Remove Additional Medicare Tax from fiitax to match NBER taxsimtest

[PavelMakarchuk]

Summary

Stop adding additional_medicare_tax to the fiitax output. NBER TAXSIM-35 (taxsimtest) reports the Additional Medicare Tax (Form 8959) in a separate addmed column rather than rolling it into regular income tax — matching Form 1040 Line 23 / Schedule 2 Line 11. PE's income_tax (which already includes NIIT) is the correct match for TAXSIM's fiitax.

Why

This was the dominant remaining federal mismatch on the eCPS comparison after #961 / #962. On the n=2000 TY 2025 sample:

• 95% of the residual federal $$ off ($412K of $432K) was traceable to records above the AddMed threshold.
• The correlation between |fed_diff| and 0.9% × (medicare_wages − threshold) was 0.997 — almost a perfect fit, indicating a single source.
• Every top-12 federal-miss record sat right on 3.8% × $... = $... for AddMed.

Verified by probing the live binary used by our compare flow:

$ cat single_500k.csv | resources/taxsimtest/taxsimtest-osx.exe
fiitax=140264.75
addmed=2355.75
v28=140264.75    # income tax before NIIT / AddMed

fiitax here is just v28 (income tax only). AddMed comes out separately. PE was reporting fiitax = income_tax + additional_medicare_tax, hence the consistent ~AddMed-sized overshoot.

Statute / form anchors

• IRC § 3101(b)(2): 0.9% Additional Medicare Tax on wages above $200K single / $250K MFJ / $125K MFS.
• IRC § 1401(b)(2): same 0.9% applied to net SE earnings above same thresholds.
• Form 8959: AddMed computation worksheet.
• Form 1040 Line 23 / Schedule 2 Line 11: AddMed reported as "Other Taxes", separate from Line 16 regular income tax.

Changes

1. policyengine_runner.py: drop the AddMed add-on from both the fiitax output assembly and the frate marginal-rate calculation (which had to mirror the fiitax definition, including a comment that referenced the now-removed addition).
2. tests/test_addmed_excluded_from_fiitax.py: two pinning tests using NBER taxsimtest smoke values as targets:
   • Single $500K wages 2024 → fiitax ≈ $140,265 ±$200
   • MFJ $1M wages 2024 → fiitax ≈ $285,321 ±$300
     Both targets are tight enough that adding back AddMed (~$2,700 / ~$6,000) would break them cleanly.
3. changelog.d/remove-addmed-from-fiitax.fixed.md.

eCPS n=2000 TY 2025 (non-S-corp, --disable-salt --assume-w2-wages)

Metric	Before	This PR	Δ
Federal $$ off	$432,768	$40,730	−$392K (−90.6%)
Federal $15 match	93.6%	97.3%	+3.7pp
Federal 1%-AGI match	99.9%	99.9%	(already saturated)
State $$ off	$121,738	$121,738	(unchanged, expected)
>$1M bucket $$ off	$325,353	$17,971	−$307K (−94%)

Per-bucket federal $$ off

Bucket	n	Before	This PR
≤$25K	560	$5	$5
$25–50K	390	$26	$26
$50–100K	374	$21,645	$12,400
$100–250K	293	$3,974	$1,918
$250K–1M	74	$81,765	$8,410
>$1M	36	$325,353	$17,971

Test plan

• Both new tests pass (test_single_500k_fiitax_excludes_addmed, test_mfj_1m_fiitax_excludes_addmed)
• Full pytest tests/ passes
• eCPS n=2000 TY 2025 confirms 90.6% federal $$ reduction with no per-bucket regression
• Existing test_otherprop_niit.py::test_otherprop_does_not_trigger_qbid continues to hold (the otherprop-NIIT target $353,188 for the synthetic probe matches taxsimtest because the synthetic record has no wages → no AddMed component)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
policyengine-taxsim	Ready	Preview, Comment	Jun 4, 2026 9:01pm