Fix North Dakota 2026 income tax thresholds

[PavelMakarchuk]

Summary

• add official TY2026 North Dakota individual income-tax thresholds for every filing status
• pin the complete official schedule in raw parameter assertions
• test the marginal schedule at and immediately above both boundaries for all five statuses

Source

North Dakota Office of State Tax Commissioner, 2026 Form ND-1ES, page 2.

Checks

• make format (includes Ruff formatting and lint)
• uv run pytest -c pyproject.toml policyengine_us/tests/core/test_state_indexed_frozen_parameters.py -k nd -q (32 passed)
• focused ND boundary YAML (37 passed)
• full ND income-tax YAML suite (73 passed)
• uv run pytest -c pyproject.toml policyengine_us/tests/test_parameter_files.py -q (6 passed)
• independent pre-PR review: CLEAN; no actionable findings

[PavelMakarchuk]

@MaxGhenis This is ready for maintainer review: exact head ec61a0344d is now 31/31 green. The sole first-attempt failure was an unrelated Hugging Face CAS download error after five retries in Baseline (rest-b); rerunning that failed job passed. Independent review and all focused ND checks are clean.

[DTrim99]

Program Review — PR #9165 (Fix North Dakota 2026 income tax thresholds)

Author: PavelMakarchuk

Source Documents

• ND Office of State Tax Commissioner, 2026 Form ND-1ES (SFN 28709, 12-2025), page 2 — "2026 Forms ND-1 and ND-EZ Tax Rate Schedules" · PDF #page=2

Branch Status

⚠ 6 commits behind main (1 ahead). Minor; review scoped to the merge-base diff.

Summary

Clean, correct TY2026 value update across all five filing statuses. 0 critical. Every 2026 bracket threshold matches ND-1ES 2026 p.2 exactly, correctly routed to the right status, correctly dated at 2026-01-01 with prior years untouched, the 3-tier scale structure (0% / 1.95% / 2.50%) preserved, and the frozen-parameter test + per-status boundary tests are aligned. CI green (31/31).

PDF / value audit — all 10 thresholds match ND-1ES 2026 p.2

Status	1.95% start	2.50% start	✓
Single	49,575	250,400	✅
MFJ + Qualifying Surviving Spouse	82,800	304,850	✅
Married filing separately	41,400	152,425	✅
Head of household	66,400	277,600	✅

Base-amount cross-checks reproduce the form's printed "your tax is" column exactly ((250,400−49,575)×0.0195 = $3,916.09; MFJ $4,329.98; separate $2,164.99; HoH $4,118.40). MFJ and QSS correctly share one schedule per the form; no status file received another's thresholds; nd_income_tax_before_credits.py routes each status to its own scale. No mismatches.

Critical

None.

Should Address (minor / cosmetic)

1. Reference title wording. The 2026 reference title on each of the five files reads "2026 North Dakota estimated income tax worksheet," but the schedule lives on the Form ND-1ES tax rate schedules page (the frozen-param test docstring correctly calls it that). The #page=2 href is correct and file-page-accurate — only the title wording is slightly off. Consider "2026 North Dakota Form ND-1ES tax rate schedules" for consistency.
2. Pre-existing (not from this PR): joint.yaml carries 0.0250 trailing-zero rate values on the 2023-01-01 top bracket (should be 0.025; the other four files are correct). Since joint.yaml is a touched file, worth a cleanup pass — but not introduced here, and the new 2026 threshold values are all correctly formatted.

Suggestions (optional)

• The 20 new test cases use free-form names rather than the file's existing Test N, … numbered convention (arguably more readable; correctly appended at the bottom).
• The "at/just-above threshold" currency cases assert fractional-cent outputs (e.g. 3_916.0875) with no absolute_error_margin; adding 0.01 would match house style (they pass exactly as-is, so this makes them stricter, not vacuous). A one-line #($1 over × 1.95%) comment would clarify the 0.0195 outputs.

Validation Summary

Check	Result
Regulatory / Value Audit	All 10 thresholds exact-match ND-1ES 2026 p.2; base-amount cross-checks reproduce; MFJ+QSS shared correctly; 0 mismatches
Reference Quality	ND-1ES cited on each of 5 files with correct #page=2; statute subsection preserved; minor title-wording nit
Code Patterns	0 critical; clean 2026-01-01 keys across all 5 statuses, scale structure intact, frozen-param test + changelog aligned; 1 pre-existing trailing-zero
Test Coverage	Tight — all 5 statuses tested at AND just-above both bracket boundaries (20 cases); would catch a transposed threshold or wrong-schedule routing
CI Status	Passing (31/31)

Review Severity: APPROVE

An exemplary, fully-verified 2026 multi-status threshold update: every value matched against ND-1ES 2026 p.2 with base-amount cross-checks, correct MFJ/QSS sharing, and comprehensive per-status boundary tests. No functional issues — only a cosmetic reference-title wording fix and a pre-existing trailing-zero cleanup. Rebase off the 6-commit lag before merge.

Review generated with Claude Code via /review-program

[DTrim99]

Small Claude nits

[PavelMakarchuk]

@DTrim99 review items addressed in f454be4 — re-requesting review.

Applied

1. Reference title (your Should Address Basic prototype #1). All five filing-status files now read 2026 North Dakota Form ND-1ES tax rate schedules instead of "estimated income tax worksheet". The href and #page=2 anchor were already correct and are unchanged.
2. joint.yaml trailing zeros (your Should Address Run black formatting on docs #2). The three 2023-01-01 top-bracket rates go 0.0250 → 0.025, matching the other four files. Also dropped a whitespace-only line in the same file.
3. Branch freshness. The branch was 507 commits behind main (not 6 — it had drifted a lot further since your review). Now merged up to date. The only conflict was additive, in policyengine_us/tests/core/test_state_indexed_frozen_parameters.py: main added a Rhode Island 2026 block while this branch adds a North Dakota one. Both are kept, RI first to match main's ordering.

Not applied, with reasoning — happy to change if you disagree

4. absolute_error_margin: 0.01 on the fractional-cent cases (your Suggestion Run black formatting on docs #2). I added the clarifying comment you asked for but skipped the margin. Two reasons: this file has no absolute_error_margin anywhere — all 22 pre-existing cases assert exact hand-calculated values under a # hand calculated to avoid whole-dollar rounding problems convention — and a margin widens an otherwise-exact comparison rather than tightening it, so it would make these cases slightly weaker, not stronger. Instead of a per-case margin I documented the derivation in a header comment above the TY2026 block (zero-rate boundary → 0; $1 above → $1 × 1.95%; top-rate boundary → (top − zero-rate) × 1.95%; $1 above → base + $1 × 2.50%), plus an inline # $1 over the threshold x 1.95% on the five 0.0195 outputs. Say the word and I'll add the margins.

5. Test naming (your Suggestion Basic prototype #1). Left as-is — the new names are descriptive and appended at the bottom, and renumbering into the Test N, … scheme would churn the whole file.

Verification on the merged head: test_state_indexed_frozen_parameters.py 33 passed (both the RI and ND blocks), full ND YAML suite 322 passed, test_parameter_files.py 7 passed, ruff format --check clean. No parameter values changed in this update — only a reference title, a rate literal's formatting, and comments.

[DTrim99]

Program Review

Source Documents

• PDF: 2026 Form ND-1ES, page 2 (6 pages)
• Year: 2026
• Scope: PR changes only

Critical (Must Fix)

None.

Should Address

None.

Suggestions

1. Add an absolute_error_margin to the household boundary suite — policyengine_us/tests/policy/baseline/gov/states/nd/tax/income/nd_income_tax_before_credits.yaml. The 20 new cases carry 4-decimal expected outputs (e.g. 3_916.0875, 2_164.9875) with no margin, relying on the file default. Float64 evaluation of count * 0.0195 can carry sub-cent representation error. The pre-existing Tests 1–17 use the same no-margin convention and pass in CI, so this is consistent and almost certainly fine; consider absolute_error_margin: 0.01 only if a future rounding change makes exact-float matches brittle. Not blocking.
2. Add a mid-bracket / deep-in-top-bracket 2026 case — every 2026 household case sits within $1 of a threshold. A single deep-in-the-top-bracket point (e.g. single at $500,000 → 3_916.0875 + (500_000 − 250_400) × 0.025 = 10_156.0875) would confirm the 2.5% rate keeps applying far above brackets[2] (with brackets[3]/[4] = inf). The pre-existing TY2023 tests already exercise this shape for prior years, so this is optional depth, not a gap.
3. Optionally harden the frozen-parameter pin — test_state_indexed_frozen_parameters.py::test_nd_2026_income_tax_brackets pins the two indexed breakpoints (brackets[1], brackets[2]) for all five statuses but not the 2026 rate values (0% / 1.95% / 2.5%) or the inf top thresholds (brackets[3]/[4]). This matches the existing NE/ME convention, and the household suite exercises the rates end-to-end, so they are not untested — just not pinned. Asserting them would guard against a future stray rate/uprating edit silently un-freezing the 2026 schedule. Not required.

PDF Audit Summary

Category	Count
Thresholds confirmed correct	10
Mismatches	0
Unmodeled / dropped items	0
Filing statuses covered	5 of 5

All 10 added TY2026 thresholds (2 breakpoints × 5 filing statuses) match the official 2026 Form ND-1ES page 2 exactly. Independent regulatory recomputation confirms both the thresholds and the printed base-tax amounts:

• Single: 49,575 / 250,400 (base 3,916.09)
• MFJ + QSS: 82,800 / 304,850 (base 4,329.98)
• MFS: 41,400 / 152,425 (base 2,164.99)
• HoH: 66,400 / 277,600 (base 4,118.40)

The 0% / 1.95% / 2.50% rate structure is unchanged; the only rate-line edit is a harmless 0.0250 → 0.025 literal normalization in joint.yaml (same value).

Validation Summary

Check	Result
Regulatory Accuracy	PASS — all 5 statuses match Form ND-1ES 2026 / NDCC 57-38-30.3; base-tax hand-recomputed
Reference Quality	PASS — 2026 Form ND-1ES reference added (title + #page=2 href), NDCC sub-citations retained
Code Patterns	PASS — 0 CRITICAL / 0 SHOULD / 3 SUGGESTION; correct period keying, .fixed changelog, value formatting
Test Coverage	PASS — 20 household boundary cases + 5 frozen-parameter assertions; all expected values re-derived
PDF Value Audit	PASS — 10 confirmed / 0 mismatches
CI Status	7 pass / 24 pending / 0 fail (at review time)

Review Severity: APPROVE

Next Steps

To auto-fix issues: /fix-pr 9165

🤖 Generated with Claude Code /review-program

[DTrim99]

Claude nits are optional, lgtm