Skip to content

Record Colorado's open AMT deduction addback question - #10832

Merged
MaxGhenis merged 3 commits into
mainfrom
co-amt-duplication-bar
Nov 15, 2026
Merged

MaxGhenis merged 3 commits into
mainfrom
co-amt-duplication-bar

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Record Colorado's AMT deduction addback question as open and preserve current formulas and parameters. The project lead still needs to decide whether filers should reduce the Colorado deduction addback for amounts already disallowed for federal AMT.

Under C.R.S. 39-22-105(2)(a), filers start with federal AMTI less the federal exemption and apply the section 39-22-104 modifications. In the same paragraph, the General Assembly bars filers from adding back state or local bond interest that federal AMTI already includes.

The Department of Revenue directs filers to exclude duplicate federal adjustments in the 2025 DR 0104AMT instructions, page 1. For federal AMT, filers cannot deduct the covered taxes or the standard deduction under 26 U.S.C. 56(b)(1)(A)(ii) and (D).

The code excludes the state income tax addback from Colorado AMTI but carries the federal deduction addback from regular-return deductions. For that regular-return addback, the Department coordinates state income taxes under Rule 39-22-104(3)(p), paragraph (3), and Rule 39-22-104(3)(p.5), paragraph (3)(a)(i). The code continues the net calculation in 2026.

The captures contain the complete first pages of the 2024 and 2025 instructions. We extracted the 2025 page with pdftotext from the official PDF (SHA256 73d9189d753e1a758acef66a186cc40bdfa3912a3ae1e084c7c3d667fba576f0), and the capture's text below its header matches that output byte for byte. The 2024 capture matches pdftotext's output for page 1 of the official 2024 PDF (SHA256 8a355627c8be22007336eb49091abe7bd42387fbab9167c2bafabda6908392e9) after whitespace normalization.

Hand computations

All four single filers earn $450,000. The itemizers deduct $5,000 of state income tax, $5,000 of property tax and $20,000 of mortgage interest, totaling $30,000.

The IRS publishes the 2025 AMT exemption, 2025 standard deduction, 2026 AMT exemption, and 2026 standard deduction. Colorado's deduction limits appear in C.R.S. 39-22-104(3)(p.5) and (p.7), pages 382-385 and the Department's January 2026 guide, page 8.

Filer and year Federal AMTI less exemption Current deduction addback Current Colorado AMTI AMTI after removing overlap
Itemizer, 2025 430,000 - 88,100 = 341,900 30,000 - 12,000 - 5,000 = 13,000 354,900 349,900
Itemizer, 2026 430,000 - 90,100 = 339,900 30,000 - 1,000 - 5,000 = 24,000 363,900 358,900
Standard deduction, 2025 450,000 - 88,100 = 361,900 15,750 - 12,000 = 3,750 365,650 361,900
Standard deduction, 2026 450,000 - 90,100 = 359,900 16,100 - 1,000 = 15,100 375,000 359,900

Normal tax equals 438,000 × 4.4% = 19,272 in 2025 and 449,000 × 4.4% = 19,756 in 2026. Multiplying each current Colorado AMTI by 3.47% gives tentative minimum tax below normal tax. All four filers owe zero Colorado AMT under either reading.

Invariants

Preserve the formulas, parameter values and numerical test assertions. The AMTI assertions distinguish the two readings; the zero-AMT assertions check integration with normal tax. The metadata check keeps the pending candidate open and defers to the shared register after registration. The capture check requires the 2025 capture's source header, operative prohibition and page context.

Validation

Pull request run 43741551793 at c02750ebef passed all 37 jobs on GitHub's merge with main f28e759188.

  • Colorado AMTI YAML: 10 passed with the locked Core version, 3.34.2.
  • Source-conflict guard: 105 passed, including both new regressions.
  • Each new guard test fails on the earlier state: state: ruled fails the metadata test, and the one-sentence excerpt fails the capture test.
  • Against main, the YAML adds two standard-deduction cases and a co_amt: 0 assertion on each itemizer case, and it keeps every existing input, expected value and tolerance.
  • Ruff format and lint leave the changed files unchanged. git diff --check passed.

Methodology choices

  • Record the question as the open candidate us-co:dr0104amt-line-2-duplication-bar, with follows: unknown, and leave the reading to the project lead. Preserve the current treatment of the state tax and federal deduction addbacks pending a ruling.
  • Record the code's scope from 2022 for itemizers and 2023 for standard deduction filers. The 2025 capture matches the official PDF's text byte for byte, and the capture of the 2024 AMT instructions matches after whitespace normalization; we have not verified the 2022, 2023 or 2026 AMT forms.
  • Attribute the regular-return state-tax reduction to the Department's rules. Continue the existing 2026 net calculation without adopting a new policy reading.
  • Add no source-conflict marker until the shared register pins an implementation, as CONTRIBUTING requires.

Microsimulation impact

This PR changes no formula or parameter and keeps every existing test expectation, so no household's result changes.

axiom: TheAxiomFoundation/rulespec-us#2034 queued (C.R.S. 39-22-105(2)(a) with the section 39-22-104 modifications and the DR 0104AMT duplicate-adjustment instruction; the corpus lacks the 2024 and 2025 DR 0104AMT instructions, so the signed encoder can't run yet)

🤖 Generated with Claude Code

MaxGhenis and others added 3 commits November 12, 2026 00:52
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Preserve current calculations pending a ruling. Capture complete instructions and add regressions for the candidate status and source context.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis MaxGhenis changed the title Record Colorado's AMT deduction addback reading Record Colorado's open AMT deduction addback question Nov 15, 2026
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 15, 2026 17:29
@MaxGhenis MaxGhenis closed this Nov 15, 2026
@MaxGhenis MaxGhenis reopened this Nov 15, 2026
@MaxGhenis MaxGhenis closed this Nov 15, 2026
@MaxGhenis MaxGhenis reopened this Nov 15, 2026
@MaxGhenis
MaxGhenis merged commit 5f31258 into main Nov 15, 2026
49 of 71 checks passed
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

US + core hub merge audit, us#10832 at c02750ebef5579fbb5db8ae32b72ea9bce7e56fc (squash): Record Colorado's open AMT deduction addback question.

  • Reviews:
    • The PR has no GitHub reviews. Two independent review rounds ran outside GitHub, with a fix round between them.
    • The first round, at 07605378c5, read C.R.S. 39-22-105(2)(a) and 39-22-104(3)(p.5) and (p.7), the coordination rules of 1 CCR 201-2, 26 U.S.C. 56(b)(1), the 2025 DR 0104AMT instructions, Rev. Proc. 2025-32 and the 2025 Form 6251. Eight households gave the same results at that head and at the merge base 935b24ff4f. The 10 YAML cases and 103 register guard tests passed, and a mutant that took the overlap out of line 2 failed the four end-to-end AMTI cases.
    • It asked for changes. The candidate carried state: ruled and follows: statute, though 39-22-105(2)(a) says nothing on this overlap and the formula already measures the (3)(d) addback against the federal AMT base. The capture held one sentence fragment that left out the operative "do not enter". It also asked to date the candidate from 2022, to credit the state income tax netting to the Department's rules, to note that the co_amt assertions pass under both readings, and to state the impact.
    • The fix round, at c02750ebef, set the candidate to state: open and follows: unknown and described both readings. It replaced the fragment with the full first pages of the 2024 and 2025 instructions, dated the candidate from 2022 and cited the rules for the netting. It added two guard tests, one for the candidate's state and one for the 2025 capture. It changed no formula, parameter or expected value.
    • The second round approved c02750ebef with no blocking finding. It read the full diff and the diff since 07605378c5, and every input and earlier expected value in the YAML matched the merge base. It recomputed all four filers by hand and matched the 2025 capture to pdftotext's output byte for byte. In the fix round's logs, 10 YAML cases and 105 guard tests passed, and each new guard test failed when a mutant restored the earlier state. It read the statute, the rules and the federal sources on government sites, and its merge of the branch into main showed no conflict.
    • It listed four minor findings. The Department's site refused its downloads, so it could not check the 2024 capture. The candidate's example still says "under the retained reading", and its note asks whether the bond-interest exception "extends to" this overlap. years.to: 2026 stops short of the later years the code covers. The new state test checks one candidate and returns once the register holds the entry. It also asked for a CI run at the head, since CI had skipped the draft; the run under the gates below supplies it.
    • On November 15 I downloaded the 2024 PDF from the Department's site (SHA256 8a355627c8be22007336eb49091abe7bd42387fbab9167c2bafabda6908392e9). The 2024 capture matches pdftotext's output for its page 1 after whitespace normalization.
    • The second round listed corrections to the description, and I made each before merge, with the 2024 check stated as it now stands. The description also states the impact, names the open candidate and uses an accepted axiom: form.
  • Methodology choices for maintainers:
    • The PR keeps main's formula and records its reading as an open question, us-co:dr0104amt-line-2-duplication-bar, with state: open and follows: unknown. The project lead has not answered it.
    • In its DR 0104AMT instructions, the Department of Revenue tells filers not to enter on line 2 or line 4 any amount that "duplicates any income or adjustment made in arriving at federal alternative minimum taxable income". The model leaves the state income tax addback out of line 2 and carries the whole federal deductions addback of 39-22-104(3)(p), (p.5) and (p.7) from the regular return. Federal AMTI already restores an itemizer's taxes and the whole standard deduction under 26 U.S.C. 56(b)(1)(A)(ii) and (D).
    • If that sentence reaches the deductions addback, Colorado AMTI falls by the overlap. The itemizer cases would expect 349,900 and 358,900 in place of 354,900 and 363,900, and the standard deduction cases 361,900 and 359,900 in place of 365,650 and 375,000. In all four cases tentative minimum tax stays under normal tax, so each filer owes $0 of Colorado AMT under both readings and no test shows a tax difference.
    • Under 39-22-105(2)(a), filers apply the section 39-22-104 modifications to federal AMTI less the exemption, and the paragraph says nothing on this overlap. The formula measures the (3)(d) state income tax addback against the federal AMT base, where it comes to zero, and takes the deductions addback from regular-return deductions.
    • The model nets the state income tax addback out of the deductions addback in 2026 as it does for 2022 to 2025. Rules 39-22-104(3)(p) and (3)(p.5) cover those years, and no rule covers (p.7) yet. Main already nets it, and the 2026 cases keep that.
    • The candidate dates the question from 2022 to 2026. The code applies the calculation to itemizers from 2022 and to standard deduction filers from 2023, and it carries the 2026 limits into later years.
    • The PR adds no marker comment. Under CONTRIBUTING, a marker follows only after the shared register lists the file.
  • Gates checked live at merge:
    • gh pr checks exits 0;
    • the latest Pull request run at the head, 43741551793, passed and tested this head merged into main at f28e759188;
    • MERGEABLE;
    • not a draft;
    • no reviewer's latest review requests changes;
    • the semantic-overlap guard: since f28e759188, main has changed no area this PR touches. Between the PR's merge base 935b24ff4f and main at c9f921640c, main changed two of the six files this PR touches. In source_conflicts/candidates.yaml it added 51 candidates and edited 20 others, among them one Colorado entry, us-co:premium-assistance-ptc-recipients. In test_source_conflicts.py it added eight KNOWN_UNMARKED lines and no test. Main changed no Colorado income tax file. Project statutory amounts from their base year with base-mode uprating #10783 moved the federal standard deduction to base-mode uprating and kept the 2025 and 2026 amounts that the standard deduction cases read. The branch merges into main at c63279b99d without conflict. The merged backlog holds main's 480 candidates unchanged plus this PR's entry, and the merged test file holds main's 23 tests plus the two new ones;
    • the merge pins the head.
  • Impact: none: the PR changes no formula or parameter and keeps every existing test expectation, so no household's result changes.
  • Next:
    • us-co:dr0104amt-line-2-duplication-bar stays open for the project lead: does the duplication sentence of the DR 0104AMT instructions reduce the federal deductions addback in Colorado AMTI? A ruling PR then keeps or changes line 2 and the four AMTI expectations.
    • The candidate doesn't cite the Department's January 2026 Individual Income Tax Guide. Its AMT steps, on PDF page 6, apply the Colorado additions and subtractions "(except for state income tax addback and the state income tax refund subtraction)".
    • No case expects Colorado AMT above $0. A case with tentative minimum tax above normal tax would show the tax difference between the readings.
    • The 2024 capture's header names "the web PDF text reader" and no hash, and its line breaks differ from pdftotext's. The captures' README bars edits once an entry pins a file, so re-extract page 1 with pdftotext before the entry goes into the rulespec-us register.
    • The ruling PR can tidy the candidate: its example says "under the retained reading", and its note should ask whether the bond-interest exception rules out other overlap relief. It can also drop years.to: 2026, since the code carries the 2026 limits forward.
    • test_co_amt_duplication_candidate_requires_a_ruling fails if a PR records a ruling in the backlog before the register holds the entry. The ruling PR should delete or rename it.
    • Colorado's AMT variables and deductions addback files still cite law.justia.com for C.R.S. 39-22-104 and 39-22-105. A follow-up can cite the olls.info compilation that the candidate uses.
  • Axiom: TheAxiomFoundation/rulespec-us#2034 queued. A comment at c02750ebef pins the two itemizer rows to this head and adds the two standard deduction rows, each with PE-US's value and the value under the other reading. The issue's blocker stands: the corpus holds the Department's DR 0104AMT web page and neither year's instructions PDF, and the release that rulespec-us pins carries no Colorado form or guidance scope.

🤖 Generated with Claude Code

@MaxGhenis
MaxGhenis deleted the co-amt-duplication-bar branch November 15, 2026 19:49
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

US + core hub merge audit, us#10832 at c02750ebef5579fbb5db8ae32b72ea9bce7e56fc: supplement on one cancelled check.

  • At merge, the gate left out one cancelled check-run, 132895452578, named "Python tests / Full Suite - Rest (Python shard ${{ matrix.shard }})". A close and reopen to restart CI cancelled run 43741381717 before its shard plan finished, so the matrix never expanded. That job had no runner and no steps.
  • Run 43741551793 then tested the same head on the same event. Its pr.yaml, python-tests.yaml and rest_python_shards.py match the cancelled run's, file for file.
  • In that run the shard plan, all 10 Rest Python shards and "Check Rest Python shards" passed. That last job fails unless the plan and every shard succeed, and python-tests.yaml sets no continue-on-error. So the job behind the cancelled check ran and passed.
  • The merge audit above did not record this check.

🤖 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