Add South Carolina fully refundable EITC contrib reform

[DTrim99]

What this does

Adds a Child Poverty Impact Dashboard contrib reform that converts South Carolina's nonrefundable EITC to fully refundable for all filers, activated by gov.contrib.states.sc.child_poverty_impact_dashboard.eitc.in_effect. This is the SC half of #8682.

SC's EITC is 125% of the federal EITC, capped per filer at $200 from 2026, and is nonrefundable by default (applied only up to remaining state income tax liability), so zero-liability filers receive nothing today.

Approach

Mirrors the corrected Utah/Missouri/Ohio refundability reforms (#8645 / #8642 / #8657) — same principles, no reintroduced bugs:

• Pay the potential, not the capped credit. sc_fully_refundable_eitc returns sc_eitc_potential (uncapped at liability), so the full credit is paid as a refund. sc_eitc is capped at liability and zeroes out for exactly the low-liability filers refundability helps.
• Rebuild the non-refundable bucket correctly. sc_non_refundable_credits re-runs ordered_capped_state_non_refundable_credits over the ordered list with sc_eitc filtered out — no add("<param-path string>") bug, and no sum − sc_eitc overstatement when the bucket binds.
• Clear inherited adds before adding a formula. sc_refundable_credits sets adds = None; subtracts = None (the baseline computes via adds) and returns the standard refundable credits plus sc_fully_refundable_eitc.

The statutory per-filer cap (gov.states.sc.tax.income.credits.eitc.max) is preserved and remains separately adjustable — the dashboard exposes it as its own lever.

New files

• policyengine_us/reforms/states/sc/child_poverty_eitc/sc_fully_refundable_eitc_reform.py (+ __init__.py)
• policyengine_us/parameters/gov/contrib/states/sc/child_poverty_impact_dashboard/eitc/in_effect.yaml
• policyengine_us/tests/policy/reform/sc_fully_refundable_eitc.yaml
• registration in policyengine_us/reforms/reforms.py; changelog fragment

Verification

• 4 YAML cases pass: full potential paid as refundable at zero liability; no double-count when liability is present; the $200 (2026) cap is respected; no effect outside SC.
• End-to-end: setting …child_poverty_impact_dashboard.eitc.in_effect = true on a 2026 SC filer with a federal EITC of $100 and zero SC liability yields sc_fully_refundable_eitc = $125, sc_refundable_credits = $125, and +$125 household net income (baseline pays $0).

Part of #8682.

[PavelMakarchuk]

Program Review — Add South Carolina fully refundable EITC contrib reform

Scope: Contrib reform (Child Poverty Impact Dashboard), SC EITC, 2026 · CI: ✅ 25/25 passing

The implementation is correct and faithfully mirrors the corrected OH/MO refundability pattern (filtered ordered-cap re-walk; adds=None cleared before the formula; no double-counting), avoiding the older UT baseline − ut_eitc bug. The statute is verified and references are sound. Requesting changes to tighten test coverage and fix misleading cap terminology before merge.

Requested changes

1. Clarify "per-filer" → "per-tax-unit" terminology. The docstring, PR description, and test header describe a "per-filer $200 cap," but the implementation (sc_eitc_potential.py, max.yaml) applies one flat $200 clamp per tax unit — a married-joint couple gets $200, not $400. The statute (SC Code 12-6-3632 / H.4216 §7) says "not to exceed two hundred dollars" with no per-filer multiplier, so the model is correct — but the wording is misleading and should say "per return / per tax unit." (This cap is pre-existing baseline behavior, not introduced here.)

2. Add missing edge-case tests. The 4 YAML cases all use 2026 + the always-on reforms: bypass and pin liability to 0 or a fully-absorbing value. Please add:

   • Partial liability — liability between 0 and the credit, confirming the full potential is paid (not the residual).
   • household_net_income assertion — every current case checks credit variables only; none confirms the refund flows end-to-end to net income. This is the highest-value missing test.
   • Married (MFJ) — pin the per-tax-unit cap behavior for 2 filers.
   • in_effect = false gating — the 5-year activation look-ahead is currently untested (all cases use the bypass).
   • Lower priority: a pre-2026 (infinite-cap) case and a case with another SC non-refundable credit present, to exercise the re-walk the docstring says prevents overstatement.

Suggestions (non-blocking)

• Inline aggregate-override variables omit reference fields (consistent with baseline SC aggregates / UT sibling — cosmetic).
• for i in range(5) → for _ in range(5) (unused loop var).

Validation Summary

Check	Result
Regulatory Accuracy	✅ Statute verified (125%, nonrefundable, $200 from 2026); reform logic correct, no double-counting
Reference Quality	✅ Toggle param matches 12 sibling conventions; statutory claims trace to H.4216 §7
Code Patterns	✅ Mirrors corrected OH/MO pattern; adds=None cleared; no hard-coding; changelog + registration present
Test Coverage	⚠️ Gaps: partial liability, net-income assertion, MFJ, gating-off
CI Status	✅ Passing (25/25)

To auto-fix: /fix-pr 8683

🤖 Generated with Claude Code

[DTrim99]

Fixes Applied

Addressed the review (no critical issues; CI was green throughout).

Should-address

• Clarified the cap is per tax unit, not per filer — corrected misleading "per-filer" wording in the reform docstring/comments and the test file. The model was already correct: the $200 refundable cap is a single flat clamp per return (MFJ also $200, not $400).
• Added edge-case tests — end-to-end refund flowing through sc_income_tax/net income, partial-liability split, MFJ cap ($200), and the in_effect=false no-op default (baseline non-refundable behavior).

Suggestions

• Added a reference (SC H.4216 §7 / SC Code 12-6-3632) to in_effect.yaml.
• for i in range(5) → for _ in range(5) (unused index).

Verification

• Reform test: 8/8 pass; regression (baseline sc_eitc + h3492 contrib): 12/12 pass.
• No expected-value corrections were needed.
• ruff format + ruff check: clean.

[DTrim99]

CI note: "Full Suite - Rest" failure was infrastructure, not a test failure

The failed Full Suite - Rest job did not contain any test failures. The logs show:

• policyengine_us/tests/policy/reform/sc_fully_refundable_eitc.yaml ........ — all 8 cases (including the new edge cases) passed
• 392 passed, 5 skipped and 67 passed across the shard's batches
• The job then ended with ##[error]The runner has received a shutdown signal ... The operation was canceled.

That is a runner shutdown / cancellation (CI infrastructure), unrelated to this PR's changes. The job is now re-running and should go green.

[PavelMakarchuk]

Re-reviewed after "Address review: clarify per-tax-unit cap wording, add edge-case tests" — all requested changes addressed:

• Wording: docstring + comments now say "per return / per tax unit" (statute caps the credit at $200 per return, not per filer).
• Edge-case tests added: end-to-end net SC tax refund (sc_income_tax: -125 at zero liability), partial liability (capped $200 credit leaves $100 owed on $300 liability), MFJ capped at $200 per tax unit (not $400), and in_effect=false gating no-op.

Verified all 8 YAML cases pass locally. The failing "Full Suite - Rest" check is a transient runner shutdown (infra), not a test failure — recommend re-running. Approving. 🤖