Correct CRFB expanded surtax premium addbacks; isolate salary-reduction reading

[MaxGhenis]

The CRFB expanded surtax base currently adds employee after-tax health premiums, which are financed from income already included in AGI, and omits the separate employer-paid premium input. This draft replaces that addback with employer-paid premiums and employee pretax payroll premiums. It holds only the CRFB parameter change and its tests, plus the required changelog fragment; the baseline input/consumer work stays in #10046.

Premium addback correction and methodology ruling

Under either reading below, adding health_insurance_premiums double-counts after-tax employee payments already financed from AGI. Replacing that source with employer_sponsored_insurance_premiums fixes the incorrect payer and restores the employer share. The expanded base also includes qualified employee salary-reduction premiums under Max's d1295 ruling on 2026-10-11.

CRFB footnote 2 specifies AGI plus employer-provided health insurance without distinguishing the employer share from employee salary reductions. Revenue Ruling 2002-3 classifies qualified salary-reduction coverage as employer-provided for federal tax purposes. Applying that classification to this contributed proposal is an inference, not a uniquely specified CRFB instruction. Max ruled on 2026-10-11 (d1295) that the expanded base includes qualified employee salary-reduction premiums together with the employer share; the code follows that ruling. Both readings remain below for the record.

For a single household in 2026 with $150,000 gross wages, $6,000 employee pretax premiums, $4,000 employee after-tax premiums, $10,000 employer-paid premiums and no other income/addbacks, AGI is $144,000:

Reading	Expanded base	Surtax under the modeled 1% rate above $100,000
Employer share only	$144,000 + $10,000 = $154,000	$540
Employer share plus qualified employee salary reductions	$144,000 + $10,000 + $6,000 = $160,000	$600
Prior expanded-base source (employee after-tax premiums)	$144,000 + $4,000 = $148,000	$480

The code implements the second reading. It does not add the $4,000 after-tax premiums. Moving all $10,000 of employee premiums to after-tax payments yields $150,000 AGI + $10,000 employer premiums = $160,000 under both readings; the mixed-payment difference is the $6,000 salary-reduction addback.

The reform implementation is policyengine_us/reforms/crfb/agi_surtax.py, with contributed parameters under policyengine_us/parameters/gov/contrib/crfb/. This is a policy proposal rather than enacted law.

axiom: n/a: contributed reform, not enacted law

Validation

Four new YAML cases cover the mixed household, all-after-tax household, after-tax double-count prevention and isolated pretax restoration. All preserved cases remain. The isolated pretax case also asserts $15,670 of income tax before credits ($15,370 regular tax plus the $300 surtax).

The independent property-test oracle now adds employer-paid and qualified salary-reduction premiums, excluding after-tax employee premiums. Both new premium inputs are generated for filers and dependents, preserving dependent-exclusion checks. A generated payment-method check holds supplied AGI fixed: moving the same amount between employer-paid and salary-reduction premiums leaves the surtax unchanged, while paying it after tax reduces the expanded base by that amount. Every draw covers a below-threshold base, crossing the initial threshold, the 1% bracket and crossing the $1 million threshold.

Reproduced the review counterexample before changing the tests: a 2026 single filer with supplied AGI $150,000, $4,000 after-tax premiums and a zero-income dependent produced $500, while the stale oracle required $540. The preserved regression now passes with the model and corrected oracle both requiring $500.

Merged main at 4b3dfb1433806102d9b716b577a4609eb818d285 before verification. Each changed test file ran alone, sequentially: test_dependent_payroll_taxes_ctc_and_surtax_invariants.py — 5 passed (3 warnings); agi_surtax.yaml — 18 passed (1 warning). make format passed before both commits. The PR remains a draft as requested; GitHub's PR workflow skips draft validation jobs.

Hub microsimulation impact

Pending hub comparison against current main. No local microsimulation was run. The employer-share replacement can affect reform results even if the dataset's pretax input is absent. Record numerical impacts, years, dataset revision and US/core heads here before landing.