[MaxGhenis]

Fix the 2020 unemployment compensation exclusion, including the strict income cutoff, each spouse's cap and dependents' separate returns. Also net dependents' estate and trust losses and apply their own excess business loss limit.

Law and calculation

Under IRC 85(c), taxpayers with exclusion AGI below $150,000 exclude up to $10,200 of their own 2020 compensation. Each spouse receives a separate cap on a joint return. The IRS leaves compensation out when calculating that AGI and applies a strict cutoff in its 2020 exclusion worksheet, lines 1–9.

For example, spouses who receive $20,000 and $5,000 exclude $10,200 and $5,000 when eligible, leaving $9,800 taxable. A taxpayer whose exclusion AGI reaches $150,000 receives no exclusion.

Changes

• Calculate taxable_unemployment_compensation for each person from total_unemployment_compensation. Compare the filers' taxable_uc_agi directly with the cutoff and test dependents against their own dependent_taxable_uc_agi.
• Sum only nondependents in tax_unit_taxable_unemployment_compensation, matching the filers' federal gross income. Indiana's unemployment deduction and the TAXSIM output receive that return total.
• Preserve modeled state unemployment insurance and reported compensation overrides from main. Leave taxable compensation at zero for a net repayment, and preserve the negative balance in tax_exempt_unemployment_compensation.
• Include estate and trust losses in the dependent income helper and limit business losses to business income plus the dependent's single or separate IRC 461(l) threshold. The helper also feeds dependent Social Security MAGI and IRA deductions.
• Pin both terms identified in review with YAML cases: positive business income within the dependent loss limit, and a dependent repayment that must not raise exclusion AGI.
• Add the promised 2020 California and Rhode Island cases for the boundary between returns. Keep a dependent's own taxable or excluded compensation off the filers' state adjustment.
• Remove the property assertion that repeats the exclusion formula. Retain checks for bounds, the AGI differential, dependents' separate returns, spouse symmetry and the exclusion ending after 2020.

Outside 2020, the exclusion amount remains zero. We still exclude dependents from the tax unit total and apply their own business loss limits.

Consumers and related work

As checked on 2026-11-13, #10187 merged on 2026-10-18, #10191 merged on 2026-11-10, and #10331 merged on 2026-11-09. We merged their California, Rhode Island, Wisconsin and Montana changes from main.

We now count gross compensation in South Carolina TANF through total_unemployment_compensation, so we removed its outdated grid and proposed follow-up.

We still count a dependent's full compensation in the qualifying relative gross income test. Directly substituting taxable compensation would create a cycle through filing status and head of household eligibility. The hub will review follow-up drafts for that calculation and § 461(l) parity.

Validation

Final head: d3f91be.

Local checks passed both property tests and all nine dependent Social Security YAML cases. The dependent unemployment API run timed out before assertions. Its isolated YAML retry remains incomplete.

We did not complete unmodified local California or Rhode Island file runs before cleanup. CI passed both files. The required local file matrix remains incomplete.

The final local source-conflict run reached its 180-second cap after two completed cases, without a file result. CI passed all 103 guard cases.

Each of four semantic mutants failed its new YAML regression case: business-income allowance, repayment floor, California return boundary and Rhode Island return boundary.

CI passed all 411 cases across the 28 requested files in Pull request run 43471044578, with no failures, errors or skips. All 36 jobs succeeded, with no whole jobs skipped or cancelled.

GitHub skipped 25 conditional steps. Quick Feedback deferred nine changed test files because its limit allows eight. Later CI jobs ran their cases in shards and batches. We verified case counts against JUnit records and exact-path test logs.

Impact

impact: pending a comparison at the final head.

axiom: TheAxiomFoundation/rulespec-us#1719 queued. us:statutes/26/85 already encodes § 85(c)(1) correctly: per-recipient $10,200 cap, joint return summed per spouse, strict <, with tests temporary_2020_exclusion_applies_per_spouse_on_joint_return and temporary_2020_exclusion_denied_at_threshold. But the § 85(c)(2) AGI and the recipient relation are free inputs. #1719 queues encoding both, with companion tests from the same worksheet cases.

🤖 Generated with Claude Code

[MaxGhenis]

Hand-off to the US + core hub. Session bebaaf35, head 9fde2d1ec4.

• Independent review: APPROVE at 9fde2d1ec4. Round 4 was served by Opus 5.5 because the Codex lanes were closed; rounds 1–3 were GPT-6.1 Sol. Reports: ~/reviews/uc-exclusion-2020/review_out.md, review2_out.md, review3_out.md, review4_out.md; author responses response_round2.md and response_round3.md.
  • Every round-1 and round-2 finding is fixed, accepted as out of scope, or withdrawn with evidence:
    • Fixed: the § 461(l) limit on the dependent's own return, non-negative taxable UC, and the citations.
    • Out of scope: the WI homestead, SC TANF and PR consumers.
    • Withdrawn: the DE/MT/DC/IA/UT/RI-1040H pooling findings. The per-cell error split and the screen-sample count show every worse case needs a dependent with both UC and an own AGI of $150k or more, and there are 0 such dependents.
  • On a merged tree of this head + Leave dependents' unemployment compensation off California, Rhode Island and Wisconsin filer returns #10187 208a87f7a9 + Leave dependents' income out of Wisconsin homestead credit household income #10191 509e5c9d71, the reviewer found both merges clean, 82/82 YAML cases passing in the 13 touched files, and 8/8 invariant tests passing.
• CI: all 26 checks pass at 9fde2d1ec4 (gh pr checks exits 0). Mergeable.
• Landing order: land after Leave dependents' unemployment compensation off California, Rhode Island and Wisconsin filer returns #10187 (CA/RI/WI state lines) and Leave dependents' income out of Wisconsin homestead credit household income #10191 (WI homestead household). This PR gives each dependent their own 2020 exclusion. Those two PRs keep a dependent's amounts off the filers' state lines. The branch does not need main merged in (no conflicts; branch protection is not strict).
• Impact: both full-file pairs ran at this head; tables are in the body. Current law 2024 moves no tax. The 2020 rules applied to 2024 data move AGI by +$278M and income tax by +$96M. A literal 2020 run would read all-zero inputs on the 2024 dataset.
• Axiom: [pe-parity] Encode IRC §85(c)(2) AGI for the 2020 unemployment exclusion and limit recipients to taxpayer and spouse (PE-US #10175) TheAxiomFoundation/rulespec-us#1719 is queued (pe-parity) for § 85(c)(2) AGI and the recipient relation.
• No hold, no decision. It is a bug fix to the statute and the IRS worksheet.
• Follow-ups, not blocking:
  • session chips for DE/DC/IA/PA/MT/UT pooling of dependents' UC and for SC TANF counting taxable UC;
  • a proposed session for Puerto Rico gross income including federally taxable UC;
  • Take the 2020 charitable deduction for non-itemizers above the line #10186 (2020 $300 charitable deduction above the line);
  • Leave dependents' own items out of DC, Delaware and Ohio subtractions #9939/Leave dependents' capital income out of state lines on the filers' return #9894 (DE/DC);
  • issues AL, AR, MS and NJ add tax unit dependents' income to the filer's state income (Iowa net income too), although federal AGI leaves it out #9814/Iowa and Mississippi move a dependent's net income onto the head's state return #9887 (IA) and South Carolina TANF counts taxable UC and taxable Social Security instead of gross amounts #9677 (SC TANF).

[DTrim99]

PR Review: Fix the 2020 unemployment compensation exclusion: cutoff test, per-spouse cap, dependents

Summary

I checked the law before reading the code.

The statute and form.

• 26 U.S.C. 85(c)(1), added by ARPA § 9042: "if the adjusted gross income of the taxpayer for such taxable year is less than $150,000, the gross income of such taxpayer shall not include so much of the unemployment compensation received by such taxpayer (or, in the case of a joint return, received by each spouse) as does not exceed $10,200."
• § 85(c)(2): that AGI is figured "(A) after application of sections 86, 135, 137, 219, 221, 222, and 469, and (B) without regard to this section."
• Conforming amendments. Pub. L. 117-2 also inserted "85(c)," into § 86(b)(2)(A) and § 219(g)(3)(A)(ii) (Cornell LII amendment notes). Taxable Social Security and the IRA limit are therefore figured with all of the compensation.
• The worksheet. The 2020 Instructions for Form 1040 (Rev. 4-2021) implement this in the Unemployment Compensation Exclusion Worksheet, file page 88:
  • line 2: "Don't include any amount of unemployment compensation from Schedule 1, line 7";
  • line 6: "This is your modified adjusted gross income";
  • line 7: "Is the amount on line 6 $150,000 or more? The $150,000 threshold applies to all filing statuses ... Yes. Stop.";
  • lines 8 and 9: the compensation paid to you and, if married filing jointly, to your spouse, "Don't enter more than $10,200" on each line.
• The joint example. File page 87 gives it: $20,000 and $5,000 of compensation, $15,200 excluded.

The PR matches the law on all three fixes:

• the cutoff AGI leaves out the compensation exactly once, and the test is a strict "<" at $150,000 for every filing status;
• each spouse's cap is their own compensation up to $10,200;
• a dependent, who files their own return, is tested on their own § 85(c)(2) AGI and excludes their own compensation there, outside the filers' return.

Code and tests.

• The person-level formula is vectorized and the entities and periods are right.
• I worked 25 expectations by hand.
• On a tree merged with today's main (479 commits ahead of the merge base), the PR's YAML files, its invariants module and the main-side guards it could trip all pass. So do the main-side tests that read the changed variables.

Open items.

• One should: two new terms survive deletion.
• Four suggestions.

Critical

None.

Should address

S1. Two new terms are unpinned: deleting either fails no test.

(a) The business income term in the dependent's IRC 461(l) limit.

• Where: dependent_taxable_ss_magi.py, dependent_income_and_deductions:

  income += business_income - min_(business_loss, business_income + threshold)

• The law: § 461(l)(3)(A) allows business deductions up to "(I) the aggregate gross income or gain of such taxpayer for the taxable year which is attributable to such trades or businesses, plus (II) $250,000".
• Why no test catches it: the only test where the limit binds is the 2025 $700,000 estate loss case, and that dependent has no business income. The 2020 case (dependent_taxable_uc_agi.yaml Case 4) has an infinite threshold. Deleting business_income + would cap the loss at the threshold alone, and every test would still pass.
• Fix: add a 2025 case to dependent_taxable_ss_magi.yaml.
  • Inputs: a dependent with dividend_income 400,000, rental_income 100,000 and estate_income -500,000.
  • Arithmetic: the allowed loss is min(500,000, 100,000 + 313,000) = 413,000, so modified AGI is 400,000 + 100,000 - 413,000 = 87,000.
  • Checks: I confirmed 87,000 on the merged tree. Without the term the result is 187,000.

(b) The floor on the compensation removed in dependent_taxable_uc_agi.

• Where: dependent_taxable_uc_agi.py:

  income -= max_(0, person("total_unemployment_compensation", period))

• Why the floor matters: the shared helper adds only positive compensation. The floor keeps the subtraction consistent with that, so a net repayment never raises worksheet line 6.
• Why no test catches it:
  • No YAML case gives a dependent a net repayment.
  • The Hypothesis invariant 2 uses dependent_taxable_uc_agi only to decide an exclusion whose cap is 0 when compensation is negative, so it can't see the difference.
• Fix: add a 2020 case to dependent_taxable_uc_agi.yaml.
  • Inputs: a dependent student with employment_income 12,000 and unemployment_compensation -500.
  • Expected: dependent_taxable_uc_agi 12,000, taxable_unemployment_compensation 0, tax_exempt_unemployment_compensation -500.
  • Checks: I confirmed all three on the merged tree. Without the floor, line 6 is 12,500.

Suggestions

G1. Extend the axiom line to the dependent-loss change.

• The gap: the line covers § 85 only. The PR also changes the dependent's own-return AGI: the § 461(l) limit and netting estate or trust losses. That AGI feeds § 86(b)(2) MAGI, dependent_traditional_ira_deduction and the new § 85(c)(2) AGI, and the changelog lists the estate-loss change as a fix.
• rulespec-us coverage: us/statutes/26 has 62 and 86 modules but no § 461 module.
• Fix: say how parity is handled for that part, either a queued issue or a note in NYC School Tax Credit Fixed Amount #1719.

G2. Refresh the PR body before squash. The squash commit will carry it.

• SC TANF is now fixed on main. Main counts gross unemployment compensation in SC TANF since Count gross unemployment compensation and Social Security in South Carolina TANF #9687 and Count South Carolina TANF unearned income at gross amounts and leave SSI recipients out of the benefit group #10184. The SC TANF grid row and its follow-up are stale.
• Two promised cases are missing. The body says that once Leave dependents' unemployment compensation off California, Rhode Island and Wisconsin filer returns #10187 merged, two 2020 cases would follow: California with a dependent who has $150,000 of dividends, and Rhode Island with a filer over the cutoff. Leave dependents' unemployment compensation off California, Rhode Island and Wisconsin filer returns #10187 merged on 2026-10-18 and is in this branch, but neither case was added. Add them or drop the promise.
• The landing-order note needs a date. It should say that Leave dependents' unemployment compensation off California, Rhode Island and Wisconsin filer returns #10187 has merged and Leave dependents' income out of Wisconsin homestead credit household income #10191 is still open.

G3. Follow-up issue: dependent_gross_income in 2020.

• The issue: it counts a dependent's full compensation for the § 152(d)(1)(B) gross income test, through the override in DEPENDENT_GROSS_INCOME_SOURCE_OVERRIDES. Yet § 85(c)(1) says "the gross income of such taxpayer shall not include" the excluded part.
• Why it isn't in scope:
  • It predates this PR and is outside the diff.
  • A direct fix would cycle through is_tax_unit_dependent.
• Why file it now: this PR computes each dependent's own exclusion, so the gap is easier to close.

G4. Invariant 2 restates the formula.

• The issue: in test_unemployment_compensation_exclusion_invariants.py, invariant 2 is np.maximum(0, received - under * cap). CLAUDE.md asks not to test a formula against itself.
• What earns the module its place: invariants 3 to 5 (the AGI differential, the return boundary and spouse symmetry) are independent checks.
• Options: drop invariant 2, or derive under independently.
• Cost: the module is cheap. At the head, Rest CI timed it at 0.49 s and 0.19 s per call, and it took 9.6 s locally.

Verified correct

Law and form.

• § 85(c)(1)-(2) text, checked on Cornell LII.
• Pub. L. 117-2 inserted 85(c) into § 86(b)(2)(A) and § 219(g)(3)(A)(ii).
• Worksheet lines 1-11, file page 88.
• The file page 87 note: the joint example, and the caution that taxable Social Security, the IRA, student loan interest and tuition deductions use the full compensation.
• Social Security Benefits Worksheet line 3 (file page 31) includes Schedule 1 line 7 in full.
• The PDF is the Rev. 4-2021 revision.

Code.

• taxable_unemployment_compensation:
  • each person's own exclusion is eligible * min(max(0, received), 10_200);
  • filers are tested on taxable_uc_agi < cutoff[filing_status], a dependent on dependent_taxable_uc_agi < cutoff[SINGLE or SEPARATE];
  • the result is floored at 0.
• Every cutoff is $150,000 in 2020. All five filing statuses are 150,000 for 2020 and 0 otherwise, and the amount is 0 outside 2020.
• Non-dependents are the head and spouse only. is_tax_unit_dependent is ~head & ~spouse, so sum_among_non_dependents matches irs_gross_income.
• The old double subtraction is gone. taxable_uc_agi already drops taxable_unemployment_compensation, and taxable_ss_magi uses total_unemployment_compensation, so Social Security is figured with all of the compensation, with no cycle.
• The dependent override counts full compensation. DEPENDENT_GROSS_INCOME_SOURCE_OVERRIDES maps taxable compensation to total_unemployment_compensation, so the dependent's income includes it in full before dependent_taxable_uc_agi removes it once.
• The new 461(l) helper keeps 2018-2020 results. It mirrors limited_business_loss, and with the infinite 2018-2020 threshold it reduces to the old netting.

Hand checks (each person's own exclusion = min(own compensation, 10,200) if their return's line 6 < 150,000):

1. taxable_unemployment_compensation Case 4. Line 6 = 150,000 pension. Not under 150,000, so taxable 8,000 and AGI 158,000.
2. Case 5. Line 6 = 149,999, so 8,000 is excluded; taxable 0, AGI 149,999.
3. Case 6 (the IRS joint example). Line 6 = 60,000. The head excludes min(20,000, 10,200) = 10,200, so 9,800 is taxable; the spouse excludes all 5,000. Total 9,800; AGI 60,000 + 9,800 = 69,800.
4. Case 7. Each spouse has 20,000 - 10,200 = 9,800 taxable. Total 19,600 = AGI.
5. Case 8. The couple's line 6 = 30,000, so the spouse's 8,000 is excluded. The dependent's own line 6 = 4,000 - 4,000 = 0, so their 4,000 is excluded. The tax unit has 0 taxable; AGI 30,000.
6. Case 9. The dependent's taxable amount is 15,000 - 10,200 = 4,800, all on their own return. The tax unit has 0.
7. Case 10. The dependent's line 6 = 150,000 + 5,000 - 5,000 = 150,000, so no exclusion and 5,000 taxable. The head's line 6 = 30,000, so 0 taxable.
8. Case 11. SS worksheet: line 3 = 140,000 + 10,000 = 150,000; plus half of benefits gives 160,000. Taxable benefits = min(0.85 × 20,000 = 17,000, 0.85 × (160,000 - 34,000) + 4,500) = 17,000. Line 6 = 157,000, so no exclusion; AGI 167,000.
9. Case 12. Line 6 = 150,400 - 500 = 149,900, so 5,000 is excluded; AGI 149,900.
10. Cases 14 and 15. For -500 of compensation: excluded = min(max(0, -500), 10,200) = 0; taxable = max(0, -500) = 0; exempt = -500 - 0 = -500.
11. tax_unit_taxable_unemployment_compensation Case 4. Head 15,000 - 10,200 = 4,800; spouse 12,000 - 10,200 = 1,800; child 11,000 - 10,200 = 800, on the child's own return. Return total 6,600; AGI 40,000 + 6,600 = 46,600.
12. dependent_taxable_uc_agi Case 2. Provisional income = 30,000 + 6,000 + 5,000 = 41,000. Taxable benefits = min(8,500, 0.85 × 7,000 + 4,500 = 10,450) = 8,500. Line 6 = 30,000 + 8,500 = 38,500.
13. Case 3. 150,000 + 5,000 - 1,000 - 5,000 = 149,000, so 5,000 is excluded.
14. Case 5. MAGI = 12,000 + 6,000 - 100 = 17,900 and 12,000 + 4,000 - 100 = 15,900. Line 6 = 11,900 each.
15. dependent_taxable_ss_magi, 2025 461(l) case. Allowed loss = min(700,000, 0 + 313,000) = 313,000. MAGI = 401,000 - 313,000 = 88,000. Taxable benefits = min(17,000, 0.85 × (98,000 - 34,000) + 4,500) = 17,000.
16. Indiana Case 2. Reduced AGI = max(0, 15,000 - 18,000) = 0, so the deduction is all 5,000.
17. Indiana Case 3. The return holds 0 of the dependent's compensation, so the deduction is 0.
18. Montana Case 10. The claimant's line 6 = 20,000, so all 8,000 is excluded and the addback is 8,000. The dependent's 155,000 already includes their full compensation, so their addback is 0. Gross household income = 20,000 + 8,000 + 1,800 = 29,800, where 1,800 is the 2020 recovery rebates: $1,200 CARES plus $600 CAA. The 20-year-old dependent adds no child amount.

Other checks.

• Changelog: top-level fix-uc-exclusion-2020.fixed.md.
• References: the amount and cutoff references corroborate $10,200 and $150,000. They give a single #page=88 with no page in the title.
• Test files: tests sit in variable-named YAML files.
• NH Add New Hampshire state supplementation for 2026 #10156 pattern: not applicable; there is no new state-gated key.
• Partner files: none touched.
• Python test cost: the PR body reports local time and memory and names the Rest core group, where it is collected automatically on main.

Axiom line assessment

Line: axiom: TheAxiomFoundation/rulespec-us#1719 queued, with an encoded-correct statement for us:statutes/26/85.

rulespec-us#1719 is valid.

• State: OPEN and labelled pe-parity.
• Dispatch-ready. It has:
  • the module path us/statutes/26/85.yaml and corpus citation us/statute/26/85;
  • the verbatim § 85(c) text;
  • the required outputs, which replace the free § 85(c)(2) AGI input and limit recipients to the taxpayer and spouse;
  • six companion tests (A-F) taken from the 2020 Form 1040 instructions, file pages 87-88 and 31 (YAML Cases 4-8, 11 and 12);
  • the recorded blocker: the signed encoder needs an approved billed run.
• Matches the head: it names taxable_uc_agi, dependent_taxable_uc_agi and the person and tax-unit outputs.

The encoded-correct claim holds. us/statutes/26/85.yaml was last changed by rulespec-us #525.

• Exclusion: each recipient's exclusion is min(max(0, unemployment_compensation), 10200), summed over the recipient relation, behind a strict < 150000.
• Tests: temporary_2020_exclusion_applies_per_spouse_on_joint_return (12,000 and 13,000 give 20,400 excluded) and temporary_2020_exclusion_denied_at_threshold (AGI 150,000 gives 0) both exist.
• Free inputs: the § 85(c)(2) AGI and the recipient relation are free inputs, as the line says.

Coverage: see G1 for the § 461(l) part.

CI status

All 27 checks pass at bceffce584 (run 39190151231, 2026-10-19). gh pr checks near the end of this review shows nothing pending.

That run predates several main-side guards: #10287 (source-conflict ratchet), #10317, #10408 and the #10439 Rest sharding. I checked them on the merged tree instead (see Branch status).

Branch status

Position: 7 commits ahead and 479 behind main, merge base 2ea9a7d3c4.

git merge-tree --write-tree against today's main is clean.

• Since the merge base, main changed one PR file, charitable_deduction.yaml (Count taxable employment disability benefits in federal gross income (IRC 105(a)) #10185). This PR only trims trailing blank lines there, and the merge is clean.
• Main also added new readers of taxable_unemployment_compensation: the Arizona property tax credit (Count every co-resident's income in the Arizona property tax credit #10580). It subtracts only non-dependents' amounts, which is consistent with this PR.
• Main dropped one reader: SC TANF now reads gross compensation.

On the merged tree (one small run at a time), these pass:

• Main-side guards:
  • Integrate modeled state unemployment insurance into income flows #9558 test_unemployment_compensation_readers.py;
  • Count military SBP annuities in state benefit income lists #10317 test_military_retirement_readers.py;
  • Count each Social Security benefit once in Pennsylvania CCW countable income #10408 test_social_security_source_lists.py;
  • the Tie source-conflict markers to the register shared with Axiom #10287 ratchet and marker tests in test_source_conflicts.py;
  • none of the PR's comments match the conflict patterns, and none of its files are in KNOWN_UNMARKED.
• The PR's own tests:
  • its six YAML files (45 cases);
  • its invariants module (2 tests).
• Main-side tests that read the changed variables:
  • the Montana nontaxable-income tests (17);
  • charitable integration.yaml;
  • Arizona household income;
  • Montana disability and unemployment;
  • SC TANF gross unearned income and integration;
  • student loan interest;
  • MA COVID premium pay;
  • total_unemployment_compensation;
  • RI additions;
  • CA AGI subtractions;
  • MA qualified unemployment deduction;
  • MT and WI unemployment subtractions;
  • the Arizona source-reform test.

Siblings (note only):

• Leave dependents' income out of Wisconsin homestead credit household income #10191 (Wisconsin homestead household) is still open. The body asks to land this PR after it.
• Fix Montana disability exclusion repeal, elderly credit receipt timing and occupancy tests #10331 (Montana elderly credit) merges cleanly on top of this PR plus main. On that three-way tree, the Montana unemployment-addback file passes, including this PR's Case 10, with the $1,800 receipt coming from the default. So the input Fix Montana disability exclusion repeal, elderly credit receipt timing and occupancy tests #10331's body asks this PR to supply is no longer needed.

0 critical, 1 should and 4 suggestions are open at this head.

Verdict: REQUEST_CHANGES

[DTrim99]

Requesting changes: 0 critical, 1 should (two new terms are unpinned: business_income in the dependent's 461(l) limit, and the max_(0, ...) removed on UC in dependent_taxable_uc_agi; the report gives a case for each). All three fixes match § 85(c) and the 2020 worksheet, and on a tree merged with current main the #9558/#10317/#10408 guards, the ratchet and the PR's 45 cases pass. Full report in the comment above.