Fix OH refundable EITC reform zero-refund and lost-credit bugs

[PavelMakarchuk]

Fixes #8656

Problem

The bundled Ohio refundable-EITC contrib reform (gov.contrib.states.oh.child_poverty_impact_dashboard.eitc.in_effect) ran without crashing but was functionally wrong on two fronts — same shape as the MO and UT bugs fixed in #8642 and #8645, but no add() string crash:

1. Zero refund at zero liability. oh_refundable_eitc.formula returned tax_unit("oh_eitc", period) — the applied nonrefundable-capped EITC. A zero-liability filer's oh_eitc is 0, so the reform's oh_refundable_eitc was also 0 — no refund paid in exactly the case refundability is meant to address.

2. Other Ohio non-refundable credits discarded. oh_non_refundable_credits.formula returned just tax_unit("oh_non_refundable_eitc", period) (= 0 under the reform), silently zeroing out every other non-refundable credit walked by the baseline ordered list (oh_cdcc, oh_senior_citizen_credit, oh_retirement_credit, oh_non_public_school_credits, oh_exemption_credit, oh_joint_filing_credit, plus oh_adoption_credit for pre-2023 years). An OH filer claiming any of those lost them all whenever the reform was active.

Fixes in oh_refundable_eitc_reform.py

oh_refundable_eitc now reads oh_eitc_potential (uncapped at liability; OH does not apply a separate cap layer above the potential, so this is the full refundable amount). Mirrors the MO mo_wftc → mo_wftc_potential change in #8642 and the UT ut_eitc → ut_eitc_potential change in #8645.

oh_non_refundable_credits now walks the same ordered list the baseline uses (gov.states.oh.tax.income.credits.non_refundable) via ordered_capped_state_non_refundable_credits, with oh_eitc filtered out of the list since the reform pays it as refundable. Identical pattern to the UT fix.

Verification

• All four updated contrib tests pass (drive from federal eitc and oh_income_tax_before_non_refundable_credits rather than injecting capped oh_eitc, in line with the new uncapped behaviour).
• New tests/policy/reform/oh_refundable_eitc.yaml adds four regression tests: full refundable payout at zero liability (asserting oh_income_tax goes negative, i.e. the refund is actually paid); confirmation that oh_joint_filing_credit still applies under the reform (proves the other credits are no longer discarded); no double-counting when liability partially absorbs other credits; and the filtered ordered walk capping multiple non-EITC credits at remaining liability.
• All 234 Ohio baseline tests still pass.
• The test_non_refundable_credit_downstream_consumers.py invariant still passes (the reform's reference to oh_eitc in the filter is unchanged).

Why this slipped through

Same test-shielding pattern as MO/UT — see the OH stanza of the bug-pattern analysis in the parent issue. The existing contrib tests pinned oh_eitc: 500 as input and only checked oh_refundable_eitc / oh_non_refundable_eitc / oh_non_refundable_credits with no other OH credits present, which dodged both broken code paths.

🤖 Generated with Claude Code

[hua7450]

Reviewed locally with a multi-validator pass — regulatory, references, code patterns, test coverage, plus a deep dataflow trace run live in a Simulation. The fix is correct. I pushed a small follow-up commit (3ba79b6) with documentation/test polish (no behavior change); details below.

Correctness verified

• Reading oh_eitc_potential yields the full refundable amount for the modeled era (2020+): Ohio's pre-2019 "50% of tax above $20,000 OH taxable income" cap was repealed by HB 62 (eff. 2019-07-03), leaving only the ordinary liability cap on the nonrefundable oh_eitc — exactly what refundability lifts. No intermediate cap is silently skipped.
• No double-counting — EITC flows through the refundable path only (oh_non_refundable_eitc → 0 and filtered out of the ordered nonrefundable walk).
• The ordered-cap rewrite preserves every other nonrefundable credit, filtering exactly oh_eitc.
• Tests are non-vacuous — each assertion fails on the pre-fix code and passes after.

Follow-up commit (polish only, 3ba79b6)

• Added an end-to-end assertion (oh_income_tax: -1_500) to the zero-liability reform test, confirming the refund actually pays out rather than just landing in the refundable bucket.
• Documented the HB 62 cap repeal in the reform docstring, explaining why reading oh_eitc_potential is safe (ORC §5747.71).
• Corrected the "six other credits" wording in the code comment, test header, and PR description — the 2021–2022 ordered list also carries oh_adoption_credit (repealed 2023), so it's "every other entry in the ordered list." The formula was already correct (it filters the live period-specific list).
• Swapped the non-OH negative test from CA to TX.

Reform tests (4) and contrib tests (4) pass locally; leaving CI to confirm.

[DTrim99]

PR Review — #8657: Fix OH refundable EITC reform zero-refund and lost-credit bugs

The Ohio member of a three-PR family of state refundable-EITC reform fixes (alongside #8642 MO and #8645 UT). Reviewed read-only for correctness, code patterns, and test coverage. Both claimed bugs are genuine; both fixes verified correct, single-counted end-to-end; CI fully green (26/26).

🔴 Critical (Must Fix)

None.

🟡 Should Address

None blocking.

🟢 Suggestions

1. Test entry-point inconsistency (minor): the new regression file tests/policy/reform/oh_refundable_eitc.yaml targets the variable directly while the existing contrib file targets the package re-export. Both resolve and pass; not worth changing.
2. Currency margin: no test carries absolute_error_margin: 0.01. All expected values are exact integers (30% of round federal EITC → 1500, 600, 900), so exact match is fine here; optional.

Findings detail

• Bug 1 — zero refund at zero liability (genuine, fixed): old oh_refundable_eitc.formula returned oh_eitc, which is applied_state_non_refundable_credit(...) — capped at OH liability → 0 at zero liability. Fix reads oh_eitc_potential (eitc × 0.3, rate constant since 2020 per credits/eitc/rate.yaml), the uncapped amount. Confirmed oh_eitc_potential.py has no liability cap. Matches the MO/UT *_potential pattern.
• Bug 2 — other OH non-refundable credits discarded (genuine, fixed): old oh_non_refundable_credits.formula returned only oh_non_refundable_eitc (= 0 under reform), silently zeroing CDCC, senior, retirement, non-public-school, exemption, joint-filing (+ pre-2023 adoption). Fix now mirrors baseline oh_non_refundable_credits exactly (ordered_capped_state_non_refundable_credits over the non_refundable param) with oh_eitc filtered out. Parameter resolved to its list before the helper iterates — no bare-string add() crash.
• No double-counting (verified end-to-end): oh_income_tax_before_refundable_credits subtracts oh_non_refundable_credits; oh_income_tax does adds=[…before_refundable], subtracts=[oh_refundable_credits]. Under the reform, oh_eitc is removed from the non-refundable bucket (does not reduce liability) and appears only via oh_refundable_eitc → oh_refundable_credits (subtracted once). EITC counted exactly once.
• Refundability treatment internally consistent: baseline OH EITC is non-refundable; the pre-2019 "50%-of-tax over $20k" limitation was repealed by HB 62 (docstring cites ORC §5747.71), so oh_eitc_potential is the correct full amount for the modeled 2020+ era.
• (b) N/A: these reform variables define formula with no inherited adds/subtracts, so no mixed-mode conflict exists — nothing to clear.
• Allowlist unchanged & correct: the fix keeps a string-literal "oh_eitc" (in the credit != "oh_eitc" filter), so the existing test_non_refundable_credit_downstream_consumers.py allowlist entry still matches — no edit needed, invariant check passes.

Validation Summary

Item	Result
Bug 1 (zero-refund) real & fixed via oh_eitc_potential	Confirmed
Bug 2 (lost credits) real & fixed via filtered ordered walk	Confirmed
No add() bare-string crash (resolves list first)	Confirmed
adds/subtracts vs formula conflict	N/A (none present)
Single-counting (EITC counted exactly once)	Confirmed via income-tax chain
Reform mirrors baseline oh_non_refundable_credits logic	Confirmed
OH EITC rate 30% / test arithmetic (1500, 600, 900)	Confirmed
Boundary tests (zero-EITC, partial-liability, multi-credit, out-of-state)	Present
Joint-filing-credit-preserved regression test	Present
End-to-end negative oh_income_tax payout test	Present (-1_500)
Changelog fragment (.fixed.md, top-level)	Present
CI (26 checks)	All SUCCESS

Review Severity: APPROVE

Both bugs are genuine, both fixes correct and consistent with the baseline and the MO/UT siblings, single-counting verified end-to-end, thorough test coverage, CI fully green. No required changes.