Keep capital gains and losses out of Massachusetts Part B income

[PavelMakarchuk]

Fixes #9953. Fixes #9957.

What changes

Capital gain netting (#9953). ma_part_b_gross_income now subtracts from ma_gross_income the amounts that irs_gross_income added for dividends and capital gains: each head's and spouse's max(0, dividend_income), max(0, capital_gains) and max(0, non_sch_d_capital_gains). Before, it subtracted ma_part_a_gross_income and ma_part_c_gross_income, which net capital losses the Schedule B and D way, so the difference stayed in Part B:

• A long-term loss against a short-term gain reduced Part B, although ma_part_a_agi already applies it on Schedule B, line 25.
• One spouse's short-term loss against the other spouse's long-term gain added phantom Part B income.

M.G.L. c. 62 § 2(b) puts all capital gain income in Part A or Part C, and § 2(d)(1)(M) disallows the IRC § 62(a)(3) loss deduction against Part B.

Capital gain distributions (#9957). non_sch_d_capital_gains (Form 1040 line 7 without a federal Schedule D) was in neither Part A nor Part C, so it stayed in Part B. Massachusetts reports it on Schedule D, line 6. ma_part_c_gross_income and ma_part_a_agi (the Schedule D, line 13 amount for line 25) now add it to long_term_capital_gains. The two changes are in one PR because each is wrong without the other: Part B must stop subtracting Part C at the same time as Part C starts to include the distributions.

Tests

gross_income/ma_part_b_gross_income.yaml has eight 2023 returns, each computed line by line from the 2023 Form 1, Schedule B and Schedule D (wages 60,000; single unless noted):

Return	Part B gross: main → PR	Tax before credits: main → PR
Dividends 1,000, short-term gain 3,000, long-term loss 1,000	59,000 → 60,000	2,850 → 2,900
Short-term gain 3,000, long-term loss 1,000	59,000 → 60,000	2,800 → 2,850
Dividends 1,000, short-term gain 1,000, long-term loss 3,000	59,000 → 60,000	not asserted¹
Joint: long-term gain 5,000; spouse short-term loss 2,000	62,000 → 60,000	2,710 → 2,610
Distributions 5,000	65,000 → 60,000	2,930 → 2,930
Distributions 5,000, short-term loss 2,000	65,000 → 60,000	2,930 → 2,830
Distributions 5,000, long-term loss 3,000	65,000 → 60,000	2,930 → 2,780
Distributions 5,000, short-term gain 3,000, long-term loss 1,000	64,000 → 60,000	3,050 → 3,135

¹ The form gives 2,680. Main gives 2,680 only because this error offsets #9905; this PR alone gives 2,730, and this PR with #9906 gives 2,680.

All eight tests fail on main. The NBER TAXSIM-35 staxbc equals the form in rows 2 and 3. In row 1 it gives 2,815 (PolicyEngine/policyengine-taxsim#1399).

All 661 tests under gov/states/ma and 746 other tests with Massachusetts households pass.

Overlap

• Subtract each spouse's Massachusetts business losses #9910 does not conflict.
• Apply Massachusetts Schedule B capital losses to dividends before long-term gains #9906 conflicts in ma_part_a_agi (one line). Its new ma_part_c_agi and ma_long_term_loss_against_dividends read long_term_capital_gains and need non_sch_d_capital_gains added too, whichever PR merges second.
• Leave dependents' capital income out of state lines on the filers' return #9894 conflicts in ma_part_a_agi and ma_part_c_gross_income (one line each): keep its tax_unit_non_dep_add and add non_sch_d_capital_gains to the list.

🤖 Generated with Claude Code

[hua7450]

Approving. Read the three formulas at 01b1873 and on main, plus ma_gross_income, ma_part_a_gross_income, irs_gross_income with gov/irs/gross_income/sources.yaml, and the capital_gains aggregate, against M.G.L. c. 62 s. 2(b) and the 2023 Form 1 instructions (Schedule D line 6 for capital gain distributions, page 27).

The new Part B formula subtracts exactly what irs_gross_income adds for the filers (each person's dividends, max_(0, capital_gains) and max_(0, non_sch_d_capital_gains), dependents excluded), so Part B is Form 1 line 10 with no capital component in any sign case and a net loss can never reduce it. The main formula subtracted ma_part_a_gross_income + ma_part_c_gross_income, whose unit-level netting differs from the per-person floor: with short-term 3,000 and long-term -1,000, gross income carries 2,000 but main removed 3,000 and understated Part B by 1,000; head gives 60,000 on 60,000 of wages, as the new case expects. Interest is unaffected because ma_part_a_gross_income never held it. Counting capital gain distributions in ma_part_c_gross_income and in the Schedule D line 13 amount inside ma_part_a_agi matches Schedule D line 6, and the three edits are consistent with each other, so the Parts still partition ma_gross_income with nothing dropped or double counted.

All eight new cases recompute by hand at head, and each differs on main. No +=, vectorized, changelog describes both changes.

Merge order with PR 9906: both rewrite ma_part_a_agi.py (textual conflict), and 9906's new ma_part_c_agi and ma_long_term_loss_against_dividends read long_term_capital_gains only. If 9906 lands after this PR without adding non_sch_d_capital_gains to those two reads, distributions drop out of the Part C base and cases 5 to 8 here fail. This PR does not move any value in 9906's tests, so merging this one first and having 9906 pick up the distributions in its rebase is the clean path.

Nits, not blocking:

1. part_a_and_c_income is a misleading name, since the comment says it is deliberately not Part A plus Part C gross income; filer_dividends_and_gains or similar would read better.
2. ma_part_b_gross_income and ma_part_c_gross_income could cite the Form 1 instructions page for Schedule D line 6 in their reference.

Pre-existing, for a follow-up: Form 4797 gains (other_net_gain) stay in Part B although Massachusetts routes them to Schedules B and D (instructions pages 25 and 27).