Skip to content

Round the Vermont veteran tax credit's phase-out steps down - #10871

Merged
MaxGhenis merged 2 commits into
mainfrom
fix-vt-veteran-credit-rounding
Nov 14, 2026
Merged

MaxGhenis merged 2 commits into
mainfrom
fix-vt-veteran-credit-rounding

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

What and why

Vermont gives a veteran with federal AGI of $25,000 or less a $250 refundable credit. For AGI above $25,000 and below $30,000, Vermont pays "$250.00 less $5.00 per $100.00" of AGI above $25,000, and it pays nothing at $30,000 or more (32 V.S.A. § 5830g(c)). The section says nothing about a partial $100. In the section before it, the legislature wrote rounding up into the child tax credit's phase-out: "for each $1,000.00, or fraction thereof" (§ 5830f(b)).

The Department of Taxes fills that gap on the 2025 Schedule IN-112, Part II, line 10: "DIVIDE Line 9 by 100, rounding down to the nearest whole number." The Schedule IN-112 instructions give the same line.

vt_veteran_tax_credit counted the $100 steps with np.ceil, so a partial $100 took $5 like a whole one. For example, at AGI of $25,050 the schedule gives $250 and the model paid $245, and at $29,950 the schedule gives $5 and the model paid $0. The formula now uses np.floor, so the credit rises by $5 for a tax unit with a veteran and AGI above $25,000 and below $30,000, unless AGI exceeds $25,000 by a whole multiple of $100. Across every whole-dollar AGI from $0 to $40,000, the old formula and the schedule differ at 4,950 values from $25,001 to $29,999, each by $5, and agree everywhere else.

The variable's reference now cites § 5830g, the schedule and its instructions in place of a news article. The five parameter files' booklet links move from page 36, which holds the Schedule IN-113 instructions, to page 34, which holds this credit's instructions.

The Department of Taxes hasn't posted a 2026 Schedule IN-112, so the model applies the 2025 schedule's rounding to 2026 under the same statute.

Tests

Pull request run 43611035458 at 6fdd72d82a passed all 36 jobs on GitHub's merge with main 9a63ae6605.

  • policyengine_us/tests/policy/baseline/gov/states/vt/tax/income/credits/vt_veteran_tax_credit.yaml gains 19 cases at each edge of the worksheet. For 2025 they cover AGI of -$1,000, $24,999, $25,000, $25,001, $25,099, $25,100, $25,101, $25,199, $25,200, $27,550, $29,899, $29,900, $29,901, $29,999, $30,000, $30,001 and $30,100. For 2026 they cover $25,099 and $29,999. Each case name carries the worksheet's arithmetic, for example "floor(4,999 / 100) = 49, pays $250 - 49 x $5 = $5".

  • policyengine_us/tests/variables/gov/states/vt/tax/income/credits/vt_veteran_tax_credit.yaml now expects $250 at AGI of $25,050 and $5 at $29,950. Four of its cases had no state code, so the default state, California, kept them from reaching the formula. They now set Vermont.

  • policyengine_us/tests/core/test_vt_veteran_tax_credit_invariants.py types the worksheet's lines from the schedule, without reading the parameters or the formula, and checks three invariants:

    • a Vermont tax unit with a veteran gets the worksheet's amount from 2025, and every other tax unit gets zero;
    • the credit stays between $0 and $250;
    • among qualifying tax units, a higher AGI never brings a higher credit.

    One simulation per year checks every whole-dollar AGI from $0 to $35,000 in 2025 and 2026, where the YAML cases sample 19 points. A Hypothesis test (12 derandomized examples of up to 20 tax units) adds negative, large and fractional AGIs, joint units, units without a veteran, New Hampshire units and 2024. The file runs in the Makefile's core Rest Python group.

Local runs

  • With the formula back on np.ceil, 10 of the 19 new baseline cases fail, each by $5, and so do the variables-folder cases at $25,050 and $29,950.

  • All 3 invariants tests fail against np.ceil. Each whole-dollar test reports 4,950 of 35,001 AGIs off by $5, and the Hypothesis test fails on a Vermont veteran with AGI of $25,000.01.

  • On macOS, the invariants file's process peaked at 1.22 GB of resident memory, and its three test calls took 2.2 s together.

  • YAML tests take their expected values from the 2025 Schedule IN-112 worksheet and § 5830g(c)(3).

  • Source conflict: not applicable. The statute says nothing about a partial $100, and the schedule fills that gap.

  • ruff format (0.15.5, the uv.lock pin) and ruff check pass on the changed Python files.

  • Changelog fragment changelog.d/fix-vt-veteran-credit-rounding.fixed.md.

Microsimulation impact

We estimate that this branch changes none of the outputs below in 2026 on the default dataset. We compared main at a93cc58d7e with this branch at 6fdd72d82a.

Output Main total Change Records changed (weighted)
vt_veteran_tax_credit $794,883 $0 0 (0)
vt_income_tax $1.45bn $0 0 (0)
state_income_tax $576.5bn $0 0 (0)
income_tax $2,594.9bn $0 0 (0)
household_benefits $2,060.5bn $0 0 (0)
household_net_income $17,247.1bn $0 0 (0)

No sampled return's credit changes, so the household tests above show what this change does.

axiom: TheAxiomFoundation/rulespec-us#2220 queued (32 V.S.A. § 5830g with the 2025 Schedule IN-112's whole $100 steps; the pinned corpus release lacks the 2025 schedule, so the signed encoder can't run yet)

🤖 Generated with Claude Code

MaxGhenis and others added 2 commits November 13, 2026 07:06
Vermont's 2025 Schedule IN-112, Part II, line 10 has filers divide AGI
above $25,000 by 100, "rounding down to the nearest whole number".
32 V.S.A. § 5830g(c)(2) sets $5 per $100 and says nothing about a
partial $100. The formula rounded up with np.ceil, so a tax unit with
AGI from $25,001 to $29,999, other than a multiple of $100, got $5 less
than the schedule gives. It now uses np.floor.

The baseline YAML file gains boundary cases at each edge of the
worksheet. The variables-folder file now expects $250 at AGI of $25,050
and $5 at $29,950, and four of its cases gain a Vermont state code so
they reach the formula. A new invariants test checks every whole-dollar
AGI from $0 to $35,000 against the worksheet. The five parameter files'
booklet links now open page 34, which holds the credit's instructions.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Put active verbs in the changelog fragment, the invariants test's
docstring and one of its comments. No code or expected value changes.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 13, 2026 22:32
@codecov

codecov Bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (aae7f98) to head (6fdd72d).
⚠️ Report is 265 commits behind head on main.

Additional details and impacted files
@@            Coverage Diff            @@
##              main    #10871   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            1         1           
  Lines           37        23   -14     
=========================================
- Hits            37        23   -14     
Flag Coverage Δ
unittests 100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@MaxGhenis MaxGhenis closed this Nov 14, 2026
@MaxGhenis MaxGhenis reopened this Nov 14, 2026
@MaxGhenis MaxGhenis closed this Nov 14, 2026
@MaxGhenis MaxGhenis reopened this Nov 14, 2026
@MaxGhenis MaxGhenis closed this Nov 14, 2026
@MaxGhenis MaxGhenis reopened this Nov 14, 2026
@MaxGhenis
MaxGhenis merged commit 6da16f7 into main Nov 14, 2026
120 of 124 checks passed
@MaxGhenis
MaxGhenis deleted the fix-vt-veteran-credit-rounding branch November 14, 2026 17:36
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

US + core hub merge audit, us#10871 at 6fdd72d82a45212be3e418e892b6f604c6393bde (squash): Round the Vermont veteran tax credit's phase-out steps down.

  • Reviews:
    • The PR has no GitHub reviews. An independent check ran outside GitHub, tried to refute the fix at 6fdd72d82a, and found no blocking problem.
    • The check fetched 32 V.S.A. § 5830g and § 5830f, Act 71 as passed, and the 2025 Schedule IN-112, its instructions and the 2025 booklet from legislature.vermont.gov and tax.vermont.gov. It confirmed that the statute sets no rule for a partial $100, that line 10 of the schedule rounds the count of $100 steps down, and that Vermont has posted no 2026 schedule.
    • It computed six cases by hand from lines 9 to 12 and ran them on the head and on main for 2025 and 2026. At AGI of $25,001, $26,250, $27,500, $27,501, $28,749 and $29,999, the head pays $250, $190, $125, $125, $65 and $5, as the schedule does. Main's formula paid $5 less at each but $27,500, and the two formulas agreed at every case it ran at or below $25,000 and at or above $30,000.
    • It ran each new test against main's formula. 10 of the 19 new baseline cases fail there, each by $5, and the other nine sit where the two roundings agree. Both changed variables-folder cases fail, and all 3 invariants tests fail, each whole-dollar test at 4,950 of 35,001 AGIs. On the head, the 27 baseline cases, the 12 variables-folder cases and the 3 invariants tests pass.
    • It checked every number and sourced sentence in the description against the code, its own fetches and its runs, and listed one correction. I made it before merge: the $5 gain also reaches fractional AGIs between $25,000 and $25,001 and between $29,999 and $30,000, so the description now states it for every AGI in the band. The description also drops a local CPU-time figure and a sentence about draft CI.
  • Methodology choices for maintainers:
    • The PR reads § 5830g(c)(2) as silent on a partial $100 and lets line 10 of the 2025 Schedule IN-112 fill the gap, so it records no source conflict. The legislature wrote a round-up as "or fraction thereof" in § 5830f(b) and left those words out of § 5830g, so the old np.ceil had no source.
    • A proportional reading, $0.05 for each dollar above $25,000, also fits "$5.00 per $100.00", and Title 32 uses "per $100.00" for a rate in § 5402. Under it, AGI of $26,250 would get $187.50, where the schedule gives $190. A maintainer who reads (c)(2) as a tie between that rate and whole steps, rather than as silence, would still follow the Department of Taxes' whole steps, and would record the choice as a source-conflict candidate.
    • The model applies the 2025 schedule's rounding to 2026. Vermont has posted no 2026 Schedule IN-112, and § 5830g hasn't changed since Act 71 added it.
  • Gates checked live at merge:
  • Impact: we estimate no change in 2026 on the default dataset. vt_veteran_tax_credit totals $794,883 at main a93cc58d7e and at 6fdd72d82a, no sampled return's credit changes, and Vermont, state and federal income tax, household benefits and household net income don't move. The household tests show the change.
  • Next:
    • § 5830g(a) grants the credit to "a resident individual" who served, and the schedule has one block of lines 8 to 12 per return. The model pays one $250 per tax unit, and the new Hypothesis test pins that for a joint return with two veterans. A follow-up can ask the Department of Taxes whether each veteran spouse may claim, and change the formula and the test if so.
    • No Vermont variable prorates refundable credits for part-year residents, as Schedule IN-112 lines 14 to 17 do with the Vermont share of income from Schedule IN-113. The gap predates this PR and covers all four of Vermont's refundable credits, so it needs an issue.
    • Before 2025, income_increment holds 0, so line 45 divides by zero and NumPy warns. where drops those values and the credit stays $0, as it did under np.ceil. A follow-up can skip the division for a zero increment.
  • Axiom: Encode Vermont's veteran tax credit: 32 V.S.A. § 5830g, with Schedule IN-112's whole $100 steps (PE-US #10871) TheAxiomFoundation/rulespec-us#2220 queued, a new pe-parity issue for § 5830g with the schedule's whole $100 steps and companion rows from this PR's YAML. The signed encoder can't run yet: the pinned corpus release carries Vermont's chapter 151 statutes but not the 2025 booklet that holds Schedule IN-112.

🤖 Generated with Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant