[MaxGhenis]

With the owners' draw reform on, the self-employed health insurance earned income limit now counts in Form 7206 line 5 the draw profits whose SECA enters the deductible SE tax, so the plan business's share of that tax no longer absorbs the draw's SECA.

The interaction

self_employed_health_insurance_earned_income_limit (#10109) follows Form 7206 lines 4–13:

• line 4 holds the Schedule C profit (total_self_employment_income);
• line 5 sums the net profit of every business subject to SE tax (adding farm income and partnership SE earnings);
• line 6 takes the plan business's share of line 5;
• lines 7–8 multiply self_employment_tax_ald_person by that share.

The owners' draw reform (#10483) overrides self_employment_tax_ald_person, so with the reform on (and half_seca_deduction at its default, true), line 7 includes the draw's SECA. Line 5 left out the draw's profits, so the Schedule C business took too large a share of that tax and the limit fell.

Hand check

Single filer, 2027, synthetic $200,000 caps:

• Inputs: $40,000 Schedule C; S-corporation profit $64,970.22, which gives draw net earnings 64,970.22 × 0.9235 = $59,999.998; $8,000 of premiums.
• SE tax: Schedule C 40,000 × 0.9235 × 15.3% = 5,651.82, plus draw OASDI 59,999.998 × 12.4% = 7,440.00 (HI on the draw defaults off). Deductible half: 6,545.91.
• Main: line 5 = 40,000, share 1, limit 40,000 − 6,545.91 = 33,454.09.
• This PR: line 5 = 40,000 + 64,970.22 = 104,970.22, share 0.3810605, line 7 = 2,494.39, limit 37,505.61.
• The $8,000 deduction doesn't change, because both limits exceed the premiums. A second case with $36,000 of premiums moves the deduction from $33,454.09 to $36,000.

The change

policyengine_us/reforms/ssa/owners_draw.py installs an override of the limit variable:

• With the reform off for the year, it returns the baseline formula.
• With it on, it repeats the baseline's lines 4–13 and adds where(half_seca_deduction, owners_draw_niit_exclusion, 0) to line 5.

The reform computes owners_draw_niit_exclusion, its "gross owners' profits subject to SECA", as owners_draw_oasdi_base divided by the net-earnings factor (q by default) with HI off, and as the whole owners_draw_gross with include_hi on. Line 5 counts only these, and only when half_seca_deduction puts their SECA into self_employment_tax_ald_person (line 7). An earlier head of this PR added the whole owners_draw_gross, which raised the limit above main in three setups where line 7 holds none or only part of the draw's SECA:

2027, $40,000 Schedule C, $64,970.22 S-corp profit, $200,000 caps	Line 7	Main	Whole gross draw	This PR
half_seca_deduction: false	2,825.91	37,174.09	38,923.16	37,174.09
Wages $300,000, HI off: the draw bears no SECA	535.63	39,464.37	39,795.89	39,464.37
Wages $150,000, HI off: draw OASDI base 13,060 of 60,000 (gross 14,141.85)	3,635.63	36,364.37	38,614.61	37,314.00

Form 7206 line 5 counts the net profit of every business subject to SE tax, and a real business above the wage base counts all of its profit. The draw becomes SE income only through this reform, so we chose to count in line 5 the part of it whose SECA line 7 carries. Line 6 then splits the deductible tax at matching rates per dollar of profit: with HI on and no OASDI room, the limit equals line 4 less Schedule C's own deductible tax (39,464.37).

We don't change the baseline variable file.

Line 5 sums net profits before the 0.9235 net-earnings factor, so the override counts gross profits, not net earnings: line 4's Schedule C profit and the K-1 partnership SE amount both come before the factor. owners_draw_niit_exclusion holds a gross profit amount (the OASDI base divided back by q), and owners_draw_net_earnings holds gross × 0.9235, the Schedule SE line 4a analogue. Using net earnings would give share 0.4 and a limit of 37,381.64 in the hand check.

We added a section on this override to docs/contributed_reforms/owners_draw.md, after the cap-priority section.

Invariants (tested with Hypothesis, plus explicit examples)

1. With the reform disabled, the reform's limit equals the baseline formula on generated vectors: Schedule C, farm, partnership SE, S-corp profit, wages and pension deduction, signed where the model allows.
2. With it enabled, the limit equals a lines 4–13 oracle, max(0, line4 − ALD × line4 / line5 − pension). The oracle derives line 5's draw term from the rule rather than from the reform's own variables: with HI on, the gross draw; with HI off, min(gross × q, room) / q, where q comes from the rates and room from the cap less taxable wages and existing capped SE earnings (or no limit with room_cap_basis 1); zero when half_seca_deduction is off. The generated inputs vary the rates, include_hi, half_seca_deduction, room_cap_basis, phase_in_share and the cap, with the hand case, the three setups above, HI on with the deduction off and $300,000 of wages, HI on with $300,000 of wages, HI on with $150,000 of wages, and a half phase-in with HI off as explicit examples.
3. Against main's formula run separately on the same SE tax, the limit never falls below main's, and it equals main's whenever the draw term is zero (no draw, no deductible draw SECA, or no OASDI room with HI off).

The earlier upper bounds (line 4 less line 8, and line 4 less the pension deduction) followed from the oracle's equality, so the tests drop them.

The invariant "never exceeds line 4 less the plan business's own deductible SE tax" fails under Form 7206's pro-rata rule. In the hand check, the Schedule C business's own deductible SE tax comes to 2,825.91, which puts that bound at 37,174.09, while the limit reaches 37,505.61. Line 6 allocates the total deductible tax by profit share. The draw bears 12.4% × 0.9235 per dollar of profit, against Schedule C's 15.3% × 0.9235, so its profit share exceeds its tax share. The tests use the line 6 share instead.

Readers of the deductible SE tax or SE profit shares

Reader	What it does	Finding
self_employed_health_insurance_earned_income_limit	Form 7206 line 6 share of SE tax ALD	Fixed here
qualified_business_income (QBID)	Prorates self_employment_tax_ald_person (via qbi.deduction_definition) across SSTB and non-SSTB QBI; qbi.income_definition includes partnership_s_corp_income	Consistent: the pool that bears the draw's SECA deduction includes its source income
self_employed_pension_contribution_ald_person (SEP/pension)	Caps contributions at Schedule C profit; doesn't read SE tax	No inconsistency of this kind (whether the draw supports plan contributions raises a separate question)
eitc_earned_income	Wages + Schedule C + farm + partnership SE − SE tax ALD	Same kind: subtracts the draw's SECA deduction from earned income that excludes the draw
ca_eitc_earned_income	Same structure as federal	Same kind
ira_compensation	SE earnings − SE tax ALD − pension	Same kind
standard_deduction_worksheet_earned_income_person	Earned-income sources − SE tax ALD	Same kind (dependents' standard deduction)
nd_mpc	Qualified income (wages, Schedule C, pensions) − SE tax ALD	Same kind
mn_marriage_credit	Income sources (wages, Schedule C, SS, pensions) − SE tax ALD	Same kind
sc_gross_earned_income (two-wage-earner credit)	earned_income (wages, Schedule C) − subtractions including SE tax ALD	Same kind
self_employment_tax_ald, the ALD lists, dependent_taxable_ss_magi, student-loan MAGI, MO AGI, IA adjustments, MI/MA/AL lists	Deduction in AGI-type totals	Consistent: an above-the-line deduction of tax actually owed
CTC refundable social_security.add	Sums payroll/SE taxes paid	Consistent

Other overrides in the reform:

• self_employment_social_security_tax and self_employment_medicare_tax add the draw's own taxes.
• ss_aime adds credited draw earnings.

Neither applies an SE profit share. We leave the earned-income readers marked "same kind" for a follow-up. Treating the draw's net earnings as SE earnings under those definitions takes a larger policy decision than this fix.

Tests

• policyengine_us/tests/policy/contrib/ssa/owners_draw_self_employed_health_insurance_earned_income_limit.yaml: 11 cases. They are the hand check, premiums above the old limit, no draw, no Schedule C, the three setups in the table above, and four more:

  • HI on with nondeductible draw SECA and $300,000 of wages: limit 39,464.37 (line 5 holds 40,000; counting the draw would give 39,795.89).
  • HI on with no OASDI room ($300,000 of wages): owners_draw_niit_exclusion 64,970.22, line 7 1,405.63, limit 39,464.37 (main 38,594.37).
  • HI on with 13,060 of OASDI room ($150,000 of wages): owners_draw_niit_exclusion 64,970.22, line 7 1,716.92, limit 38,283.08 (counting only the OASDI-taxed 14,141.85 of draw would give 36,671.24; main 35,494.37).
  • phase_in_share 0.5 with HI off: gross draw 32,485.11, OASDI base 30,000, line 7 4,685.91, limit 37,414.14 (main 35,314.09; the whole S-corp profit in line 5 would give 38,214.38).

• policyengine_us/tests/reforms/test_owners_draw.py:

  • The two Hypothesis properties above. The generated SEHI test runs an explicit @example for each YAML case with both a draw and a Schedule C business. Those cases include three with HI on (the deduction off, no OASDI room, and some OASDI room) and a half phase-in with HI off.
  • BASELINE_VARIABLES includes the baseline limit, so the existing disabled-equality tests, including the generated real-Simulation one, also compare it.
  • The pure harness stores integer money inputs as floats, as Core does.
  • The whole file: 38 passed in 188 s at bed691b7ff, run in its own process.

• policyengine_us/tests/code_health/test_source_conflicts.py: 103 passed.

• Mutants on the line 5 draw term at owners_draw.py:304-305, each in its own process, with the code restored after each:

  Mutant	YAML (10 cases)	test_owners_draw.py -k "sehi or disabled" (5 tests)
  M6: gate on half_seca_deduction | include_hi	1 failed: "With HI on, nondeductible draw SECA still keeps the draw out of line 5" (39,795.90 vs 39,464.37)	test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 failed
  Whole s_corp_income in place of the draw term	3 failed: above the OASDI cap with HI off (39,795.90), partly within the cap (38,614.61), half phase-in (38,214.39)	same test failed
  Gate on include_hi in place of half_seca_deduction	5 failed: the hand check, premiums above the old limit, partly within the cap, HI on with nondeductible SECA, half phase-in	same test failed

  The tests killed the round-1 mutants (ignore the gate, whole owners_draw_gross) at an earlier head, and the code they mutate hasn't changed.

Overlap

• Let farm, partnership and S corporation plans set the self-employed health insurance limit #10120 and Let farm and partnership plans take the self-employed plan deduction #10131 (drafts) edit the baseline self_employed_health_insurance_earned_income_limit formula. This PR copies that formula into the reform's override, so whichever lands second must carry the change into the copy.
• Add annual Social Security earnings histories and per-year reforms #10560 rewrites policyengine_us/reforms/ssa/owners_draw.py and edits docs/contributed_reforms/owners_draw.md.

impact: none at baseline. create_structural_reforms_from_parameters calls create_owners_draw_reform (reforms.py:372), which returns the reform only when some gov.contrib.ssa.owners_draw.in_effect value is true and otherwise None, which reforms.py:754 filters out; in_effect defaults to false.

axiom: n/a: contributed reform

CI: Pull request run 41537180716 at bed691b7ff completed with conclusion success (35/35 jobs succeeded; Documentation run 41537180467 also succeeded).

🤖 Generated with Claude Code

[MaxGhenis]

Ready for hub review

Head: 30fd2d7f6ac30f225b4bcf52c8a3990d1955d47d (branched from main 469ad12cb9, which includes #10483 and #10109).

What the item asked and how it's resolved

• Interaction verified before changing anything. With the reform on, line 7 uses the reform's self_employment_tax_ald_person, which includes the draw's SECA when half_seca_deduction is true (the default). Line 5 didn't include the draw. Hand check (also in the body): main gives 33,454.09; Form 7206 logic with the draw in line 5 gives 37,505.61. The $8,000 deduction is unchanged, because the limit isn't binding at $8,000; a $36,000-premium case moves the deduction from 33,454.09 to 36,000.
• Fix: a reform override of the limit (policyengine_us/reforms/ssa/owners_draw.py:290, registered at :359). When the reform is off for the year it returns the baseline formula. When on, it adds max_(0, owners_draw_gross) to line 5. The baseline variable file is untouched. Changelog: changelog.d/owners-draw-sehi-line5.fixed.md.
• Reader check: the table is in the PR body. EITC, CA EITC, IRA compensation, the dependent standard-deduction earned income, ND MPC, MN marriage credit and SC two-wage-earner earned income all subtract the deductible SE tax, which now includes the draw's SECA, from an earnings measure that leaves out the draw. QBID, the ALD and AGI lists, and the CTC payroll-tax add-back are consistent. The pension deduction doesn't read SE tax.

Judgment calls

1. Line 5 uses owners_draw_gross, not owners_draw_net_earnings. Form 7206 line 5 sums net profits before the 0.9235 factor, like line 4's Schedule C profit and the K-1 partnership SE amount. The draw's matching figure is its gross; net earnings is gross × 0.9235. In the hand case, net earnings would give share 0.4 and a limit of 37,381.64. Both differ from main by more than 3,900. The YAML pins gross.
2. The brief's second invariant is false as first worded. Read as "line 4 less the plan business's own deductible SE tax", the bound is 40,000 − 2,825.91 = 37,174.09, but Form 7206's pro-rata result is 37,505.61. Line 6 allocates by profit share, and the draw bears 12.4% × 0.9235 per dollar of profit against Schedule C's 15.3% × 0.9235. The Hypothesis test instead asserts four things:
   • the limit never exceeds line 4 less the line 6 share of deductible SE tax;
   • it is never below main's draw-free result;
   • it is never above line 4 less the pension deduction;
   • it equals an independent lines 4–13 oracle.
3. Test harness change. The pure harness in test_owners_draw.py now converts integer inputs to float, as Core stores them. Without that, np.zeros_like produced an integer output array and the limit's division raised on an integer probe in test_generated_disabled_reform_matches_public_baseline.

Tests (one file per process)

File	Result
policyengine_us/tests/policy/contrib/ssa/owners_draw_self_employed_health_insurance_earned_income_limit.yaml	4 passed: hand check, binding premiums, no draw (unchanged, 37,174.09), no Schedule C (0)
policyengine_us/tests/reforms/test_owners_draw.py	38 passed, 143 s. Includes test_generated_disabled_sehi_limit_equals_baseline (line 748), test_generated_sehi_limit_counts_the_draw_in_line_5 (line 759), and the existing disabled-equality tests, which now also compare the limit, including the generated real-Simulation test

Mutants (each in its own process; restored and checked with cmp)

• Draw dropped from line 5: killed by test_generated_sehi_limit_counts_the_draw_in_line_5 and by the 2 YAML cases with a draw and a Schedule C.
• Draw added twice: killed by the same tests.

CI

Draft, so CI hasn't run. No microsimulation was run; the hub confirms impact: none at baseline (reform off by default).

Proposed follow-up

• Treat owners' draw net earnings as SE earnings in earned-income definitions under the reform. EITC, CA EITC, IRA compensation, dependent standard-deduction earned income, ND MPC, MN marriage credit and SC two-wage-earner income subtract the draw's deductible SECA from earnings that leave out the draw. Decide whether the reform's draw counts as net earnings from self-employment for these definitions, and include it or exclude its SECA consistently.

---

Generated by Claude Code

[MaxGhenis]

Picking this up for the US hub. I'll post a hand-off comment here when it's ready.

---

Generated by Claude Code

[MaxGhenis]

Ready for hub review

Head: 8d96db39acbf1129f20c036d9dab0c8b17998565 (one commit on 30fd2d7f; nobody else pushed in between).

Findings

P2-1, line 5 counts draw profits whose SECA isn't in line 7. Fixed in policyengine_us/reforms/ssa/owners_draw.py:302-315. Line 5 now adds where(draw.half_seca_deduction, owners_draw_niit_exclusion, 0), where draw = parameters(period).gov.contrib.ssa.owners_draw. Before using them I read owners_draw_niit_exclusion (owners_draw.py:225-241: owners_draw_oasdi_base / owners_draw_net_earnings_factor with HI off, owners_draw_gross with HI on, 0 when disabled) and half_seca_deduction.yaml (default true). I rewrote the comment at :294-301 to describe what the code does now.

P2-2, nothing tests these setups.

• YAML: three cases in owners_draw_self_employed_health_insurance_earned_income_limit.yaml, each asserting the limit. They also assert the intermediate draw variables.
• Hypothesis: test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 (test_owners_draw.py:795; its policy strategy and examples start at :757) now varies half_seca_deduction, the cap (1 to 400,000, applied to both the payroll cap and current_law_cap), room_cap_basis, phase_in_share and include_hi, along with the rates. It has the hand case and all three witnesses as @examples.
• The oracle builds line 5's draw term from the rule, not from owners_draw_niit_exclusion. With HI on it is the gross draw. With HI off it is min(gross × q, room) / q, where q comes from the rates and room = max(0, cap − taxable_earnings_for_social_security − social_security_taxable_self_employment_income), or infinity with room_cap_basis 1. It is zero when half_seca_deduction is off.

P3 items.

• Docs: added a "Self-employed health insurance limit" section to docs/contributed_reforms/owners_draw.md.
• Bounds: I dropped the two upper bounds because they follow from the oracle's equality. I kept two checks against main's formula, which is run separately: the limit is never below main's, and it equals main's wherever the draw term is zero. That second check is wider than the old no-draw check.
• PR body: rewritten and read back. It drops the draft/no-CI text and adds the witness table and an overlap section (Let farm, partnership and S corporation plans set the self-employed health insurance limit #10120, Let farm and partnership plans take the self-employed plan deduction #10131, Add annual Social Security earnings histories and per-year reforms #10560). It keeps the impact: and axiom: lines.
  • The impact: line cites gates I read myself. reforms.py:372 calls create_owners_draw_reform, which returns None unless some in_effect value is true. reforms.py:754 filters out None. in_effect.yaml defaults to false.

Hand values

All five setups use 2027, $40,000 Schedule C, $64,970.22 S-corp profit, $200,000 caps and q = 0.9235. They match the review's table.

Case	Line 7	Main	Old head	This head
half_seca_deduction: false	2,825.91	37,174.09	38,923.16	37,174.09
Wages $300k, HI off	535.63	39,464.37	39,795.89	39,464.37
Wages $150k, HI off	3,635.63	36,364.37	38,614.61	37,314.00

How the $150k case works out:

• The draw's OASDI room is 200,000 − 150,000 − 36,940 = 13,060, so the draw's OASDI tax is 1,619.44.
• The draw term is 13,060 / 0.9235 = 14,141.85, so line 5 is 54,141.85.
• The limit is 40,000 − 3,635.63 × 40,000 / 54,141.85 = 37,313.997.

The existing hand case (37,505.61) and the four existing YAML values are unchanged.

Tests (one file per process)

• uv run pytest -q -p no:cacheprovider policyengine_us/tests/reforms/test_owners_draw.py: 38 passed in 179 s.
• policyengine_core ... test owners_draw_self_employed_health_insurance_earned_income_limit.yaml -c policyengine_us: 7 passed.
• make format and ruff check are clean.

Mutants (each applied, run and reverted)

• Gate on draw.include_hi instead of draw.half_seca_deduction: killed by test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 and by 3 of the 7 YAML cases. Those are the cases with a nonzero draw term at defaults: the hand case, premiums above the old limit, and the partial-room case.
• Ignore the gate (True): killed by the same generated test.
• The old whole-owners_draw_gross term: killed by the same generated test.

CI

Pull request run https://github.com/PolicyEngine/policyengine-us/actions/runs/41397745379 was still in_progress 60 minutes after the push. At that point every job had succeeded except Contrib tests / Full Suite - Contrib (other-shard-1), which was still running. Documentation passed.

Judgment calls

• Beyond the brief's half_seca_deduction and cap, I also added room_cap_basis and phase_in_share to the Hypothesis inputs. They are cheap, and room_cap_basis 1 exercises the infinite-room branch.
• The oracle reads the reform's owners_draw_gross and the baseline's capped SE and wage bases as upstream inputs; other tests in this file cover those variables. It derives everything after them independently.
• I reworded the changelog fragment to say only SECA-taxed, deductible draw profits count.
• I left the copied baseline formula as it is.
• Overlap is listed in the body:
  • Let farm, partnership and S corporation plans set the self-employed health insurance limit #10120 and Let farm and partnership plans take the self-employed plan deduction #10131 edit the baseline limit formula, which this PR copies.
  • Add annual Social Security earnings histories and per-year reforms #10560 edits owners_draw.py and owners_draw.md.
• I pushed the same commit to this session's own working branch as well as to owners-draw-sehi-line5.

Follow-up

Owners' draw: count the draw in earned-income readers that subtract SE tax ALD. eitc_earned_income, ira_compensation and the other readers marked "same kind" in the PR body subtract the draw's SECA deduction from earned income that excludes the draw. Deciding whether the draw is earned income there is a policy call, separate from this fix.

---

Generated by Claude Code

[MaxGhenis]

Ready for hub review
VERDICT: REQUEST_CHANGES
8d96db3
METHODOLOGY_CHOICE: yes

Independent round-2 review. The head hasn't moved past 8d96db39ac. I found no **Hub review in progress** comment from an earlier session after 2026-11-02T17:55:30Z, so this review starts from the beginning.

Summary

The code is right. I hand-computed ten setups from the Form 7206 rule before reading the code, including the two this round asks about (phase_in_share 0.5, and HI on with the OASDI room used up). The real model matches every value. Both r1 P2s are fixed in the code. The defect reproduces at main, and the baseline can't move.

The one blocking gap is in the tests. Neither test file pins the limit with HI on and half_seca_deduction false, or with phase_in_share between 0 and 1 and HI off. The generated test spends its random examples almost entirely on one combination. A plausible mutant in exactly that gap survives every test in both files: it counts the whole draw in line 5 whenever HI is on. The fix is three YAML cases (given below, already run at this head) plus matching @examples. There are also two small docs fixes.

Findings

P2-1: no test pins the limit with HI on and half_seca_deduction false, or with a partial phase-in and HI off, and a mutant survives there.

• Where: policyengine_us/tests/reforms/test_owners_draw.py:756-793 (_SEHI_POLICY and the @examples) and policyengine_us/tests/policy/contrib/ssa/owners_draw_self_employed_health_insurance_earned_income_limit.yaml. All seven YAML cases use phase_in_share: 1 and HI off.

• Coverage probe. I called the real test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 with a wrapper around _calculate that records each row with a positive draw and a positive Schedule C. That gave 44 examples and 35 such rows:

  Rows	phase_in_share	include_hi	half_seca_deduction	Draw's OASDI room
  28	0.5	true	true	none
  1	0.5	false	false	partial
  1 each	1	false	true	full / partial / none (the @examples)
  1	1	false	false	full (the @example)
  2	1	false	true	full, room_cap_basis 1

  It never reaches HI on with half_seca_deduction false, HI on with any OASDI room left, or share 0.5 with HI off and the deduction on.

• Mutant M6 survives. Changing owners_draw.py:304 from draw.half_seca_deduction, to draw.half_seca_deduction | draw.include_hi, means "with HI on, the whole draw bears SECA, so count it." That passes test_owners_draw.py (38 passed) and the PR's YAML (7 passed). My 6-case witness file, run against M6, gives 1 failed, 5 passed. Only one of its cases has HI on and the deduction off, the one setup M6 changes. By hand, M6 gives 39,795.89 there instead of 39,464.37: line 7 holds none of the draw's SECA, but line 5 counts all of its profit.

• Fix. Add these three cases to the PR's YAML file. I ran them at this head, and they pass (part of the 6-case witness file below). Then add matching @examples with include_hi=True, half_seca_deduction=False and phase_in_share=0.5. A st.sampled_from that gives each combination an @example would avoid relying on the random draw.

- name: With HI on, nondeductible draw SECA still keeps the draw out of line 5
  period: 2027
  absolute_error_margin: 0.02
  reforms: policyengine_us.reforms.ssa.owners_draw.owners_draw
  input:
    gov.contrib.ssa.owners_draw.current_law_cap: 200_000
    gov.contrib.ssa.owners_draw.in_effect: true
    gov.contrib.ssa.owners_draw.phase_in_share: 1
    gov.contrib.ssa.owners_draw.include_hi: true
    gov.contrib.ssa.owners_draw.half_seca_deduction: false
    gov.irs.payroll.social_security.cap: 200_000
    gov.simulation.aime_lookback_years: 0
    payroll_tax_gross_wages: 300_000
    self_employment_income: 40_000
    s_corp_income: 64_970.22
  output:
    # 59,999.998 x 2.9% = 1,740.00 of draw HI, none of it deductible.
    owners_draw_hi_tax: 1_740
    # Schedule C HI only: 36,940 x 2.9% / 2 = 535.63.
    self_employment_tax_ald_person: 535.63
    # Line 5 is 40,000: 40,000 - 535.63. Counting the draw would give 39,795.89.
    self_employed_health_insurance_earned_income_limit: 39_464.37

- name: With HI on and no OASDI room, the whole gross draw is in line 5
  period: 2027
  absolute_error_margin: 0.02
  reforms: policyengine_us.reforms.ssa.owners_draw.owners_draw
  input:
    gov.contrib.ssa.owners_draw.current_law_cap: 200_000
    gov.contrib.ssa.owners_draw.in_effect: true
    gov.contrib.ssa.owners_draw.phase_in_share: 1
    gov.contrib.ssa.owners_draw.include_hi: true
    gov.irs.payroll.social_security.cap: 200_000
    gov.simulation.aime_lookback_years: 0
    payroll_tax_gross_wages: 300_000
    self_employment_income: 40_000
    s_corp_income: 64_970.22
  output:
    owners_draw_oasdi_base: 0
    owners_draw_hi_tax: 1_740
    owners_draw_niit_exclusion: 64_970.22
    # (1,071.26 + 1,740.00) / 2 = 1,405.63.
    self_employment_tax_ald_person: 1_405.63
    # Line 6: 40,000 / 104,970.22 = 0.3810605; 40,000 - 535.63 = 39,464.37.
    # Main (draw left out of line 5) gives 38,594.37.
    self_employed_health_insurance_earned_income_limit: 39_464.37

- name: A half phase-in counts the same half in line 5 as line 7 taxes
  period: 2027
  absolute_error_margin: 0.02
  reforms: policyengine_us.reforms.ssa.owners_draw.owners_draw
  input:
    gov.contrib.ssa.owners_draw.current_law_cap: 200_000
    gov.contrib.ssa.owners_draw.in_effect: true
    gov.contrib.ssa.owners_draw.phase_in_share: 0.5
    gov.irs.payroll.social_security.cap: 200_000
    gov.simulation.aime_lookback_years: 0
    self_employment_income: 40_000
    s_corp_income: 64_970.22
  output:
    owners_draw_gross: 32_485.11
    owners_draw_oasdi_base: 30_000
    owners_draw_niit_exclusion: 32_485.11
    # (5,651.82 + 30,000 x 12.4%) / 2 = 4,685.91.
    self_employment_tax_ald_person: 4_685.91
    # Line 5: 72,485.11; line 6: 0.5518375; 40,000 - 2,585.86 = 37,414.14.
    # Main gives 35,314.09; the whole S-corp profit in line 5 would give 38,214.38.
    self_employed_health_insurance_earned_income_limit: 37_414.14

P3-1: the new docs section takes over a paragraph about the cap. docs/contributed_reforms/owners_draw.md:99-110.

• The section went in above the paragraph at :112 ("For the current-law cap basis, the taxed draw is nonnegative … The 1 cap option intentionally permits earnings above C."). That paragraph closes the "Gross draw, adjusted earnings, and cap priority" section. It now reads as part of "Self-employed health insurance limit", which it has nothing to do with.
• Fix: move lines 99-111 (the heading through the blank line after "Disabled years use the baseline formula.") to just before ## Benefit credit and active NIIT interaction (:120).
• Nit, same section: "draw_oasdi_base / q" is right only when net_earnings_factor is 0. The code divides by owners_draw_net_earnings_factor, which follows the explicit override when one is set. The NIIT section at :135 uses the same shorthand, so either both should say "divided by the net-earnings factor (q by default)" or neither.

P3-2: the hand-off's CI line is stale. It says run 41397745379 was still in_progress. It has now finished, and all 34 jobs on 8d96db39ac succeeded, none cancelled or skipped (Pull request run 41397745379; Documentation 41397745317). The body's "check its Pull request run before landing" is accurate. No change is needed beyond knowing this.

r1 findings at this head

r1	Status	Witness I ran
P2-1 line 5 counts draw profits whose SECA isn't in line 7	Resolved	Real model, PR YAML at head: half_seca_deduction: false 37,174.09; wages $300k, HI off 39,464.37; wages $150k, HI off 37,314.00 (7 passed). The same three at main give 37,174.09, 39,464.37 and 36,364.37. My hand script gives identical values, and 38,923.16 / 39,795.89 / 38,614.61 for the old whole-gross term.
P2-2 nothing tests these setups; gate mutant survives	Resolved for HI off, open for HI on and partial phase-in (P2-1 above)	M2 (ignore the gate) and M1 are now killed. M6 survives.
P3-1 copied formula can drift; overlap	Guarded	M7 (unfloor farm in the baseline file, standing in for #10120/#10131) is killed by test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5's check against main. #10560 now merges with this PR without conflict (git merge-tree exit 0), and the merged owners_draw.py keeps _enabled and owners_draw_niit_exclusion. #10120 and #10131 conflict only in files this PR doesn't touch; they already conflict with main.
P3-2 docs not updated	Resolved, with P3-1 above
P3-3 stale body; footer; redundant bounds	Resolved	The body ends with the required line and no longer says draft/no CI. The redundant bounds are gone.

The questions this round asked

phase_in_share 0.5. Line 5's draw term and line 7's tax use the same share. owners_draw_gross is share × max(0, active income). Both owners_draw_net_earnings (line 7's tax base) and owners_draw_niit_exclusion (line 5's term) derive from it. Witness (real model, head): 2027, $40,000 Schedule C, $64,970.22 S-corp profit, $200,000 caps.

• Gross 32,485.11 and OASDI base 30,000, so line 7's draw tax is 3,720 = 12.4% × 32,485.11 × 0.9235.
• Line 5's term is 32,485.11, the same half.
• The limit is 37,414.14 (main 35,314.09).
• With wages of $150,000 and share 0.5, the term is 14,141.85 and the limit 37,314.00, the same as share 1, because the room binds either way.

So line 5 never counts more draw than line 7 taxes. Counting only the phased-in half, rather than the whole S-corp profit, is a methodology choice; see below.

HI on with OASDI room used up. Witness (wages $300,000, HI on): owners_draw_oasdi_base 0, owners_draw_hi_tax 1,740.00 = 2.9% × 59,999.998 (HI on all of the gross × q), and owners_draw_niit_exclusion 64,970.22 (all of the gross). self_employment_tax_ald_person is 1,405.63 = (1,071.26 + 1,740.00) / 2, so half of the draw's HI is deductible. Line 5 then counts the whole gross. The limit is 39,464.37, equal to line 4 less Schedule C's own deductible tax, as it should be: both businesses then bear 2.9% × 0.9235 per profit dollar. With HI on and the room free (wages 0), both bear 15.3% × 0.9235 and the limit is again exactly the own-tax figure, 37,174.09.

The oracle. It now derives line 5's draw term from the rule (test_owners_draw.py:812-830): gross with HI on, min(gross × q, room) / q with HI off, room from the cap less the baseline's capped bases, and zero with the deduction off. It never reads owners_draw_niit_exclusion, owners_draw_oasdi_base, owners_draw_oasdi_room or owners_draw_net_earnings. It still takes three things from upstream:

• owners_draw_gross, so the phase-in and the $400 floor come from the code under test;
• line 7 (self_employment_tax_ald_person);
• q derived from the rates. The harness fixes net_earnings_factor at 0, so the override path isn't exercised here.

Those are covered by other tests in the file, so I don't count this as a finding. The two remaining checks against main's formula (:850-854) follow from the oracle's equality while the copy matches the baseline. They can fail, and do: M7, a change to the baseline file only, is caught by them, not by the oracle. They are the drift guard r1 asked for.

Is this a methodology choice? Yes. Form 7206 line 5 counts the net profits of businesses subject to SE tax, and a real business above the wage base still counts its whole profit. The PR instead counts only the draw profits whose SECA is in line 7, which makes line 6 allocate the deductible tax at matching per-dollar rates. The HI-on witnesses above show that this reproduces each business's own tax exactly when rates match. That choice has no statute behind it either way. I think it's the defensible reading for a parallel draw base, but the hub should own it.

Checked and sound

• Defect at main. I ran the PR's YAML file with main's expected values (origin/main 4bfc548de3, worktree, PYTHONPATH checked) and got 7 passed: hand case 33,454.09, binding premiums 33,454.09, partial room 36,364.37, other cases unchanged. Main has no change to owners_draw.py, the limit variable, the docs or the test file since the merge base 469ad12cb9.

• Hand values, computed in a standalone script from the rule (Schedule C SE tax, the draw's OASDI room after wages and Schedule C, line 5 = Schedule C + SECA-taxed gross draw if deductible) before reading the override.

  Setup	Line 7	Draw term	Limit	Main
  Hand case	6,545.91	64,970.22	37,505.61	33,454.09
  half_seca_deduction false	2,825.91	0	37,174.09	37,174.09
  Wages $300k, HI off	535.63	0	39,464.37	39,464.37
  Wages $150k, HI off	3,635.63	14,141.85	37,314.00	36,364.37
  Share 0.5	4,685.91	32,485.11	37,414.14	35,314.09
  Share 0.5, wages $150k	3,635.63	14,141.85	37,314.00	36,364.37
  HI on, wages $300k	1,405.63	64,970.22	39,464.37	38,594.37
  HI on, wages $300k, deduction off	535.63	0	39,464.37	39,464.37
  HI on, wages 0	7,415.91	64,970.22	37,174.09	32,584.09
  Share 0	2,825.91	0	37,174.09	37,174.09

  Each matches the PR's YAML or my witness YAML run on the real model at this head (6 passed: share 0.5, share 0.5 with wages $150k, HI on with wages $300k with and without the deduction, HI on with wages 0, share 0).

• The brief's own-tax invariant. Rejecting it is still right. The hand case's 37,505.61 is above line 4 less Schedule C's own deductible tax (37,174.09).

• Tests at head, one file per process:

  • policyengine_core ... test owners_draw_self_employed_health_insurance_earned_income_limit.yaml -c policyengine_us: 7 passed;
  • pytest -q -p no:cacheprovider policyengine_us/tests/reforms/test_owners_draw.py: 38 passed (230 s).

• Mutants, each in its own detached worktree and process. I ran pytest -k "sehi or disabled" (5 tests) and the PR's YAML, then removed the worktree.

  Mutant	Killed by
  M1 gate on include_hi (hand-off)	generated SEHI test; 3 YAML cases
  M2 no gate (hand-off)	generated SEHI test; 1 YAML case
  M3 whole owners_draw_gross (hand-off)	generated SEHI test; 2 YAML cases
  M4 mine: owners_draw_oasdi_base (net, not gross)	generated SEHI test; 3 YAML cases
  M5 mine: farm loss unfloored in the override's line 5	generated SEHI test only
  M6 mine: count the whole draw whenever HI is on	survives (38 passed, 7 passed); see P2-1
  M7 mine: baseline formula drifts (farm unfloored there)	generated SEHI test's main check

  git status is clean, and the only worktrees left are the main checkout and the main comparison.

• Coverage. _enabled is checked per period, and disabled years call baseline_sehi_limit.formula. The variable is per person, so filing status doesn't enter. abolish_self_emp_tax makes line 7 zero, so the draw term in line 5 then has no effect.

• Readers.

  • The only reader of the limit is self_employed_health_insurance_ald_person (min(premiums, limit)). From there it reaches the SEHI ALD lists, QBI deduction_definition, dependent_taxable_ss_magi, self_employed_health_insurance_ald_excluded_premiums, and SC, MS, OH, MO and MT items. None treats it differently from main. Under the reform they all get a larger deduction where premiums bind, which is the intended correction.
  • owners_draw_niit_exclusion and half_seca_deduction are only read, not changed. Their other reader, s770_active_business_income, is unaffected.
  • The PR's table of earned-income readers that subtract the SE tax ALD matches my reading of r1's list. Those mismatches exist on main with the reform on, so they stay in the proposed follow-up.

• Baseline. The override is installed only by the reform, and the reform is created only when an in_effect value is true (body's reforms.py:372/:754 claim). All 34 CI jobs on this head succeeded, including every baseline shard and Household API Partners. The diff can't move any baseline output, so the hub's reform-on microsim is the only population check needed. It should see the limit rise wherever premiums bind it for a Schedule C filer with a SECA-taxed draw.

• Body lines. impact: and axiom: n/a: contributed reform are present. The docs sentences match the code, apart from P3-1: line 5 adds owners_draw_niit_exclusion only with half_seca_deduction true; with it false the reform's line 7 equals the baseline deduction, max(0, SE tax − draw taxes) × 0.5, so "follows the baseline formula" holds. Commits carry the required trailer.

• Scope. Nothing under .github/ or tests/policy/baseline/partners/. Overlap search on the changed file names finds Add annual Social Security earnings histories and per-year reforms #10560, Let farm, partnership and S corporation plans set the self-employed health insurance limit #10120 and Let farm and partnership plans take the self-employed plan deduction #10131; see the r1 table.

What I couldn't check

• Sources. The session's network policy blocks irs.gov, govinfo.gov, uscode.house.gov, ecfr.gov, law.cornell.edu and ssa.gov (proxy 403 / DNS refusal). I couldn't confirm Form 7206 line 5's wording or 26 U.S.C. 162(l)(2)(A) myself, and I quote neither. My hand rule follows the line structure the PR and the baseline variable's docstring describe, which r1 confirmed against the forms.
• Population impact with the reform on: no microsimulation, as instructed.
• The hand-off's own logs on the hub machine. I reran every test it cites instead.

---

Generated by Claude Code

[MaxGhenis]

Picking this up for the US hub. I'll post a hand-off comment here when it's ready.

---

Generated by Claude Code

[MaxGhenis]

Ready for hub review

Head: bed691b7ff86066710cc469a62c608abc309f447

Round-2 fix round on the review at #10571 (comment). owners_draw.py is unchanged.

Merge

Fast-forwarded to 8d96db39ac (no new commits since the item), then merged origin/main with a merge commit (a6d18eb5). No conflicts. Main touched nothing in owners_draw.py, the SEHI limit variable, the owners' draw docs or tests (its changes under reforms/ssa/ are trustees_2026.py and __init__.py).

Findings

1 (P2): HI on with the deduction off, and a half phase-in with HI off.

• Added the review's three cases, as written, to policyengine_us/tests/policy/contrib/ssa/owners_draw_self_employed_health_insurance_earned_income_limit.yaml:163, :186 and :210.
• Hand values, from the parameter files (12.4% / 2.9% SE rates, employer share 0.5, percent deductible 0.5, so q = 0.9235), cap $200,000:
  • HI on, deduction off, wages $300,000. No OASDI room. Schedule C HI 36,940 × 2.9% = 1,071.26. Draw net 64,970.22 × 0.9235 = 59,999.998, HI 1,740.00, none deductible. Line 7 = 1,071.26 / 2 = 535.63. Line 5 = 40,000, limit 39,464.37. Under M6, line 5 = 104,970.22, share 0.3810605, limit 39,795.89.
  • HI on, wages $300,000. Line 7 = (1,071.26 + 1,740.00) / 2 = 1,405.63. Line 5 = 104,970.22, line 8 = 1,405.63 × 0.3810605 = 535.63, limit 39,464.37 (main 38,594.37).
  • Share 0.5, HI off. Gross 32,485.11, net 30,000.00, room 163,060, so the OASDI base is 30,000 and the draw tax 3,720. Schedule C SE tax 36,940 × 15.3% = 5,651.82. Line 7 = 4,685.91. Line 5 = 72,485.11, share 0.5518375, line 8 = 2,585.86, limit 37,414.14 (main 35,314.09; whole S-corp profit 38,214.38).
  • My run of the real model at head agrees: 10 passed.
• Added three @examples to test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 (policyengine_us/tests/reforms/test_owners_draw.py:795-813), following the existing rows:
  • $300,000 wages, include_hi=True, half_seca_deduction=False;
  • $300,000 wages, include_hi=True;
  • the hand case with phase_in_share=0.5 (HI off, deduction on).
  • I kept st.booleans() for the flags rather than switching to st.sampled_from. With these three, the explicit examples now cover every setup the YAML pins, and the change stays to @example rows only.

2 (P3): docs section placement. Moved "Self-employed health insurance limit" to just before ## Benefit credit and active NIIT interaction (docs/contributed_reforms/owners_draw.md:107). The cap paragraph ("For the current-law cap basis …", :100-105) now closes "Gross draw, adjusted earnings, and cap priority".

3 (P3): / q. I read owners_draw_net_earnings_factor (owners_draw.py:116-134): it returns net_earnings_factor when set, otherwise 1 − employer share × (OASDI + HI rates). owners_draw_niit_exclusion (:237-244) divides by that variable. Both doc spots now say "divided by the net-earnings factor (q by default)" (:114-115 and :136-137). I rewrapped the two edited paragraphs. The PR body's "The change" section had the same shorthand, so I changed it there too.

4: the body. Written with the PATCH call from a file and checked against GitHub's readback. Every line matches except the last: the write path replaced the closing "🤖 Generated with [Claude Code]" line with its own --- / "Generated by Claude Code" footer. I left that as it is.

• The methodology paragraph is added after the owners_draw_niit_exclusion paragraph and its table. The 39,464.37 matches my run of the second new case.
• The Tests section has this round's counts and mutant table, M6 included.
• CI: names Pull request run 41537180716 at this head, success.
• Kept the impact: line, axiom: n/a: contributed reform and the closing line. I also updated one sentence in invariant 2 to list the new explicit examples.

Tests at head (one file per process)

File	Result
owners_draw_self_employed_health_insurance_earned_income_limit.yaml	10 passed (241 s)
policyengine_us/tests/reforms/test_owners_draw.py	38 passed (188 s; 4 min 19 s wall)
policyengine_us/tests/code_health/test_source_conflicts.py	103 passed
make format	no changes; ruff check clean

Mutants

Each ran in its own process: the YAML file, then test_owners_draw.py -k "sehi or disabled" (5 tests). git checkout restored the code after each, and git status was clean.

Mutant (owners_draw.py:304-305)	YAML	pytest
(a) M6: draw.half_seca_deduction | draw.include_hi	1 failed: "With HI on, nondeductible draw SECA still keeps the draw out of line 5" (39,795.90 vs 39,464.37)	test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5 failed (1 failed, 4 passed)
(b) s_corp_income in place of owners_draw_niit_exclusion	3 failed: "A draw above the OASDI cap with HI off …" (39,795.90), "A draw partly within the OASDI cap …" (38,614.61), "A half phase-in …" (38,214.39, the comment's 38,214.38 within the 0.02 margin)	same test failed
(c) gate on draw.include_hi	5 failed: hand check and premiums case (33,454.09), partly within the cap (36,364.37), HI on nondeductible (39,795.90), half phase-in (35,314.09)	same test failed

CI

Pull request run 41537180716 at bed691b7ff: completed, success, 35/35 jobs succeeded, none cancelled or skipped. Documentation run 41537180467: success.

Judgment calls

• Kept st.booleans() and added explicit @examples (see finding 1).
• I applied the "(q by default)" wording to the PR body as well as the docs, since the body repeated the shorthand.
• The new YAML comments compare against main and against the mutants. They don't use any source-conflict phrasing, and test_source_conflicts.py passes. I see this as a methodology choice for a contributed reform, not a conflict between sources, so it has no register entry.
• Setup: uv sync alone doesn't install hypothesis, so I ran uv sync --extra dev before running the Python tests.
• I didn't fetch Form 7206 or 26 U.S.C. 162(l) this round, and I quote neither. The methodology paragraph's account of line 5 is the brief's wording, and it matches the review's.

No new follow-up beyond the earned-income readers the body already lists.

---

Generated by Claude Code

[MaxGhenis]

Hub review in progress (superseded)

The round-3 review is complete. See #10571 (comment).

---

Generated by Claude Code

[MaxGhenis]

Ready for hub review
VERDICT: APPROVE
bed691b
METHODOLOGY_CHOICE: yes

Independent round-3 review. The head hasn't moved past bed691b7ff. The range since round 2 is 8d96db39ac..bed691b7ff: the merge of main a6d18eb5 and two fix commits, 80f3fd9dff (tests) and bed691b7ff (docs).

Summary

Every round-2 finding is closed at this head. The merge adds nothing beyond main, owners_draw.py is unchanged, and I recomputed the three new YAML cases by hand. The defect reproduces at main. All three hand-off mutants are killed, and so is one of my two. The other (M9) survives every test in both files. It lives in a setup no test reaches: HI on with part of the draw inside the OASDI room. It is a coverage gap, not a wrong result, and the variable it reads is pinned elsewhere, so I rank it P3 and don't block on it. CI is green on every job.

Findings

P3-1: no test reaches HI on with OASDI room left, and mutant M9 survives there.

• Where: policyengine_us/tests/policy/contrib/ssa/owners_draw_self_employed_health_insurance_earned_income_limit.yaml (both HI-on cases use $300,000 of wages, so there is no room) and the @example rows at policyengine_us/tests/reforms/test_owners_draw.py:795-813.

• Coverage probe. I wrapped _calculate in a pytest plugin and ran test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5. It made 47 examples: the 7 @examples plus 40 generated draws, which are fixed because of derandomize=True. Rows with a positive draw and a positive Schedule C, by (phase_in_share, include_hi, half_seca_deduction, room_cap_basis, the draw's OASDI room):

  (0.5, True,  True,  0, none)    28
  (1,   False, True,  1, full)     2
  (1,   False, True,  0, full)     1
  (1,   False, False, 0, full)     1
  (1,   False, True,  0, none)     1
  (1,   False, True,  0, partial)  1
  (1,   True,  False, 0, none)     1   <- new @example
  (1,   True,  True,  0, none)     1   <- new @example
  (0.5, False, True,  0, full)     1   <- new @example
  (0.5, False, False, 0, partial)  1
  

  So the answer to "can a generated draw now reach each setup" is no. The 40 generated draws are the same as in round 2, and each of the three new setups is reached once, by its @example. Keeping st.booleans() is fine for those three, since the @examples pin them. No example of any kind has HI on with full or partial room.

• Mutant M9 (mine). At owners_draw.py:305 I replaced person("owners_draw_niit_exclusion", period), with
  where(draw.include_hi & (person("owners_draw_oasdi_base", period) > 0), person("owners_draw_oasdi_base", period) / person("owners_draw_net_earnings_factor", period), person("owners_draw_niit_exclusion", period)),
  That means "with HI on, count only the OASDI-taxed part whenever some room is left." Results:

  • the YAML file: 10 passed;
  • pytest -k "sehi or disabled": 5 passed;
  • the whole test_owners_draw.py: 38 passed.

  The witness below kills it: 1 failed, giving 36,671.24 against 38,283.08. At head it gives 1 passed.

• Why P3, not P2. The HI-on branch of owners_draw_niit_exclusion is pinned elsewhere, with room left and with none: owners_draw_net_earnings.yaml, "HI taxes all adjusted draw earnings and makes M7 exclude the full actual gross draw", which has wages of $100,000 and $250,000. (test_vector_invariants_and_phase_monotonicity pins the HI-off branch.) So only an edit to this PR's three-line override that depends on the room can get past the tests, and M9 is built to do exactly that. The code is right in this setup. A mutant that makes the same mistake without depending on the room (the term becomes oasdi_base / q whenever HI is on) is caught by the existing HI-on case with no room.

• Fix (optional). Add this case. I hand-computed it, then ran it at head (passes) and under M9 (fails). Also add the matching @example(rows=[(40_000, 0, 0, 64_970.22, 150_000, 0)], rates=(124, 29, 500), policy=_SEHI_WITNESS_POLICY | {"include_hi": True}).

- name: With HI on and part of the draw in the OASDI room, the whole gross draw is in line 5
  period: 2027
  absolute_error_margin: 0.02
  reforms: policyengine_us.reforms.ssa.owners_draw.owners_draw
  input:
    gov.contrib.ssa.owners_draw.current_law_cap: 200_000
    gov.contrib.ssa.owners_draw.in_effect: true
    gov.contrib.ssa.owners_draw.phase_in_share: 1
    gov.contrib.ssa.owners_draw.include_hi: true
    gov.irs.payroll.social_security.cap: 200_000
    gov.simulation.aime_lookback_years: 0
    payroll_tax_gross_wages: 150_000
    self_employment_income: 40_000
    s_corp_income: 64_970.22
  output:
    # Room 200,000 - 150,000 - 36,940 = 13,060 of the draw's 59,999.998.
    owners_draw_oasdi_base: 13_060
    owners_draw_hi_tax: 1_740
    owners_draw_niit_exclusion: 64_970.22
    # (4,580.56 + 1,071.26 + 1,619.44 + 1,740.00) / 2 = 4,505.63.
    self_employment_tax_ald_person: 4_505.63
    # Line 5: 104,970.22; line 6: 0.3810605; 40,000 - 1,716.92 = 38,283.08.
    # Counting only the OASDI-taxed 14,141.85 would give 36,671.24; main 35,494.37.
    self_employed_health_insurance_earned_income_limit: 38_283.08

P3-2: three small wording slips in the body and one YAML comment.

• The body's Tests section says the generated test has "explicit @examples for each setup the YAML pins … so none of them depends on the random draw." Two of the ten YAML cases, "no draw" and "no Schedule C", have no @example. The test is also derandomized, so none of its draws is random. Suggested replacement: "The generated SEHI test has an explicit @example for each of the YAML file's draw setups (the hand case, the three OASDI/deduction setups, HI on with and without the deduction, and the half phase-in). Its 40 generated draws reach none of the three new setups, so those rest on the @examples."
• The impact: line cites reforms.py:372. At this head the call owners_draw = create_owners_draw_reform(parameters, period) is at :371, and :372 is the AFA reform. :754 (the None filter) is right.
• In the third new YAML case, the comment "the whole S-corp profit in line 5 would give 38,214.38" should say 38,214.39. My hand script gives 38,214.39, and the model under mutant (b) gives 38,214.387. The hand-off already noted this falls within the 0.02 margin. It's cosmetic.

r2 findings at this head

r2	Status	Witness
P2-1 no test with HI on and the deduction off, or a partial phase-in with HI off; M6 survives	Closed	The three cases are at YAML :163, :186 and :210, with matching @examples at test_owners_draw.py:795-813. M6 is now killed by "With HI on, nondeductible draw SECA still keeps the draw out of line 5" (39,795.895 vs 39,464.37) and by test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5.
P3-1 docs section splits the cap paragraph; / q shorthand	Closed	docs/contributed_reforms/owners_draw.md: the cap paragraph (:100-105) now closes "Gross draw, adjusted earnings, and cap priority", and "Self-employed health insurance limit" (:107-120) sits just before "Benefit credit and active NIIT interaction". Both sentences (:114-115, :136-137) now say "divided by the net-earnings factor (q by default)".
P3-2 stale CI line	Closed	The body cites run 41537180716 at this head, success, 35/35. I confirmed that below.

Checked and sound

• The merge. git merge-tree --write-tree 8d96db39ac f8142b0223 gives tree e41006f9, the same tree as a6d18eb5. So the merge commit is the clean automatic merge, and nothing was hand-edited in it. git diff --stat f8142b0223 a6d18eb5 lists only this PR's five files. git diff --quiet 8d96db39ac bed691b7ff -- policyengine_us/reforms/ssa/owners_draw.py exits 0.

• Hand values, computed from the parameter files before reading the cases' code path. Rates: SE OASDI 0.124, HI 0.029, employer share 0.5, percent deductible 0.5, so q = 0.9235. Schedule C net earnings are 36,940.

  Case	Draw OASDI base	Line 7	Line 5 draw term	Limit	Main
  HI on, deduction off, wages $300k	0	535.63	0	39,464.37	39,464.37
  HI on, wages $300k	0	1,405.63	64,970.22	39,464.37	38,594.37
  Share 0.5, HI off	30,000	4,685.91	32,485.11	37,414.14	35,314.09
  Hand case (reference)	60,000	6,545.91	64,970.22	37,505.61	33,454.09

  Under M6 the first row gives 39,795.89. With the whole S-corp profit, the third gives 38,214.39. Every value matches the YAML comments, apart from the 0.01 noted in P3-2.

• The defect at main. I copied this head's YAML file into an origin/main (d4d391788d) worktree and ran it there with PYTHONPATH set to that worktree (and confirmed the import path): 5 failed, 5 passed. The failures are:

  • the hand case: 33,454.09;
  • the premiums case: 33,454.09;
  • partly within the cap: 36,364.37;
  • HI on with no room: 38,594.37;
  • the half phase-in: 35,314.09.

  Each is main's value as the YAML comments state it. The five that pass are the cases where the PR says main is already right.

• Tests at head, one file per process:

  • the YAML file: 10 passed (196 s);
  • test_owners_draw.py: 38 passed (151 s);
  • test_source_conflicts.py: 103 passed.

• Mutants. Each ran in a separate detached worktree at this head, in its own processes: the YAML file, then pytest -k "sehi or disabled". After each, git checkout restored the code and git status --porcelain was empty.

  Mutant (owners_draw.py:304-305)	YAML (10)	pytest (5)	Killed by
  (a) M6 half_seca_deduction | include_hi	1 failed	1 failed	"With HI on, nondeductible draw SECA …" (39,795.895); generated SEHI test
  (b) s_corp_income for the draw term	3 failed	1 failed	above-cap HI off (39,795.895), partly within cap (38,614.605), half phase-in (38,214.387); generated SEHI test
  (c) gate on include_hi	5 failed	1 failed	hand case and premiums (33,454.09), partly within cap (36,364.37), HI on nondeductible (39,795.895), half phase-in (35,314.09); generated SEHI test
  M8 (mine) draw term × _share(period, parameters) (phase-in applied twice)	1 failed	1 failed	"A half phase-in …" (36,667.355); generated SEHI test
  M9 (mine) see P3-1	10 passed	5 passed (whole file 38 passed)	only my witness case

  These match the hand-off's mutant table case for case. Afterwards I removed every worktree, and the main checkout's git status is clean.

• The rule, in every setup the change reaches. _enabled is checked per period, and disabled years call baseline_sehi_limit.formula. The variable is per person, so filing status doesn't enter. The draw term is owners_draw_niit_exclusion, gated on half_seca_deduction. With HI on that is owners_draw_gross; with HI off it is owners_draw_oasdi_base / owners_draw_net_earnings_factor, and the factor follows the override when one is set. So it is the draw profit whose SECA is in line 7, under every combination of flags.

• Docs against code. I read owners_draw_net_earnings_factor (owners_draw.py:116-134): it returns the override when nonzero, else 1 - employer_share × (OASDI + HI). That is the docs' q at :63-66, which counts HI even when HI on the draw is off. owners_draw_niit_exclusion (:237-244) divides by that variable. Both "divided by the net-earnings factor (q by default)" sentences are therefore accurate. draw_oasdi_base is the docs' own symbol (:82). The rest of the moved section matches the override: "the whole actual gross draw D with HI on", "line 5 omits the draw and the limit follows the baseline formula" with the deduction off, and "Disabled years use the baseline formula".

• Readers. The only reader of self_employed_health_insurance_earned_income_limit is self_employed_health_insurance_ald_person. From there it reaches:

  • the SEHI ALD totals and dependents_self_employed_health_insurance_ald;
  • qbi/deduction_definition, dependent_taxable_ss_magi and self_employed_health_insurance_ald_excluded_premiums;
  • SC two-wage-earner subtractions, MS, OH, MO (AGI and premiums) and MT.

  Each uses the deduction just as on main. Under the reform each gets a larger deduction where premiums bind, which is the intended correction. No reader is made worse. The earned-income readers the body lists as "same kind" subtract the SE tax ALD and don't read this variable. They behave the same on main with the reform on, so they belong to the proposed follow-up.

• Baseline can't move. create_structural_reforms_from_parameters (reforms.py:371) calls create_owners_draw_reform, which returns the reform only when some in_effect value is true (owners_draw.py:377-380). in_effect defaults to false, and reforms.py:754 filters out None. Every baseline shard and Household API Partners passed. The only population check needed is the hub's reform-on microsimulation: the limit should rise only for Schedule C filers with a SECA-taxed draw whose premiums bind.

• Body lines. impact: and axiom: n/a: contributed reform are present. The methodology paragraph's 39,464.37 matches my run of the second new case. The mutant table matches my reruns.

• Methodology choice: yes. Line 5 counts only the draw profit whose SECA is in line 7, rather than all of the S-corp profit, and only the phased-in share. That is one reading where the law offers no direct rule for a parallel draw base. The HI-on, no-room case is consistent with it: the limit then equals line 4 less Schedule C's own deductible tax.

• Order with Add annual Social Security earnings histories and per-year reforms #10560. Add annual Social Security earnings histories and per-year reforms #10560 (head 2166886f48) merges with this head without conflict: git merge-tree exits 0. In the merged tree, the SEHI override still uses _enabled, which Add annual Social Security earnings histories and per-year reforms #10560 keeps. Add annual Social Security earnings histories and per-year reforms #10560's kernel computes niit_exclusion the same way: gross with include_hi, else oasdi_base / net_earnings_factor (tools/social_security/owners_draw.py:264-270). I ran the merged tree:

  • this YAML file: 10 passed;
  • the whole test_owners_draw.py: 38 passed.

  Once this lands, Add annual Social Security earnings histories and per-year reforms #10560 needs only a merge of main. Its doc hunk (the benefit-credit paragraph) lands just after the moved SEHI section and merges cleanly. Its test_installed_formulas_delegate_to_the_same_kernel_and_honor_inputs iterates only the owners_draw_* outputs, so this override doesn't need to delegate to the kernel.

• Scope. Nothing under .github/ or policyengine_us/tests/policy/baseline/partners/. The files are the changelog fragment, the docs, owners_draw.py, the YAML file and test_owners_draw.py. The open-PR search on the changed file names finds Add annual Social Security earnings histories and per-year reforms #10560 (above) and the drafts Let farm, partnership and S corporation plans set the self-employed health insurance limit #10120 and Let farm and partnership plans take the self-employed plan deduction #10131. Those two edit the baseline limit formula, so whichever lands second must carry the change into the override's copy. test_generated_sehi_limit_counts_seca_taxed_draw_in_line_5's check against main is the guard for that (round 2's M7).

• CI at this head. Pull request run 41537180716: completed, success, 35 of 35 jobs succeeded, none cancelled or skipped. Documentation run 41537180467: success. Its build succeeded, and "Deploy documentation to GitHub Pages" was skipped, as it is for every PR.

What I couldn't check

• Sources. This session's network policy refuses irs.gov, govinfo.gov, uscode.house.gov, ecfr.gov and law.cornell.edu (proxy 403 or DNS refusal). I couldn't read Form 7206, its instructions or 26 U.S.C. 162(l), and I quote none of them. My hand rule follows the line structure in the baseline variable and the PR, which round 1 confirmed against the forms.
• Population impact with the reform on: no microsimulation, as instructed.
• Add annual Social Security earnings histories and per-year reforms #10560's own test_owners_draw_kernel.py at the merged tree; I ran only this PR's two test files there.
• The hand-off's local timings. I reran every test it cites instead.

---

Generated by Claude Code

[MaxGhenis]

US + core hub merge audit, us#10571 at 8e10a070d9474f0df7090701d50065cd220101a7 (squash): Count the owners' draw in the SEHI limit's line 5 under the owners' draw reform.

• Review: cloud review r3 (Opus 5.5, independent) approved at bed691b7ff with P3 notes only. It found every round 2 finding closed, recomputed the three new YAML cases by hand, and killed the three hand-off mutants and one of its own two. The other, M9, survived every test, so the hub added the review's witness YAML case and @example at 8e10a070d9; the YAML case failed under M9 in the review's run. The hub also revised the body's wording. The review's other two wording notes don't hold: reforms.py:372 holds the call, and exact arithmetic gives 38,214.38.
• Gates checked live at merge:
  • gh pr checks exits 0;
  • the latest Pull request run at the head concluded success;
  • MERGEABLE;
  • not a draft;
  • no reviewer's latest review requests changes;
  • the semantic-overlap guard;
  • head pinned.
• Impact: none at baseline: create_owners_draw_reform returns the reform only when some gov.contrib.ssa.owners_draw.in_effect value is true, and that parameter defaults to false.
• Order: this PR lands before Add annual Social Security earnings histories and per-year reforms #10560, which rewrites the same owners_draw.py and merges main afterwards.
• Axiom: n/a: contributed reform