Count a Montana casualty loss once per joint return and only for the spouse who had it

[MaxGhenis]

Stacked on #9877 (fix-mt-separate-filing-allocations), which moves the child and dependent care deduction out of the same Montana miscellaneous deduction list. Until #9877 merges, this diff includes its commits; the changes reviewed here are the commits after 6eefe271dc. Merge #9877 first.

The bug

mt_misc_deductions is a Person variable whose adds is the parameter list gov.states.mt.tax.income.deductions.itemized.misc_deductions. For 2021–2023 that list included casualty_loss_deduction, a TaxUnit variable. for_each_variable would raise on a person-from-group aggregation, but adds does not go through it. policyengine-core's _run_formula evaluates each adds entry with calculate(var, period, map_to=variable.entity.key), and map_result handles group→person with no how as entity_pop.project(values) (policyengine_core/simulations/simulation.py, _run_formula and map_result). So every member of the tax unit got the full unit casualty deduction:

• the joint path (mt_itemized_deductions_joint and mt_itemized_deductions_for_federal_itemization_joint, which add mt_misc_deductions over the tax unit) counted it once per member: twice for a couple, three times with one child;
• the separate path (mt_itemized_deductions_indiv and its federal-itemization twin) gave each spouse the full amount, whoever had the loss.

Verified on main with a 2022 Montana couple: wages $60,000 and $40,000, and a $30,000 casualty_loss on the head. The casualty deduction has to be switched on, because gov.irs.deductions.itemized.casualty.active is false from 2018:

	casualty_loss_deduction	mt_misc_deductions	mt_itemized_deductions_joint (head)	mt_itemized_deductions_indiv
baseline	0	[0, 0]	8,481	[5,000, 3,392]
casualty.active: true	20,000	[20,000, 20,000]	48,481 (casualty counted twice)	[25,000, 23,392] (full amount to each spouse)
same, plus a child	20,000	[20,000, 20,000, 20,000]	66,481 (three times)

Under current law the bug is latent, because the casualty deduction is suspended for every year the Montana line exists. It changes results under any reform that enables casualty losses for 2021–2023.

The rule

2022 Montana Form 2 instructions, Itemized Deductions Schedule, line 15 (PDF pp. 35–36). The 2021 (PDF p. 34) and 2023 (PDF p. 39) instructions say the same:

You may deduct a casualty and theft loss from a federally declared disaster (other than net qualified disaster losses). If you have a casualty and/or theft loss, complete federal Form 4684 to figure the amount of the Montana loss that you can enter on this line. To determine your casualty and/or theft loss, use your Montana Adjusted Gross Income in place of the Federal Adjusted Gross Income. If you and your spouse are filing separately, you each should complete a separate federal Form 4684 to determine the amount of your separate casualty and/or theft loss.

Deductions for all other casualty and theft losses are suspended until tax years beginning after December 31, 2025.

The same instructions, Filing Separately When Married (PDF p. 11): "Deductions that are attributable to only one spouse must be claimed by that spouse."

Form 4684 (2022) has line 11 ("Enter $100") subtracted from each loss on line 12, and line 17 takes 10% of adjusted gross income. IRS Publication 547 (2022, PDF p. 10) says: "If you and your spouse file a joint return, you are treated as one individual in applying the $100 rule … If you and your spouse have a casualty or theft loss and you file separate returns, each of you must reduce your loss by $100 … If one spouse owns the property, only that spouse can claim a loss deduction on a separate return." The 2024 instructions (PDF p. 8) list the Itemized Deductions Schedule under discontinued schedules.

The fix

• mt_casualty_loss_deduction_joint (TaxUnit) figures the return's loss as on Form 4684, lines 10–18: (the spouses' casualty_loss − $100), less 10% of mt_agi_joint, floored at 0. It is counted once.
• mt_casualty_loss_deduction_indiv (Person) figures each spouse's own: (own loss − $100), less 10% of their own mt_agi_indiv.
• A dependent's loss is left to the dependent's own return on both paths, as the dependent's income already is: adjusted_gross_income_person, and so Montana AGI, leaves it out. Pub. 547 also allows only the owner to claim a loss.
• Both are multiplied by the federal suspension flag gov.irs.deductions.itemized.casualty.active, which the old path also used. Both are gated by the new gov.states.mt.tax.income.deductions.itemized.casualty_loss.applies (true 2021–2023, false from 2024), which replaces the misc list's 2024-01-01: [].
• A new federal parameter, gov.irs.deductions.itemized.casualty.per_casualty_reduction, holds the $100 per-casualty reduction (26 U.S.C. § 165(h)(1); Form 4684 line 11: $100 in 2008, $500 in 2009, $100 from 2010).
• Former MCA § 15-30-2131(1)(a) (MCA 2021), which allowed the items of IRC § 161 including § 165 losses, is cited on the new parameter and variables.
• The four Montana itemized aggregators add the new path-specific variable where they added mt_misc_deductions. After Align Montana separate-filing deductions and the elderly credit's income with the forms #9877, the misc list held only the casualty loss, so the list parameter and mt_misc_deductions are removed. A repo-wide grep finds no other references.

Methodology notes

• PE records one loss amount per person, not separate casualty events. Each return's losses (joint) or each spouse's losses (separate) are treated as one casualty, so the $100 reduction applies once per return or once per spouse with a loss. ar_casualty_loss_deduction_* already does the same.
• PE has no input that distinguishes federally declared disaster losses, which are the only losses line 15 allows. So, like the federal deduction, no personal casualty loss is deductible from 2018 unless gov.irs.deductions.itemized.casualty.active is switched on.
• The federal casualty_loss_deduction does not apply the $100 reduction. That gap only touches 2010–2017, since the deduction is suspended from 2018. It is left for a separate PR; see the follow-ups below.

Tests

• New casualty_loss/mt_casualty_loss_deduction_joint.yaml (9 cases) and …_indiv.yaml (5 cases) cover:

  • losses pooled once on a joint return;
  • the Montana AGI floor rather than federal AGI (Montana AGI $80,000 with federal AGI $100,000);
  • the boundaries at the floor ($10,100 → $0 and $10,101 → $1 on Montana AGI of $100,000);
  • a loss under $100;
  • each spouse's own loss and floor;
  • a working dependent's loss left off both paths (natural inputs, no AGI overrides);
  • the suspension;
  • the 2024 end on both paths. The 2024 separate-path case allows filing status 2a; it fails if the applies gate is removed, which I checked by deleting the gate.

  The headers quote line 15, Form 4684 and Pub. 547.

• Six aggregator cases use natural inputs (wages plus casualty_loss) in the two joint and two separate aggregator files. All six fail on main with the double count and pass here:

  Case	main	This PR
  joint, couple	[40,000, 0]	[19,900, 0]
  joint, couple and child	[60,000, 0, 0]	[19,900, 0, 0]
  separate, couple	[20,000, 20,000]	[23,900, 0]

  The joint cases appear in both mt_itemized_deductions_joint and its federal-itemization twin, and the separate case in both _indiv files. The main column shows main's actual failing outputs, run on upstream/main 4c900b6d3d.

• Existing aggregator tests that used casualty_loss_deduction or a per-person mt_misc_deductions as inputs now use the path-specific variable. The joint "Amount only distributed to the head" case puts its $100 at the tax unit, counted once, so it expects 2,012 where it expected 2,212.

• New tests/unit/test_mt_casualty_loss_deduction_properties.py is a grid property test in the style of test_ar_misc_casualty_deduction_properties.py. The grid has 432 couples, with 0–2 dependents and dependent losses, plus 18 single filers: 1,314 people in 450 tax units, all in one simulation for 2021–2024. It checks these invariants for every grid input:

  • both variants equal the closed-form Form 4684 reference, max(max(loss − 100, 0) − 10% × Montana AGI, 0);
  • joint: the per-unit sum of mt_itemized_deductions_joint equals the unit deduction, whatever the member count, and only the head carries it;
  • separate: each spouse's amount uses only their own loss and AGI, a spouse without a loss gets 0, and dependents get 0;
  • a dependent's loss changes neither path;
  • 0 ≤ deduction ≤ max(loss − 100, 0);
  • the deduction rises with the loss and falls with Montana AGI;
  • for single filers, the joint and separate variants agree;
  • 2024 gives 0.

  Local cost of the committed file, measured on macOS (not Linux CI): 10 tests in 9.5 s, 23.7 s wall. Peak RSS was 2.44 GB, against 13.1 s and 0.94 GB for a collect-only run of the same file. So the file adds about 11 s and about 1.5 GB at peak, mostly for one reform system and the 1,314-person simulation. It runs in the Rest remaining group.

• All of these pass locally, run one file per process: the 6 touched or new MT itemized YAML files (36 cases), 15 more MT tax files (76 cases), the cross-state state_itemized_deductions.yaml and state_standard_deduction.yaml (31 cases), and the code-health checks (parameters.py, variable_names.py, the reference-URL checks, the builtin-sum and cached-array checks, and test_rest_python_groups.py).

Impact

No microsimulation was run, because baseline values cannot change. The new variables are multiplied by gov.irs.deductions.itemized.casualty.active, which is false from 2018, and the Montana line exists only for 2021–2023. The removed path was casualty_loss_deduction, which is p.active * …, so it was also 0 in every year it was listed. Reforms that enable casualty losses for 2021–2023 change, as described above.

Follow-ups (out of scope, filed separately)

• The federal casualty_loss_deduction does not apply the § 165(h)(1) $100 per-casualty reduction (2010–2017 only, while the deduction is active), and it sums dependents' losses. New York's pre-TCJA copy also lacks the $100 reduction and is not gated by active. Both are left for a follow-up PR.

• A repo-wide check of Person variables whose adds pull group-level variables (core projects each onto every member) found five more real double counts. Each reproduces on a household run, and each is left for its own PR:

  • Ohio's insured medical deduction is counted once per member, so Ohio AGI is understated.
  • Illinois AABD/HBWD gives each member the unit's state withholding.
  • Riverside General Relief counts cash and vehicles once per member, and earners deduct each other's withholding.
  • LA General Relief counts TANF once per member.

  MS (ms_agi_adjustments) and WV (wv_senior_citizen_disability_deduction_total_modifications) are already being fixed in other PRs. A code-health guard against new cases of this pattern is also planned.

axiom: TheAxiomFoundation/rulespec-us#1516 queued (former MCA 15-30-2131(1)(a) via 2021–2023 Form 2 line 15 and 26 U.S.C. 165(h)(1), (2), (4)(B); blocked on ingesting the historical Montana statute and Form 2 instructions, as in #1512, and on Max's approval for signed runs)

Review:

• An in-session adversarial review with three lenses (law and sources, code and entities, tests) approved head 9e7c5d162b. Its two minor findings and five nits are addressed in e352ed05bb: the dependent rule, the suspension wording, the MCA citation, the Form 4684 wording, the $100 history, the 2024 gate test and the grid size.
• The independent Subfleet Opus review (review/standard) is pending; results will be posted here.

🤖 Generated with Claude Code

[DTrim99]

PR Review: Count a Montana casualty loss once per joint return and only for the spouse who had it

PR #9888 (author: MaxGhenis, head e352ed05bb). The PR's own changes are its three commits after 6eefe271dc, the round-1 head of #9877. The branch also carries #9877's pre-merge commits and a merge of #9877's branch; #9877 itself was squash-merged to main on 2026-10-12 as b686c9f0e3. #9912 is stacked on this PR and is reviewed separately.

Summary

For 2021–2023, Montana's Itemized Deductions Schedule line 15 allows a casualty and theft loss figured on federal Form 4684, with Montana AGI in place of federal AGI. Spouses filing separately on the same form each complete their own Form 4684.

On main, mt_misc_deductions is a Person variable whose adds list held the TaxUnit variable casualty_loss_deduction. Core projected the unit amount onto every member. So the joint aggregators counted the loss once per member, and the separate aggregators gave each spouse the whole loss, whoever had it.

The PR replaces that path with two variables:

• mt_casualty_loss_deduction_joint (TaxUnit): the spouses' losses less $100, above 10% of mt_agi_joint.
• mt_casualty_loss_deduction_indiv (Person): each spouse's own loss less $100, above 10% of their own mt_agi_indiv.

Both leave dependents' losses out. Both are gated by a new casualty_loss.applies parameter (2021–2023) and by the federal casualty.active flag. The misc list and mt_misc_deductions are deleted, and the four Montana itemized aggregators add the path-specific variable instead.

The law and code are right. The stage-A source audit found no critical problems, and I re-checked the code independently.

• All 36 YAML cases in the six touched files pass on the extracted head tree, and so do all 10 property tests.
• Twelve expectations check by hand.
• CI is 35/35 green.
• The axiom issue is open, labelled pe-parity and dispatch-ready.
• Under current law the PR changes no baseline value: casualty.active is false from 2018, and the Montana line exists only for 2021–2023. Results change only under reforms that enable casualty losses.

The branch cannot merge yet. It is 189 commits behind main, and GitHub reports it CONFLICTING. A rebase is more than mechanical (S1):

• Two files on main reference the deleted mt_misc_deductions and will fail CI, but have no textual conflict to flag them.
• The branch still holds Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's pre-merge commits. If the conflicts are resolved by taking the branch's side, Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's later changes would be silently reverted.

There is one should-address item and five optional suggestions.

Critical (must fix)

None.

Should address

S1. Rebase onto main by replaying only this PR's own commits, then clear main's three newer uses of the deleted mt_misc_deductions.

• Where (on main):

  1. policyengine_us/tests/code_health/test_no_group_variables_in_person_adds.py:114-115. Guard against person variables that add group variables (every member gets the whole group amount) #9897 added # PolicyEngine/policyengine-us#9888. and ("mt_misc_deductions", "adds", "casualty_loss_deduction") to KNOWN_VIOLATIONS.
  2. policyengine_us/tests/core/test_mt_mortgage_interest_federal_limit_properties.py:54. Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004 lists "mt_misc_deductions" in ZERO_PERSON_COMPONENTS, and every person in its grid gets that variable as an input.
  3. Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's mortgage-interest YAML cases set mt_misc_deductions: 0:
     • mt_itemized_deductions_joint.yaml and federal_itemization/mt_itemized_deductions_for_federal_itemization_joint.yaml: lines 86, 91, 120, 125 and 153;
     • mt_itemized_deductions_indiv.yaml and federal_itemization/mt_itemized_deductions_for_federal_itemization_indiv.yaml: lines 86 and 93.

• What's wrong:

  • Item 1 is a ratchet. Its docstring says "KNOWN_VIOLATIONS holds open bugs. Fixing one makes this file fail until its entry is deleted." Once the variable is gone, test_known_violations_only_shrink fails with "No longer projected; delete these from KNOWN_VIOLATIONS".
  • Item 2 would raise on an unknown input variable.
  • Items 1 and 2 merge with no textual conflict, so resolving only the conflicted files still fails CI.
  • Item 3 sits inside the conflict hunks. Keeping both sides puts a deleted variable into the test inputs.

• The Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 hazard: the branch holds Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 at round 1. main has Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's later commits, which:

  • move the care deduction from head_or_spouse * 0.5 to spouse_allocation_rate;
  • move the charity split in mt_itemized_deductions_for_federal_itemization_indiv.py from * 0.5 to spouse_allocation_rate;
  • add Case 29 and mt_childcare_itemization_boundary.yaml.

  Taking the branch's side of those conflicted files, together with their YAML, would revert all of that, and CI would stay green.

• What survives a clean rebase: I replayed the PR's own diff onto main with a three-way merge based at 6eefe271dc. Only 4 files conflict, not 17: the four aggregator YAML files.

  • The four aggregator .py files auto-merge to mt_casualty_loss_deduction_*. They keep main's deductible_mortgage_interest and spouse_allocation_rate.
  • The deletion of misc_deductions.yaml applies cleanly.
  • per_casualty_reduction.yaml is already on main, byte-identical (Allow non-disaster casualty and theft losses in California and Minnesota itemized deductions #10068), so it drops out of the diff.
  • Nothing that the PR's expectations rely on changed on main: mt_agi_joint, mt_agi_indiv, adjusted_gross_income_person, tax_unit_non_dep_add and the casualty parameters.
  • So the PR's intent survives a straightforward resolution.
  • I could not run the merged tree locally: current main imports a policyengine-core helper that the local 3.26 install lacks. CI on the rebased branch is the check.

• Fix:

  1. Rebase 95d94db81e, 9220847cfe and e352ed05bb onto main. Drop Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's commits, the merge 9e7c5d162b, and Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's changelog fragment.
  2. In the four YAML conflicts, keep both sides' cases. In Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's mortgage cases, delete the mt_misc_deductions: 0 lines.
     • The casualty variables are already 0 without a casualty_loss input, so no replacement is needed.
     • In the two _indiv files, git interleaves Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004's mortgage case and the PR's casualty case line by line. Rebuild each case whole.
  3. Delete the KNOWN_VIOLATIONS entry and its comment (lines 114-115).
  4. Remove "mt_misc_deductions" from ZERO_PERSON_COMPONENTS (line 54).
  5. Update the PR body:
     • drop "Stacked on Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 … Merge Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 first";
     • stop calling per_casualty_reduction "a new federal parameter", because it is now Allow non-disaster casualty and theft losses in California and Minnesota itemized deductions #10068's file;
     • remove or fill "The independent Subfleet Opus review … is pending; results will be posted here".
  6. Then update Apply the Form 4684 $100 per-casualty reduction and drop dependents' losses in the federal, NY, AL and HI casualty deductions #9912, which is stacked on this branch.

• Why this is a should and not just a branch-status note: Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's round-2 conflict was a single append-append hunk. Here, the obvious resolution either fails CI in two files nothing points to, or silently undoes a merged PR. I'd like to see the rebased branch green before approving.

Suggestions

All five are optional.

1. State the joint path's one-casualty convention in a documentation string.

• Where: mt_casualty_loss_deduction_joint.py:34-42. The separate path makes the same per-column choice at mt_casualty_loss_deduction_indiv.py:44-46.
• Law: Pub. 547 (2022), PDF p. 10: "If you have more than one casualty or theft loss during your tax year, you must reduce each loss by $100." Its example has two storms with losses of $75 and $90, and "you aren't entitled to any deduction for these storms."
• Effect (TY2022, active on): a couple with zero income, a head loss of $80 and a spouse loss of $90 from different storms.
  • The law gives $0.
  • The joint path gives $70 = 170 − 100.
  • The separate path gives [0, 0].
• Why it is not blocking: PE has one casualty_loss per person and no casualty count. The code comments and the PR body say so, and AR, CA and MN on main use the same convention. A documentation string would surface it in the API metadata.

2. Baseline gives $0 for 2021–2023 even for federally declared disaster losses (documented deferral).

• Law: line 15 allows "a casualty and theft loss from a federally declared disaster" (2022 Form 2 instructions, PDF p. 35). § 165(h)(5)(A) allows the same losses federally.
• Model: both variables are multiplied by gov.irs.deductions.itemized.casualty.active, which is false from 2018. This keeps the prior behavior, because the removed path read casualty_loss_deduction, which is also p.active * …. It is documented in both formulas and the PR body.
• Follow-up: a person-level federally declared disaster loss input would fix the federal deduction and every state that inherits § 165(h)(5).

3. Optional readable YAML cases for edges that only the property test covers.

Case	Inputs	Expected (head)
Only the spouse had a loss (2022)	wages 90,000 / 20,000; spouse casualty_loss 12,000; active on	mt_casualty_loss_deduction_indiv [0, 9,900] = 12,000 − 100 − 2,000; mt_casualty_loss_deduction_joint 900 = 12,000 − 100 − 11,000
A year other than 2022 (2023)	the joint YAML's first case with period: 2023	19,900; and [23,900, 0] for the first indiv case

• All casualty YAML cases run in 2022 or 2024. The property test covers 2021 and 2023, and covers spouse-only losses through its grid.
• Stage A reproduced both rows on the head tree.

4. Source hygiene.

• PR body: it says the 2021 (PDF p. 34) and 2023 (PDF p. 39) instructions "say the same" as 2022.
  • The 2023 text is identical.
  • The 2021 text reads "a casualty loss and theft from a federally declared disaster", and its suspension sentence ends "after December 31, 2026" (probably a typo for 2025).
  • rulespec-us#1516 already records this. It does not affect the model.
• References (optional, when rebasing): the variables cite the 2022 instructions as separate entries per page: #page=35, #page=36 and #page=11 in _indiv (lines 16-19), and #page=35 and #page=36 in _joint (lines 14-15).
  • That followed the per-page rule when the PR was written.
  • main's CLAUDE.md (Document the reference page-citation convention #10009, 2026-10-07) now asks for one entry per source document, with a # PDF pages 11, 35-36 comment above the href.
  • Either form is acceptable here.

5. Itemized Deductions Schedule lines 16–18 have no hook once the misc list is gone (pre-existing, out of scope).

• Line 16, political contributions: "limited to a total of $100 for yourself and $100 for your spouse".
• Line 17, gambling losses: "allowed only to the extent of the amount of gambling winnings" (2023 instructions, PDF p. 39).
• Line 18, other miscellaneous deductions.
• PE already has a gambling_losses input. The PR did not cause the gap, and deleting the list is fine. File a follow-up issue if wanted.

Verified correct

• Diagnosis and fix, reproduced by stage A (TY2022; wages $60,000 / $40,000; $30,000 loss on the head; active on).

  	Parent 6eefe271dc	Head
  Casualty amount per member	mt_misc_deductions [20,000, 20,000]	joint 19,900, counted once
  mt_itemized_deductions_joint (head)	48,481 (66,481 with a child)	28,381 = 8,481 + 19,900, with or without a child
  mt_itemized_deductions_indiv	[25,000, 23,392.40]	[28,900, 3,392.40]

  With active off, both paths give 0.

• Law (stage A, quotes checked at the anchored pages):

  • Former MCA 15-30-2131(1)(a) (MCA 2021) allows "the items referred to in sections 161 … and 211 of the Internal Revenue Code", and is "Repealed effective January 1, 2024".
  • Form 2 line 15 is in the 2021 instructions at PDF p. 34, 2022 at pp. 35–36 and 2023 at p. 39.
  • 2022 PDF p. 11: "Deductions that are attributable to only one spouse must be claimed by that spouse."
  • 2024 PDF p. 8 lists the discontinued schedule.
  • Form 4684 lines 11-12 and 17-18, and § 165(h)(1), (2), (3)(B) and (4)(B).
  • Pub. 547 p. 10 states the joint "one individual" rule and the separate-return and owner rules.

• Code patterns (every changed file):

  • Entities: TaxUnit _joint and Person _indiv, matching mt_medical_expense_deduction_joint/_indiv. Names follow the Montana _joint/_indiv convention.
  • defined_for: _indiv uses mt_married_filing_separately_on_same_return_eligible and _joint uses StateCode.MT, the same as their aggregators.
  • No hard-coded values: the $100 comes from per_casualty_reduction and the 10% from casualty.floor. REDUCTION/FLOOR appear only in the test's independent reference.
  • Vectorized: the formulas use max_ and role masks. if not p.casualty_loss.applies: return 0 gates on a period-level scalar parameter; 13 state formulas on main use the same early return.
  • The floor can't go negative: mt_agi_joint and mt_agi_indiv are both floored at 0 (max_(agi + additions - subtractions, 0)), so the 10% floor never increases the deduction.
  • Dependents: they are left out through tax_unit_non_dep_add (joint) and is_tax_unit_head_or_spouse (indiv). This is consistent with adjusted_gross_income_person and with CA/MN on main.
  • No dead code: there is no leftover reference to mt_misc_deductions or the misc list at the head.

• Parameters:

  • casualty_loss/applies.yaml follows the boolean template ("…, if this is true."), with unit: bool, period: year, values true 2021 and false 2024, and five single-URL references that corroborate.
  • per_casualty_reduction.yaml (100 / 500 / 100 from 2008 / 2009 / 2010) is byte-identical to main's.

• Callers at the head:

  • The two new variables feed only the four aggregators.
  • The aggregators feed mt_tax_unit_itemizes, mt_deductions_indiv, mt_taxable_income_joint, state_itemized_deductions and the MFS generator, and their interfaces are unchanged.
  • The federal casualty_loss_deduction, casualty_loss, casualty.floor and casualty.active are untouched.
  • The other casualty users (AL, AR, CA, DC, DE, HI, MN, NY) are unaffected. CA and MN read the same per_casualty_reduction.

• Tests run on the extracted head tree, one file per process (local policyengine-core 3.26):

  • mt_casualty_loss_deduction_joint.yaml 9/9 and mt_casualty_loss_deduction_indiv.yaml 5/5;
  • the aggregator files: mt_itemized_deductions_joint 6/6, _indiv 5/5, …_for_federal_itemization_joint 6/6 and _indiv 5/5;
  • tests/unit/test_mt_casualty_loss_deduction_properties.py 10/10 in 90 s, of which 60 s is the module fixture (model load plus the 1,314-person simulation).

• The property test works:

  • It uses a fixed grid with no randomness or hypothesis, so no derandomization is needed, and no importorskip is needed.
  • Each test guards against a vacuous pass (max() > 0).
  • The joint per-unit sum and the own-loss-only checks would fail on the old mt_misc_deductions projection.
  • It is collected by the Rest remaining group. The PR body reports cost measured on macOS, as main's CLAUDE.md asks.

• Test hygiene: test periods are plain (2022, 2024). The PR adds two parameter-only override combos to the Montana tree (active, and active with 2a allowed). main's batcher weights those at 0.25 each against a 5.0 budget, so the tree goes from 1.0 to 1.5.

• Hand checks (12):

  1. Joint: (30,000 − 100) − 10% × 100,000 = 19,900.
  2. Pooled: (35,000 − 100) − 10,000 = 24,900, one $100.
  3. Montana AGI floor: 29,900 − 8,000 = 21,900; a federal-AGI floor would give 19,900.
  4. At the floor: 10,100 − 100 − 10,000 = 0.
  5. One dollar over the floor: 10,101 → 1.
  6. Under $100: max(80 − 100, 0) = 0.
  7. Joint with a dependent: Montana AGI is 100,000 (the dependent's $5,000 wages are excluded), giving 19,900. Counting the dependent's loss would give 29,900.
  8. Indiv: [29,900 − 6,000, 0] = [23,900, 0].
  9. Indiv, both spouses: [23,900, 4,900 − 4,000] = [23,900, 900].
  10. Indiv with a dependent: [23,900, 0, 0]. The dependent's own column would otherwise be 9,900.
  11. "Amount only distributed to the head": 100 interest + 100 casualty + 412 medical + 800 SALT + 600 federal tax = 2,012. The old 2,212 counted the misc 100 three times.
  12. Natural-input aggregator cases: joint [19,900, 0] and [19,900, 0, 0]; indiv [23,900, 0]. Montana AGI is computed from wages. The PR body shows main at [40,000, 0], [60,000, 0, 0] and [20,000, 20,000].

• Changelog: changelog.d/fix-mt-casualty-loss-misc-deduction.fixed.md is a top-level file of type fixed, and its sentence matches the change.

• Partner tests: nothing under tests/policy/baseline/partners/** is touched, and no partner file sets casualty_loss or mt_misc_deductions. Household API Partners passed.

Axiom line assessment

The PR body reads axiom: TheAxiomFoundation/rulespec-us#1516 queued (former MCA 15-30-2131(1)(a) via 2021–2023 Form 2 line 15 and 26 U.S.C. 165(h)(1), (2), (4)(B); blocked on ingesting the historical Montana statute and Form 2 instructions …). The line is valid.

• State: or_income_tax_after_refundable_credits -> or_income_tax #1516 is OPEN and labelled pe-parity, last updated 2026-10-06T16:38:59Z (after the head commit).
• Dispatch-ready on the model of Add AMT parameters for 2020-2023 #1467:
  • module paths us/statutes/26/165/h.yaml and us-mt/statutes/15-30-2131/1/a.yaml;
  • federal corpus citations re-pinned to the signed release scope;
  • verbatim MCA, line 15, p. 11, 2024 p. 8, § 165(h), Form 4684 and Pub. 547;
  • required outputs;
  • companion tests derived from Form 4684 arithmetic, not PolicyEngine output.
• Blocker: the line and the issue both record the blocker (ingesting the Montana corpus), as main's CLAUDE.md now requires for queued. This is the same arrangement the Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 review accepted for Correct MA income tax Part A AGI calculation #1512.
• Match: the companion table matches the head's 9 + 5 cases. Two lines are stale but harmless: "stacked on Align Montana separate-filing deductions and the elderly credit's income with the forms #9877", and Apply the Form 4684 $100 per-casualty reduction and drop dependents' losses in the federal, NY, AL and HI casualty deductions #9912 reusing per_casualty_reduction "from Count a Montana casualty loss once per joint return and only for the spouse who had it #9888" (it is on main via Allow non-disaster casualty and theft losses in California and Minnesota itemized deductions #10068).

CI status

• All 35 checks pass at e352ed05bb (run 37470564352, 2026-10-06 13:24 UTC). That covers every Baseline and Contrib shard, Rest (Python + variables), Microsimulation, Household API Partners, Lint, the changelog check and the smoke imports.
• The run predates Guard against person variables that add group variables (every member gets the whole group amount) #9897, Count only deductible mortgage interest in Montana, give Maryland spouses their own losses, report the whole Virginia EITC #10004, Allow non-disaster casualty and theft losses in California and Minnesota itemized deductions #10068 and Align Montana separate-filing deductions and the elderly credit's income with the forms #9877 landing on main, so it has not tested the merge. It will fail on the two silent references in S1 until they are removed.

Branch status

• The branch is 189 commits behind main and 7 ahead. The merge-base is bab1a47c60. GitHub reports CONFLICTING / DIRTY.

• git merge-tree --write-tree PolicyEngine/main e352ed05bb reports 17 conflicted files:

  • 1 modify/delete: misc_deductions.yaml, which Count refundable credits received in Montana's elderly homeowner/renter credit income #9915 edited on main;
  • 2 add/add: Align Montana separate-filing deductions and the elderly credit's income with the forms #9877's changelog fragment and mt_federal_income_tax_deduction_share.py;
  • 14 content conflicts.

  Eight of the 17 are Align Montana separate-filing deductions and the elderly credit's income with the forms #9877-only files, where main must win. In the four aggregator .py files, the conflict is one line, mt_misc_deductions against mt_casualty_loss_deduction_*, and the PR's side is right.

• Replaying only the PR's three commits leaves 4 conflicts (the aggregator YAML files), resolved by keeping both sides as described in S1. The two silent breakages in S1 still need edits.

• Before merge: rebase as in S1, let CI re-run, then update Apply the Form 4684 $100 per-casualty reduction and drop dependents' losses in the federal, NY, AL and HI casualty deductions #9912.

Verdict: REQUEST_CHANGES

[DTrim99]

The law and code are right. The new joint/indiv casualty deductions apply Form 4684 with Montana AGI ($100, then the 10% floor), leave dependents out, cover 2021-2023 only, and replace the person-level projection. The tests pass and rulespec-us#1516 is dispatch-ready. One should item (see the comment above): the branch is 189 behind and conflicting in 17 files. Rebase only this PR's three commits, don't take the branch's side of #9877's files (that would silently revert spouse_allocation_rate), and fix the two non-textual CI breaks (main's KNOWN_VIOLATIONS entry and the mt_misc_deductions name in the mortgage-interest property test). Then update the PR body and #9912.