[MaxGhenis]

Summary

Kentucky's household and dependent care service credit is 20% of "the federal credit allowed under Section 21 of the Internal Revenue Code" (KRS 141.067). KRS 141.010(21) fixes that Code at a reference date. 2026 Ky. Acts ch. 161 (HB 757), sec. 6 moved the date from December 31, 2024 to December 31, 2025, for taxable years beginning on or after January 1, 2026. The One Big Beautiful Bill Act (P.L. 119-21) was enacted on July 4, 2025, so Kentucky's 2026 Code includes its section 21 rate increase. The same Kentucky act's changes to KRS 141.019 (sec. 7) and 141.0101 (sec. 55) do not decouple section 21, and KRS 141.067 is unchanged.

The model still read the pre-OBBBA federal credit from 2026, through a hard-coded period.start.year >= 2026 branch added in #8933 (when the reference date was still December 31, 2024). A 2026 single parent with $50,000 of earnings and $1,000 of care for a child of five got $40; the law gives $70. Found by the hub's review of #10045 (should-fix 3).

Kentucky's reference date by tax year

Read from each amending Act (each is the KRS 141.010 definition of "Internal Revenue Code"):

Tax years	Code in effect on	Source
2018	Dec 31, 2017 (with P.L. 115-97)	2018 Ky. Acts ch. 207
2019-2021	Dec 31, 2018	2019 Ky. Acts ch. 151, sec. 35 (2021 Ky. Acts ch. 31, sec. 1 renumbered it only)
2022	Dec 31, 2021, excluding P.L. 117-2 sec. 9673	2022 Ky. Acts ch. 212, sec. 42
2023	Dec 31, 2022	2023 Ky. Acts ch. 92, sec. 17
2024	Dec 31, 2023	2024 Ky. Acts ch. 166, sec. 14
2025	Dec 31, 2024	2025 Ky. Acts ch. 98, sec. 30
2026 on	Dec 31, 2025	2026 Ky. Acts ch. 161, sec. 6

Only the 2026 date includes the Act.

Changes

• New parameter gov.states.ky.tax.income.irc_conformity.obbba: Kentucky adopts the Act's Code amendments if true. False from 2018, true from 2026. It cites KRS 141.010(21) and both Acts, and notes that Kentucky's specific decouplings (KRS 141.019 and 141.0101, such as sections 174A, 224 and 225) are outside it.
• ky_cdcc_potential reads cdcc when the parameter is true and pre_obbba_cdcc otherwise, replacing the hard-coded year.
• pre_obbba_cdcc_potential, pre_obbba_cdcc, pre_obbba_capped_cdcc now return their live variables (cdcc_potential, cdcc, capped_cdcc) for tax years before the Act's section 21 amendment applies. Before, they applied the 2025 rate schedule to any year, which was wrong for 2021 (the refundable American Rescue Plan Act credit) and ignored cdcc inputs.
  • They key on the Act's statutory effective date, P.L. 119-21 sec. 70405(b) ("shall apply to taxable years beginning after December 31, 2025"), held as the constant OBBBA_SECTION_21_EFFECTIVE beside the existing frozen-schedule instant.
  • The switch is a fixed date, so no reform moves it. The frozen rates are read from the simulation's own parameters at 2025-01-01 (as on main), so a rate reform dated 2026 or later leaves them alone, while one that also changes 2025's rates reaches them. A YAML case switches off the amended second phase-out in 2026 and checks the mirrors stay at $200.
  • The expense base and liability limit stay live, since the Act did not amend section 21(c)-(e) or section 26. A reform of those, such as a higher expense cap, still reaches the mirrors, and so does the Act's own section 129 change (see Known gaps).
  • Invariant 6 checks that, over 2022-2035, the baseline schedule departs from the frozen one exactly from that date.
  • With this, Kentucky reads the mirror in 2018-2025 and gets the live credit, so pre-2026 results don't change, and a cdcc input still flows through (the existing 2023 ky_cdcc.yaml case and the 2025 integration.yaml case).
  • Maine and Vermont read the mirrors only from 2026, so their baseline results are unchanged (their CDCC files pass). Under a reform switching off the amended second phase-out, they now keep the pre-OBBBA credit, as their static conformity requires; round 1 would have given them the live credit.

Audit of Kentucky's other federal-conformity readers

I grepped Kentucky's variables and parameters for frozen or year-gated federal reads (pre_obbba, period.start.year, parameter instants, frozen). ky_cdcc_potential was the only one. Every other Kentucky reader of federal law reads the live federal variables, so none is pinned to a stale Code. Some have their own gaps (below):

• ky_itemized_deductions_unit adds itemized_deductions_less_salt and subtracts medical_expense_deduction. So from 2026 it includes the Act's 0.5% charitable floor, which Kentucky now adopts (KRS 141.019(2) allows Chapter 1 deductions "except" a list that doesn't name section 170). It also has existing gaps unrelated to the reference date. Mortgage interest on a second home passes through, though KRS 141.019(2)(j) limits it to one principal residence from 2026 (Cap Kentucky qualified residence interest at principal-residence interest from 2026 #10003 handles that). Non-wagering casualty losses pass through, though KRS 141.019(2)(b) excludes section 165 losses other than wagering losses. The review also notes that wagering losses are not in the imported list; I have not verified that.
• ky_tuition_tax_credit_potential is 25% of the live American Opportunity and Lifetime Learning credits.
• ky_agi starts from federal AGI.
• The combined-separate helper (_ky_combined_separate.py) reads ky_cdcc_potential, so it picks up the fix.

No other provision changes output with the new reference date, so the "another affected provision" tests are downstream ones: ky_income_tax for the reviewer's household, and the federal mirror in 2021.

Before and after

YAML expectations are hand-computed from IRC 21(a)(2), IRC 26 and KRS 141.067. On unchanged main (fab18b7d6b), the first column fails where it differs.

Case (ky_cdcc.yaml unless noted)	Main	This PR
2026 single parent, $50,000, one child of 5, $1,000 care: ky_cdcc_potential	$40	$70 (35% x $1,000 = $350 federal)
Same household, ky_income_tax	$1,592.40	$1,562.40 ($46,640 x 3.5% less $70)
Same household in 2025	$40	$40 (pre-OBBBA year, unchanged)
2026 joint, $50,000, two children, $6,000 care (liability-capped)	$240	$356 (20% of the $1,780 capped federal credit; was the old 2026 test's $240)
2026 joint, $160,000 (second phase-down)	$240	$384 (32% x $6,000 = $1,920 federal)
2026 joint, $220,000 (bottom of the second phase-down)	$240	$240 (both schedules give 20%)
2026 single parent, reform: Kentucky conformity off	n/a (no parameter)	$40 (20% of the $200 pre-OBBBA credit)
2025 single parent, reform: Kentucky conformity on	n/a (no parameter)	$40 (the Act doesn't change 2025)
pre_obbba_cdcc.yaml: 2026 single parent, reform: amended second phase-out off	$200	$200 (live $350; failed at this PR's round-1 head)
pre_obbba_cdcc.yaml: 2021 joint, $50,000, two children, $6,000	$1,200	$3,000 (live refundable 2021 credit)

New-case results: 3 of the 7 baseline ky_cdcc.yaml cases and the 2021 pre_obbba_cdcc.yaml case fail on main (the two Kentucky reform cases need the new parameter, so they can't run there); all 9 and all 4 pass here. The amended-phase-out reform case fails at the round-1 head (9d2bddeb60) with $350, as the review predicted, and passes now.

Invariants

Stated in test_ky_cdcc_federal_match.py's docstring and checked there on a 16-unit boundary grid and 10 Hypothesis batches of up to 20 Kentucky tax units (the reform lever is in YAML):

1. ky_cdcc_potential = 0.2 * cdcc in 2025, 2026 and 2027.
2. Where the Act doesn't change section 21 (2021, 2025), each pre-OBBBA mirror equals its live variable.
3. From 2026, cdcc >= pre_obbba_cdcc and cdcc_potential >= pre_obbba_cdcc_potential: the Act never lowers the credit, so conformity never lowers Kentucky's.
4. cdcc_potential equals an independent reference written from IRC 21(a)(2), and from 2026 the Act's schedule, times the relevant expenses (a differential test of the federal schedule Kentucky now inherits).
5. 0 <= ky_cdcc <= ky_cdcc_potential.
6. From the Act's effective date on, and only then (2022-2035 checked), the baseline section 21 parameters depart from the frozen pre-OBBBA schedule in both the (max, min) rates and amended_structure.applies.

The conformity lever is checked in ky_cdcc.yaml. With the parameter off in 2026, Kentucky gets 20% of pre_obbba_cdcc ($40); with it on in 2025, nothing changes. Invariant 3 shows conformity never lowers the credit.

Mutation checks (local), each made and reverted:

• Making ky_cdcc_potential always read pre_obbba_cdcc (the old behaviour) fails invariant 1.
• Making the mirrors always frozen (the old federal behaviour) fails invariant 2.
• Restoring the round-1 marker (amended_structure.applies) fails the amended-phase-out reform case in pre_obbba_cdcc.yaml.

Test cost

The hub asked for a property test of the 20% match. Python is needed for the Hypothesis generator, and for a reference rate written from the statute rather than from the model's parameters. The reform lever needs no Python: YAML parameter inputs cover it.

• Design. Each batch is one Simulation holding all four years, and no reformed system is built in Python.
• Local measurements (macOS arm64, fresh process, /usr/bin/time -l, thread caps 4):
  • New module alone: 3 passed, 12.3 s in pytest, 25.8 s wall, 2.13 GiB peak RSS.
  • The existing test_ky_non_refundable_credit_order.py alone: 19.0 s wall, 1.50 GiB peak.
  • Both in one process: 37.7 s wall, 2.22 GiB peak. So the new module adds about 19 s and 0.72 GiB in a fresh process.
  • Most of the memory is per-year parameter caches on the shared system, which stay resident after each Simulation is freed. Other tests in a CI process warm those years too, so the in-group increment should be smaller still.
  • The three YAML parameter-input cases each build a reformed system: ky_cdcc.yaml went from 8.3 s to 25.5 s and pre_obbba_cdcc.yaml from 9.3 s to 17.3 s locally, in the state and IRS YAML shards.
• CI group. The Python file falls in Rest Python shard 2, group policy. In CI run 39155269843 (2026-10-18) that group took 7:40 at an 11.03 GiB peak (rest-python-policy-resources.txt). At this head, CI run 39173805147 put the same group at 9:16.50 and an 11.09 GiB peak (11,632,368 KB), about 0.06 GiB (0.5%) above that baseline. These are Linux CI numbers, separate from the local ones above.

Known gaps (not changed here)

• 2019-2021. The reference date was December 31, 2018, so Kentucky did not adopt the American Rescue Plan Act's 2021 section 21 expansion (2021 Form 740 instructions: "new form 2441-K") or the Consolidated Appropriations Act's 2021 Lifetime Learning credit phase-out (Form 8863-K lines 9(a)-(e)). The model reads the live 2021 credits; the code comment in ky_cdcc_potential says so. Separate follow-up.
• Section 68 from 2026. Kentucky's 2026 Code includes section 68 as the Act rewrote it (the 2/37 limitation), and KRS 141.019(2) allows the Chapter 1 deductions without excepting it. ky_itemized_deductions_unit applies no section 68 reduction. I have not checked how Kentucky's forms apply it (no 2026 Schedule A exists yet), so this needs its own check. Separate follow-up.
• Maine and Vermont keep the same hard-coded 2026 branch from CDCC coupling accuracy: SC § 21(e)(4), ID/GA reform gating, KY/ME/VT static conformity #8933. Whether their 2026 conformity laws moved past the Act needs its own check.
• Section 129 in the mirrors (from 2026, as on main). P.L. 119-21 sec. 70404 raised the section 129 dependent care assistance exclusion cap from $5,000 to $7,500. Section 21(c) reduces the expense limit by that exclusion, and the mirrors use the live cdcc_relevant_expenses, so their expense base reflects the Act. Example: a 2026 joint couple with $50,000 of wages each, two qualifying children aged 3 and 6, $6,000 of care and $6,000 of employer benefits gets a $0 mirror credit, against $200 with the pre-Act $5,000 cap ($6,000 less $5,000, at 20%). This affects eligible filers with more than $5,000 of employer benefits. Married couples filing separately can't claim the credit unless treated as unmarried, and then use the $5,000 single cap. It reaches Maine and Vermont, whose static conformity reads the mirrors, and the Kentucky conformity-off reform, but not Kentucky's baseline, which reads the live credit from 2026. It goes to the Maine and Vermont follow-up.

Methodology choices

None new. One existing reading stays: "the federal credit allowed" is cdcc, the section 21 credit after the IRC 26 limit (the smaller of the computed credit and the Form 2441 Credit Limit Worksheet amount), as before this PR.

Hub microsimulation impact (current head)

A real microsimulation on the default dataset, main at the merge base fab18b7d6b against this PR's head f78b6e0d97, 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)
ky_cdcc	+0	+2.23M	0 (0)	29 (50,431)
ky_income_tax	+0	-2.23M	0 (0)	29 (50,431)
ky_non_refundable_credits	+0	+2.23M	0 (0)	29 (50,431)
cdcc	+0	+0	0 (0)	0 (0)
state_income_tax	+0	-2.23M	0 (0)	29 (50,431)
income_tax	+0	+0	0 (0)	0 (0)
household_net_income	+0	+2.23M	0 (0)	29 (50,431)
household_benefits	+0	+0	0 (0)	0 (0)

2025 shows no change, as expected: Kentucky's conformity date moves for 2026 only.

Comparison files are in fixes/10267-impact/ on the hub host.

Impact

The hub runs the microsimulation impact. From 2026 the Kentucky credit rises for filers whose federal credit allowed rises: a 2026 section 21 rate above 20%, and federal liability above the pre-OBBBA credit. For example, a joint couple with $30,000 of wages, two children and $6,000 of care has a 42% rate but no federal liability, so both credits stay $0. 2018-2025 are unchanged.

Review round 1

GPT-6.1 Sol reviewed 9d2bddeb60 and requested changes. It confirmed the law, every reference date and anchor, and every YAML expectation; 50 tests passed. Responses:

1. Marker could follow a reform (should-fix): fixed. The mirrors key on the statutory effective date, and a new YAML case covers the reform that broke round 1.
2. YAML supports parameter reforms (should-fix): fixed. The lever is tested in ky_cdcc.yaml, the Python reformed system is gone, and the justification above is corrected.
3. Axiom queued lacked a blocker and corpus check (should-fix): rulespec-us#1793 now records the blocker (signed, billed encoder runs dispatched by the Axiom eng hub after production-signing approval, d1142/d1159) and the corpus facts. The KRS 141.010 source in scope 2026-09-14-income-tax-chapter is byte-identical to the official PDF. The union release lineage from us-rulespec-2026-09-14-wave4-union through us-rulespec-2026-10-13-w7-bundle-gaps-union carries that scope; four later releases do not, so the issue names them and asks for a compatible release to be pinned at dispatch. The citation path at dispatch is left as a stated check.
4. Audit claim too broad (nit): narrowed; the mortgage, casualty and wagering gaps are listed.
5. Impact overstated (nit): qualified with the liability-cap example.
6. Stale YAML comment; "the Act" in the issue (nit): fixed.

Review round 2

GPT-6.1 Sol approved 338fd8db43 (~/reviews/us-hub/reviews-gen/us-10267/review-r2.md): every round-1 finding fixed or answered, all 13 YAML cases agree with hand arithmetic, and the new reform case fails against the round-1 code as intended. It left three nits, all addressed:

1. Not every corpus release after 2026-09-14 carries the Kentucky chapter scope: rulespec-us#1793 now names the releases without it and asks for a compatible release to be pinned.
2. The mirrors' reform immunity was overstated: the body now separates the fixed switch from the frozen rates (read at 2025-01-01, so reforms dated 2025 or earlier reach them) and the live expense base and liability limit. Maine and Vermont are qualified as unchanged in the baseline.
3. Invariant 6's docstring now states its 2022-2035 scope (f78b6e0d97, docstring only).

Review round 3 (delta)

A delta review (Opus 5.5, ~/reviews/us-hub/reviews-gen/us-10267/review-r3.md) approved the code at f78b6e0d97: docstring only, matching the test, ruff clean. It ran the head and found two body claims still too strong, both fixed above with no code change:

• Rate reforms dated 2025 or earlier do reach the frozen rates.
• The Act's section 129 increase reaches the mirrors' expense base; this is now a Known gaps entry.

It also tightened the invariant 6 and release-lineage wording. A read-only confirmation (GPT-6.1 Sol, review-r4.md) verified those fixes and asked for one more precision fix to the section 129 example (the household's children and wages, and the married-filing-separately case), applied here.

axiom: us-ky/statute/krs/141.010 (KRS 141.010(21) composed into us-ky/statutes/11/141/067) TheAxiomFoundation/rulespec-us#1793 queued. The 141.067 module computes 20% of a caller-supplied federal credit, and no module encodes the reference date or connects us:statutes/26/21#cdcc; the issue is dispatch-ready under d1142 with companion tests from the same statute arithmetic as the YAML.

🤖 Generated with Claude Code

[MaxGhenis]

Hand-off to the US + core hub

Head: f78b6e0d971e5302516dd9e587a154489d710b34

Independent review (briefs, outputs and staged primary sources in ~/reviews/us-hub/reviews-gen/us-10267/):

Round	Reviewer	Head	Verdict
1	GPT-6.1 Sol	9d2bddeb60	Request changes: 3 should-fix, 3 nits; all fixed or answered at 338fd8db43
2	GPT-6.1 Sol	338fd8db43	Approve; 3 documentation nits, fixed (one docstring-only commit, f78b6e0d97)
3	Opus 5.5 (delta)	f78b6e0d97	Code delta approved; 4 PR-body text fixes, applied (no code change)
4	GPT-6.1 Sol (body text, read-only)	f78b6e0d97	Verified round 3's fixes; asked for one more precision fix to the section 129 example in the body, applied as written (no code change)

Rounds 2-4 agree on the code: f78b6e0d97 differs from the approved 338fd8db43 only in a docstring, which round 3 approved. Round 4's single request was body text, now applied.

Passed locally, one file per process:

• At f78b6e0d97: ky_cdcc.yaml (9), pre_obbba_cdcc.yaml (4), test_ky_cdcc_federal_match.py (3), vt_nonrefundable_cdcc.yaml (3), federal cdcc.yaml (11), capped_cdcc.yaml (4), integration.yaml (8), cdcc_rate.yaml (13), ky_non_refundable_credit_order.yaml (1).
• At 338fd8db43 (same code): Kentucky integration.yaml (25), vt_cdcc.yaml (6), vt_low_income_cdcc.yaml (4), Maine child_care_credit.yaml (6).

On fab18b7d6b, 3 Kentucky cases and the 2021 mirror case fail as intended. Ruff 0.15.5 is clean.

CI at hand-off (run 39173805147): 12 pass, 5 pending (Baseline contrib-hhs, reform, ssa-usda; Microsimulation; Rest Python shard 1), none failing. Rest Python shard 2, which holds the new test's policy group, starts after shard 1. GitHub reports the PR mergeable.

Pending / owed:

• Microsimulation impact: the hub runs it (disk hold until about 2026-10-19 16:00Z).
• The body promises this PR's Rest Python shard-2 policy group peak against the 11.03 GiB baseline from run 39155269843. It is in artifact rest-python-test-timings-shard-2, file rest-python-policy-resources.txt; please add it to the body at landing.
• Axiom: Encode KRS 141.010(21) and compose it into KRS 141.067: Kentucky's 2026 dependent care credit is 20% of the OBBBA section 21 credit (PE-US PR 10267) TheAxiomFoundation/rulespec-us#1793 is queued (pe-parity, d1142). The Axiom eng hub dispatches the signed encoder after production signing (d1159).

Depends on: none. Decision: none; the only reading is the existing one ("federal credit allowed" = cdcc after IRC 26).

Follow-ups (proposed as session cards; notes in the shared side-session folder ky-irc-conformity-10267.md):

1. Kentucky 2021 credits on the December 31, 2018 Code (ARPA CDCC, Form 2441-K; CAA 2021 Lifetime Learning phase-out, Form 8863-K).
2. Kentucky itemized deductions: non-wagering casualty losses pass through despite KRS 141.019(2)(b); wagering losses; the 2026 section 68 limitation.
3. Maine and Vermont: 2026 conformity date for their hard-coded pre-OBBBA branch, and the OBBBA section 129 increase reaching the pre-OBBBA mirrors' expense base.

🤖 Generated with Claude Code

[MaxGhenis]

US + core hub merge audit, us#10267 at f78b6e0d971e5302516dd9e587a154489d710b34 (squash): Use the OBBBA federal credit for Kentucky's 2026 dependent care credit.

• Review: GPT-6.1 Sol r2 (reviews-gen/us-10267/review-r2.md) APPROVED the code at 338fd8d after executing the tests: all six round-1 findings fixed or answered, 13 YAML cases matching hand arithmetic, 32 tests passing. This head f78b6e0 changes only a docstring in one test file; the hub verified the file is AST-identical with docstrings stripped. An Opus delta review (r3) and a read-only Sol pass (r4) approved the docstring and body-text follow-ups.
• 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 at this exact head against fab18b7, 2025 and 2026, in the description: 2026 ky_cdcc +$2.23M (29 records), with ky_income_tax and household_net_income moving by the same amount; 2025 unchanged. CI memory for the Rest Python shard-2 policy group: 11.09 GiB peak against the 11.03 GiB baseline.
• Order: No open dependency or decision. The only reading is the existing one: 'federal credit allowed' means cdcc after IRC 26.
• Axiom: us-ky/statute/krs/141.010 (KRS 141.010(21) composed into us-ky/statutes/11/141/067) Encode KRS 141.010(21) and compose it into KRS 141.067: Kentucky's 2026 dependent care credit is 20% of the OBBBA section 21 credit (PE-US PR 10267) TheAxiomFoundation/rulespec-us#1793 queued. The 141.067 module computes 20% of a caller-supplied federal credit, and no module encodes the reference date or connects us:statutes/26/21#cdcc; the issue is dispatch-ready under d1142 with companion tests from the same statute arithmetic as the YAML.