Fix UT fully refundable EITC reform crash and zero-refund bug

[PavelMakarchuk]

Fixes #8644

Problem

Triggering the bundled Utah fully-refundable-EITC contrib reform (gov.contrib.states.ut.child_poverty_impact_dashboard.eitc.in_effect) crashed the simulation. After working around the crash, the reform also failed to actually pay the EITC as refundable for the low-liability filers refundability is meant to help — same dual-bug profile as the Missouri reform fixed in #8642.

Three bugs fixed in ut_fully_refundable_eitc_reform.py

1. add() received a parameter path string (crash). ut_non_refundable_credits.formula passed the literal string "gov.states.ut.tax.income.credits.non_refundable" to add(), which iterates it character by character — get_variable("g") returns None → AttributeError: 'NoneType' object has no attribute 'entity'. Now resolves the parameter to its list of variable names first.

2. Mixed computation modes (crash on core ≥3.26.8). ut_refundable_credits redeclared a formula but inherited adds from the baseline variable, tripping the engine's strict mode check (ValueError: ... mixes computation modes). Now sets adds = None and subtracts = None explicitly.

3. Refundable credit capped at tax liability (silent zero-refund). ut_fully_refundable_eitc returned ut_eitc — the applied credit capped at UT tax liability — so a zero-liability filer received $0 even though refundability should pay the full amount. Now uses ut_eitc_potential (uncapped at liability; the W-2 wages cap mandated by Utah Code § 59-10-1044 still applies as designed). Same functional fix the MO PR applied via mo_wftc_potential.

Verification

Reproduced the issue's exact script — runs without raising. Verified the reform's effect against baseline using the YAML regression test below: a Utah filer with eitc = 5_000, employment_income = 30_000, and ut_income_tax_before_non_refundable_credits = 0 (i.e. zero state liability) now receives the full ut_eitc_potential = 1_000 as ut_refundable_credits, where the buggy reform returned $0.

Tests

• New tests/policy/reform/ut_fully_refundable_eitc.yaml: verifies the full potential UT EITC is paid as refundable at zero liability, the nonrefundable portion is zeroed, and there's no double-counting when liability already absorbs the credit.
• Existing contrib tests at tests/policy/contrib/states/ut/child_poverty_impact_dashboard/eitc/ut_fully_refundable_eitc.yaml updated to drive from federal eitc (rather than injecting the capped ut_eitc), in line with the new uncapped behaviour.
• All 180 Utah baseline tests still pass.
• The test_non_refundable_credit_downstream_consumers.py invariant still passes (the reform's reference to ut_eitc in ut_non_refundable_credits is unchanged).

🤖 Generated with Claude Code

[DTrim99]

PR Review — #8645: Fix UT fully refundable EITC reform crash and zero-refund bug

The Utah member of a three-PR family of state refundable-EITC reform fixes (alongside #8642 MO and #8657 OH). Reviewed read-only for correctness, code patterns, test coverage, and the one red CI check. All three claimed fixes verified correct.

🔴 Critical (Must Fix)

None. The fixes are correct. (The red CI check is an unrelated runner timeout — see CI diagnosis.)

🟡 Should Address

1. Re-run the failed Full Suite - Contrib (states-shard-2) — it is not a real failure. The shard died with exit code 143 (SIGTERM / runner shutdown) during unrelated Rhode Island tests (ri/ctc_reform_test.yaml, ri/exemption_reform_test.yaml) in Batch 15 — a job timeout, not a UT failure. The UT reform tests themselves passed earlier in the same job (ut_fully_refundable_eitc.yaml …, batch reported 35 passed). It's on the latest commit (a7b5961), so a maintainer re-run is warranted before merge; expect green.
2. Add absolute_error_margin: 0.01 to the currency assertions in both tests/policy/reform/ut_fully_refundable_eitc.yaml and the contrib test. Expected values are exact integers (0/300/400/500/1000) so tests pass without it, but the one-cent margin is convention and guards float brittleness.

🟢 Suggestions

1. Consider asserting ut_non_refundable_credits directly in one zero-liability case to lock the "EITC removed from the bucket, counted once" behavior at the aggregator level (currently inferred via ut_income_tax: 0).

Findings detail

• Bug (a) — string→add() crash: old ut_non_refundable_credits.formula passed the bare param path to add(), which iterates it char-by-char → get_variable("g") → AttributeError. Fix resolves the non_refundable parameter to its list first, then feeds the ordered-cap helper. Correct.
• Bug (b) — adds = None/subtracts = None: baseline ut_refundable_credits uses adds = "…credits.refundable"; the reform replaces it with a formula, so clearing both is the right defensive pattern. (The "mixes computation modes" ValueError guard is core ≥3.26.8; still correct and forward-compatible.)
• Bug (c) — capped→uncapped, no double-count: ut_eitc is liability-capped → $0 at zero liability; fix returns ut_eitc_potential (uncapped, still applies the §59-10-1044 W-2-wages cap). The reform's new ut_non_refundable_credits walks the same ordered list as baseline minus ut_eitc, so EITC is counted exactly once — refundable side only. The switch from a naive sum − ut_eitc to the ordered walk (2nd commit) is correct and necessary so the ordered cap frees EITC's slot for later credits. "Fully refundable for all filers incl. childless" semantics match intent.
• Code patterns: period usage correct; valid import of ordered_capped_state_non_refundable_credits; no hard-coding (rate 0.2 from params; 5000×0.2=1000, 2500×0.2=500 verified). No code-health allowlist change in the diff.

CI diagnosis

Check	Full Suite - Contrib (states-shard-2) (run 27589687418, job 81568058302, head a7b5961)
Root cause	Job timeout / runner shutdown (exit 143, SIGTERM) during unrelated RI tests in Batch 15; make: *** Terminated, "runner received a shutdown signal"
UT tests in that job	PASSED (ut_fully_refundable_eitc.yaml, 35 passed); shards 1 / other-1/2 / congress all green
Verdict	Infrastructure timeout, not caused by this PR — re-run to clear

Validation Summary

Check	Result
Bug (a) add()-string crash fix	PASS
Bug (b) adds/subtracts cleared	PASS
Bug (c) uncapped ut_eitc_potential, counted once	PASS — no double-count
§59-10-1044 W-2 wages cap preserved	PASS
"Fully refundable" semantics	PASS — matches intent
Period / naming / no hard-coding	PASS
Test coverage of all 3 bugs	PASS (boundary cap, partial-liability, SE-only)
absolute_error_margin: 0.01 on currency	MISSING (cosmetic)
Changelog fragment (.fixed.md, top-level)	PASS
CI shard-2 failure	Infra timeout (exit 143), not UT-related — re-run

Review Severity: APPROVE

The three fixes are correct, no double-counting, and well-tested. The red check is a runner timeout on unrelated RI tests, not a defect in this PR — re-run to clear. Adding the currency margin is minor optional polish.