_cdcc_rate gives the same values as before for 2022 to 2025.
• Ordinary California households (no deferrals, HSA, losses or employer benefits) get the same 2022+ result as main's federal cdcc path.
• Main's five newer commits touch nothing this PR reads.

I recomputed every new and changed case in the full diff, and they match.

One thing is left before I can approve.

New York 2026: the add-back goes on after the zero floor

ny_cdcc_qualifying_expenses.py:34-42 adds employment_income - irs_employment_income back onto head_earned and spouse_earned, which are already floored at zero. When a self-employment loss wipes out a person's taxable wages, the deferred wages come back on top of the zero. This is the C2 pattern, still in place for New York.

TY2026, NY, joint return. The head has $90,000 of wages. The spouse has $4,000 of wages with a $1,000 traditional 401(k) deferral and a $4,000 Schedule C loss. Two children aged 5 and 3, and $9,000 of care:

• At head: spouse earned = max(3,000 − 4,000, 0) + 1,000 = 1,000, so qualifying expenses are 1,000 and the refundable ny_cdcc is about $365.
• On main: max(4,000 − 4,000, 0) = 0, so the credit is 0.

This doesn't depend on how the open 606(c-2)(2)(C) question is settled. Netting the loss against gross wages, the reading the PR adopts to keep main's results, gives 0. The federal 32(c)(2) measure also gives 0. A literal "items of gross income" reading gives at least 4,000. No reading gives 1,000. The PR body lists this under known gaps, but the inputs exist and main handled this household.

The helper you added for C2 fixes it directly. Per person, use max_(0, earnings_before_zero_floor(person, period, parameters, person("employment_income", period))). With gross wages as the wage base, excluded_wages is 0 and the loss offsets the wages before the floor. Then sum by head and spouse, as ca_head_earned and ca_spouse_earned do. Please add a 2026 case to ny_cdcc_qualifying_expenses.yaml with a spouse whose loss offsets all taxable wages (expected 0, margin 0.01), and drop the add-back bullet from the known gaps.

Nits, none blocking

1. 2026 California rate. This isn't a gap. R&TC 17052.6(a)(1) ties the credit to section 21 "as applicable for federal income tax purposes for the taxable year". That specific provision overrides 17024.5's January 1, 2025 specified date, so the federal 2026 cdcc_rate is right. Please drop it from the known gaps, and cite 17052.6(a)(1) in the ca_cdcc_rate comment, which now says only that the FTB has not published a 2026 table. A 2026 ca_cdcc_rate case would also pin the second phase-out you added. For example, head of household, AGI $80,000: max(.50 − .33, .35) − .03 = 0.32.
2. Employer benefits from 2022. From 2022 the California credit reaches FTB 3506 Part IV. For a household with employer dependent care benefits and expenses below the dollar limit, it now nets the benefits through federal Form 2441 line 31, so the result moves off main. That matches the form, but no case pins it, and the body says results change only for payroll-deduction filers.
3. PR body.
   • Line 27 should also mention California child-AMT filers (A1, every year) and the employer-benefit change above.
   • The 10501 rebase note and the new-case count are out of date.
   • The state table leaves out AZ (2026+), MD and SC, which read cdcc_relevant_expenses and also change.
   • The Minnesota row quotes the pre-2023 text of 290.067 subd. 1. Subd. 1a(d) now imports 21(d).
   • The 2017 Form 6251 instructions share Form 8615's "personal services" sentence, but only Form 8615 points to Form 1040 line 7.
   • The cdcc_single_parent_work_requirement reform no longer flows into the California credit from 2022, because California now uses ca_cdcc_relevant_expenses. A sentence would record that.
4. Tests.
   • ca_cdcc.yaml: the 2021 and 2022 blocks are now one class, and their cdcc: 0 input feeds nothing.
   • ca_amt_exemption.yaml: seven existing cases (lines 11, 30, 48, 66, 90, 113, 134) still input head_earned, which the formula no longer reads. ca_cdcc_relevant_expenses.yaml renamed the same kind of input to ca_cdcc_min_head_spouse_earned.
   • The new CA AMT case gives a dependent 17-year-old payroll HSA contributions and asserts federal head_earned 8,000. Under Pub. 969 a dependent can't receive tax-free HSA contributions. A 401(k)-only child would test the line 8 arithmetic without that question. ca_adjusted_earnings.yaml already covers HSA with an adult.
   • The edited section 129 case in ca_cdcc_relevant_expenses.yaml inputs only ca_cdcc_min_head_spouse_earned. The California exclusion it describes is therefore 0, not 5,000; 1,000 still passes for another reason. Adding min_head_spouse_earned: 50_000, or real dependent_care_employer_benefits, plus an exclusion assertion, would test it.
   • One new comment still describes the round-1 post-floor HSA addition.
   • test_eitc_elective_deferral_invariants.py now checks amt_exemption, but at YEAR 2026 the child cap is off, so that assertion can't fail.
5. Code.
   • max_(ca_taxable_wages, irs_employment_income) never binds, since ca_taxable_wages subtracts a subset of the federal pre-tax list. Use ca_taxable_wages directly, or keep it with a NOTE.
   • ca_dependent_care_assistance_exclusion works out to min(benefits, dollar cap, ca_cdcc_min_head_spouse_earned), so it could say that directly, as FTB 3506 lines 21 to 26 do.
   • The federal amt/exemption/child/in_effect.yaml comment still says California reads the federal child amount.
   • The new parameter's label could follow the source: "minimum exemption amount for certain children".
   • Uprating it after 2025 gives unrounded amounts, while the published ones are multiples of $50 (59(j)(2)).
   • child_dependent_care/input.yaml still selects federal cdcc before 2021. California tax runs for backfilled years, so those keep the C1 behavior. Either start the California entry earlier, or say the routing covers 2021 on.

Follow-ups for separate PRs (all the same on main)

• ca_federal_capped_cdcc still caps the base at federal tax. R&TC 17052.6(a)(2) applies California's percentage "without taking into account whether there is a federal tax liability".
• ca_cdcc calls p.rate.calc(agi) on band-top thresholds without right=True. AGI of exactly $40,000, $70,000 or $100,000 drops a band (for example, $100,000 gets 0 instead of .34).
• Federal cdcc_relevant_expenses skips Form 2441 line 30, which subtracts excluded benefits. The federal exclusion is also not limited to qualified expenses (lines 16 and 17).
• mn_cdcc does not reduce its dollar limit by the section 129 exclusion.

[PavelMakarchuk]

@hua7450, thank you for the re-review of e3bb2ad416 (6:52pm EDT, October 29). All the fixes below are in one commit, 9959a00e7e, which is the new head. Main merges into the branch with no conflict, so I did not merge it again.

New York 2026: the loss now offsets the wages before the floor

• policyengine_us/variables/gov/states/ny/tax/income/credits/cdcc/ny_cdcc_qualifying_expenses.py:31-44: I used your fix. Per person, the formula takes max_(0, earnings_before_zero_floor(person, period, parameters, person("employment_income", period))), then sums it for the head and for the spouse. With gross wages as the wage base, excluded_wages is 0. This is the same as adjusted_earnings on main, so New York results are as on main for every household.
• ny_cdcc_qualifying_expenses.yaml:172-202, case 11: your TY2026 household (spouse $4,000 wages, $1,000 deferral, $4,000 Schedule C loss) gives 0, with a margin of 0.01.
• ny_cdcc_qualifying_expenses.yaml:204-234, case 12: a $3,500 loss gives 500, not 1,000.
• Both cases gave 1,000 at e3bb2ad416.
• Cases 1-6 now set employment_income instead of head_earned, because the formula no longer reads head_earned.
• In the PR body, I removed the add-back bullet from the known gaps and rewrote the New York paragraph.

Nits

1. 2026 California rate. Done.

• ca_cdcc_rate.py:20-25: the comment now cites R&TC 17052.6(a)(1), "as applicable for federal income tax purposes for the taxable year". I also added that section to the references.
• ca_cdcc_rate.yaml:49-60: a 2026 head of household with AGI $80,000 gives 0.32. Main gives 0.35, but that rate did not reach the credit on main.
• I removed the 2026 rate from the known gaps.

2. Employer benefits. Done. ca_cdcc.yaml:192-223 pins the FTB 3506 Part IV path in TY2025. The household has $60,000 wages, $2,000 of benefits, one child and $2,500 of care. Line 29 is 1,000, line 30 is 500 and line 31 is 500. So ca_cdcc_relevant_expenses is 500 and ca_cdcc is 43. Main gives 86. The body now lists this change.

3. PR body. All six points are done:

• The "What changes" section now lists the California child AMT filers, the employer benefit change, business losses and the years before 2021.
• The Phase out the California AMT exemption at 25% instead of the federal 2026 rate #10501 note is current. The old case count is replaced by a statement that each table row was run against main (bf2ec882ba).
• The table has AZ (from 2026), MD and SC rows.
• The Minnesota row cites 290.067, subd. 1a(d).
• The AMT paragraph says that only the Form 8615 instructions point to Form 1040 line 7.
• The body has a sentence on the cdcc_single_parent_work_requirement reform.

4. Tests. Done:

• ca_cdcc.yaml: I removed the eight cdcc: 0 inputs. I kept both the 2021 and the 2022 cases, because 2021 still has its own rate branch.
• ca_amt_exemption.yaml: the seven head_earned inputs are now ca_head_earned.
• ca_amt_exemption.yaml:142-171: the child now has a 401(k) deferral and no HSA. head_earned and ca_head_earned are 9,000, and the exemption is still 18,750.
• ca_cdcc_relevant_expenses.yaml:79-93: the section 129 case now sets min_head_spouse_earned: 50_000 and checks ca_dependent_care_assistance_exclusion 5,000.
• ca_cdcc_relevant_expenses.yaml:370: the round-1 comment now describes ca_adjusted_earnings.
• test_eitc_elective_deferral_invariants.py: I removed amt_exemption and the AMT inputs from the 2026 runs. The eitc_earned_income.yaml case "AMT kiddie-tax exemption caps at taxable earnings before 2018" still covers the pre-2018 cap.

5. Code.

• max_: done. ca_adjusted_earnings.py:34-36 now uses ca_taxable_wages directly, with a comment that explains why.
• ca_dependent_care_assistance_exclusion: not changed. It gives the same values as min(benefits, cap, ca_cdcc_min_head_spouse_earned). I keep the federal exclusion as the base because FTB 3506 Part IV repeats Form 2441 Part III. When the federal lines 16-17 follow-up lands, California then gets that fix too.
• amt/exemption/child/in_effect.yaml:10-11: done. The comment now says that California reads its own amount.
• Label: done. It is now "California AMT minimum exemption amount for certain children".
• Rounding: not changed. The published amounts do not follow a plain step of gov.states.ca.cpi (7,850 x 322.043 / 297.447 = 8,499, but the 2022 amount is 8,300). So I cannot tell which rule the FTB uses after its CPI step. The federal child amount also uprates without rounding. I will set 2026 when the FTB publishes it.
• child_dependent_care/input.yaml:3-8: done. The California entry now starts in 2012, because R&TC 17052.6(a)(1) has applied the computation since 2000. Your TY2025 household in 2020 now gives ca_cdcc 272 instead of 204.

Follow-ups for separate PRs (the same on main)

These four items are now in the known gaps of the PR body:

1. ca_federal_capped_cdcc caps the base at federal tax, but R&TC 17052.6(a)(2) applies California's percentage "without taking into account whether there is a federal tax liability".
2. ca_cdcc calls p.rate.calc(agi) without right=True, so AGI of exactly $40,000, $70,000 or $100,000 drops a band.
3. Federal cdcc_relevant_expenses skips Form 2441 line 30. Also, the federal exclusion is not limited to qualified expenses (lines 16 and 17).
4. mn_cdcc does not reduce its dollar limit by the section 129 exclusion (290.067, subd. 1a(c)).

Checks on 9959a00e7e (Python 3.14, macOS)

• gov/states/ca/tax: 312 passed.
• gov/states/ny/tax: 441 passed.
• gov/irs/credits: 732 passed. Two batches timed out while the Mac slept. I ran those two folders again directly, and their 69 cases passed in 13 seconds.
• tests/policy/baseline/partners: 623 passed. No partner value moved, and no partner file changed.
• adjusted_earnings.yaml, the federal AMT folder, reform/cdcc_single_parent_work_requirement.yaml, contrib/states/ny/a06774 and contrib/states/ca/prop3: 168 passed.
• The invariant file and seven code-health files: 260 passed.

I re-requested your review.

[hua7450]

Approving at 9959a00. Thanks, Pavel.

The round-2 blocker is fixed. ny_cdcc_qualifying_expenses now nets the loss against gross wages before the floor, so the example household gives 0 and the $3,500 loss case gives 500, the same as main. I recomputed every new and changed case in this round, and they match, including the 2026 ca_cdcc_rate of 0.32 and the employer-benefit ca_cdcc of 43 (500 × .20 × .43). The move of the California entry to 2012 is right: your 2020 example gives 272 by hand.

Before merging: please merge main. The branch is 103 commits behind. I compared the trees and found no textual conflict, and no test on main that this should break, but that was by reading, not running. The one failed check, Contrib tests (other-shard-1), was cancelled at the 60-minute limit, and PR 10556 on main addresses that timeout.

Of your two declined items, leaving ca_dependent_care_assistance_exclusion as it is holds: the values are equal, and your reason about the federal lines 16 and 17 follow-up is sound. The rounding reason doesn't hold; see nit 2.

Nits, none blocking:

1. Child AMT amount before 2021. amt/exemption/child/amount.yaml starts in 2021, so 7,850 backfills to 2015-2020. FTB's Schedule P worksheet line 7 prints 7,400, 7,400, 7,500, 7,600, 7,750 and 7,950 for those years. Main's federal amount matched 2015-2019 and was $50 low for 2020, so this PR moves those years off the published amounts. I'm not treating it as a gap because the child branch needs is_qualifying_child_dependent set on a head by hand, and the other California AMT parameters also backfill flat from 2021. Adding the six dated values would fix it, and would make the body's "in every year" true for line 7. The effect is at most $31.50 of AMT (2015 and 2016).
2. Child AMT amount from 2026. You're right that California CPI doesn't produce the series, but there's no California CPI step to find: R&TC 17062(b)(7) reaches the paragraph (5) and (6) amounts, and the child limit comes through 17062(b)(1) from IRC 59(j). The 2022-2025 amounts match 59(j)(2) on unchained CPI-U, rounded to the nearest $50, which would give 10,000 for 2026. uprating: gov.states.ca.cpi is therefore the wrong index, though the effect is only about $2 of tax. Either use CPI-U with $50 rounding, or wait for FTB's figure and correct the reason.
3. ca_cdcc.yaml, 2021 and 2022 blocks. The 2021 rate branch you kept them for is in ca_cdcc_rate, and these eight cases input ca_federal_capped_cdcc, so they never reach it. ca_cdcc has no year branch, so lines 37-72 repeat lines 1-36.
4. Employer-benefit comment, ca_cdcc.yaml:192-195 and :218. Lines 29, 30 and 31 at 1,000, 500 and 500 are Form 2441 lines. On FTB 3506 they are 2,000, 1,000 and 2,500, then line 32 is 500 and line 33 is 500. The expected value is right.
5. Untested paths. No ca_cdcc case runs before 2021, so nothing pins the new 2012 route; the 2020 household at 272 would. Nothing reaches the deemed-income branch of ca_cdcc_min_head_spouse_earned either. For example, TY2025, head wages $60,000, full-time student spouse with no wages, one child, $4,000 of care: deemed 3,000, ca_cdcc 258.
6. Comments and descriptions.
   • input.yaml still ends "based on year", though it now has one value.
   • ny_cdcc_qualifying_expenses.py:37 binds earned to a person array, and line 45 rebinds it to a tax-unit array. person_earned would be clearer (rename the reads on lines 43 and 44 too).
   • The ca_cdcc_rate comment explains 2026 but not why 2021 differs. R&TC 17052.6(f), and FTB's section 70405 entry for 2026, would cover both.
   • test_eitc_elective_deferral_invariants.py:218 loops over a one-item tuple.
7. PR body.
   • The AMT paragraph still says the AMT variables can "add back the excluded wages the way New York does below". New York no longer adds back; earnings_before_zero_floor on gross wages is the pattern now.
   • The Minnesota row cites subd. 1a, which applies only from TY2026. Earlier years rest on subd. 1 ("pursuant to the provisions of section 21"). My round-2 "pre-2023" was wrong on that.
   • The Arizona row says "Correct", but A.R.S. 43-1022(34) doesn't state the 21(c) and (d) limits. The Maryland row's wording fits better: it reads cdcc_relevant_expenses and keeps those limits, as on main.
   • The Form 8615 sentence you quote follows "In that case", for business income from personal services. 2017 Pub. 929 states the same rule without that qualifier.
   • California's "care provided in this state" limit (R&TC 17052.6(d)(1); FTB 3506 line 2(e)) isn't modeled, and there's no input for it. It belongs in the known gaps now that the body describes this path as California's own limits.

Follow-up, outside this PR: filer_adjusted_earnings has no formula reader left on head or main, and wa_working_families_tax_credit.yaml still inputs it, so WA Case 3 passes through earnings > 0 rather than the ceiling.

Not verified: I didn't read the pre-2025 wording of R&TC 17052.6, and nothing here was run.