[MaxGhenis]

Correct Missouri TANF grants, Guam and Virgin Islands SNAP income standards, Hawaii child care income, and SSI food support timing.

impact: in 2026 on the default dataset, household benefits rise by $1,246 and household net income by $1,183, by up to $4 per household; Missouri TANF falls by $19,685, and taxes and SNAP don't change.

The affected populations include Missouri TANF units, Guam and Virgin Islands SNAP households, Hawaii child care applicants and recipients who receive TANF, and SSI recipients, including those who receive state supplements.

Missouri TANF

Missouri prints whole-dollar maximum grants in 13 CSR 40-2.310(13) and DSS Appendix B.
Use that row for the maximum grant and percentage-of-need income test: sizes 1–12 receive $136, $234, $292, $342, $388, $431, $474, $514, $554, $595, $635 and $675 before subtracting countable income.
Add $40 for each person above 12. For example, a size-16 unit receives $675 + 4 × $40 = $835; multiplying its $2,420 standard by 0.34526 gives $835.52920.
Keep the parameter's 1991-01-01 start date, matching the existing standard of need. Remove the unused maximum_benefit/percentage.yaml parameter.
We confirmed the regulation's 2016 amendment and the manual's identical rows from July 1990 and July 1993; we did not fetch earlier regulatory editions.

Retain the section (14) whole-dollar floor and $10 minimum. DSS also directs staff to round downward in Manual 0210.010.15.
Round the capped deficit to four decimal places before flooring. This precision choice can change deficits within $0.00005 of a whole dollar; deficits in whole cents keep their floor.
For example, with a direct countable-income input of $127.00001 and a $234 maximum, float32 stores a $106.99999237060547 deficit; the guarded floor pays $107.
A bounded arithmetic search over 5,985,732 combinations of earnings in whole cents for two new applicants found no witness, so the regression supplies countable income directly.
The search covered two earners with at least $120 each per month and combined exact countable income from $1 to $282.

SNAP in Guam and the Virgin Islands

Map GU and VI to contiguous-state poverty guidelines for SNAP income standards under 7 CFR 273.9(a).
Keep their state groups, allotments and deductions, along with main's existing large-household increment rounding.
HHS sets a $21,640 guideline for two people and a $5,680 increment in its 2026 guidelines.
For FY2027, the two-person net limit equals ceil(21,640 / 12) = $1,804; the gross limit equals ceil(1.3 × 21,640 / 12) = $2,345.
All ten territory regressions key state_code to 2026 so October calculations exercise GU or VI.

Hawaii child care

Count Hawaii TANF throughout the modeled source list, starting 2017-08-01.
Hawaii counts public assistance under HAR 17-798.2-10(b)(11), with a March 8, 2008 effective date and a December 31, 2017 compilation date.
Hawaii continues counting it under HAR 17-798.3-10(b)(11) from August 6, 2021.
Merge the identical dated source lists and cite both chapters. Hawaii counts $1,100 for families with $1,000 earnings and $100 Hawaii TANF before and after the chapter change.

hi_ccap_tanf_income reads hi_tanf under the take-up and abolition gates. Reading national tanf would traverse other states' child care deductions back into Hawaii CCAP.
A generic tanf input does not override this source; supply hi_tanf or hi_ccap_tanf_income.
The abolition test supplies a $100 Hawaii grant and expects zero TANF income and $1,000 total CCAP income.
We keep the adult-care TANF deduction from #10254 and update its parameter metadata.

SSI monthly inputs and food timing

SSA excludes food received on or after September 30, 2024 under 89 FR 21199.
Use the second prior income month for the food parameter and transition-date flag under 20 CFR 416.420(a).
Read food receipt in that income month when it falls in the payment year. For January and February, use payment-month food receipt because one-year household inputs contain no prior-year history.
We choose this dating to match food receipt with its governing rule while retaining one-year input coverage.
For pre-exclusion years, we also shift food-only PMV timing by two months when food receipt changes within the year.
Keep shelter inputs in the payment month. Value the PMV at the payment month's federal benefit rate, consistent with section 416.420's Example 2.

Define ssi_pmv_applies, ssi_pmv_amount, ssi_shelter_support_value, ssi_in_kind_support_and_maintenance and ssi_countable_income monthly.
Callers supplying a single month's amount must supply a monthly value. Year-keyed monetary inputs split evenly across months, and annual monetary amounts sum monthly values.
We re-key supplement and CAPI inputs without changing their amounts or expected outputs, and keep the annual exclusions wrapper's annual interface.

For continuing food-only support in 2024, PMV = $943 / 3 + $20 = $334.3333. Monthly countable income equals $314.3333 in January–November and zero in December.
October and November SSI equal $628.6667; December SSI equals $943. Annual ISM equals 11 × $334.3333 = $3,677.6667, and annual SSI equals $7,858.3333.
If food support starts in October, SSA pays $943 SSI in October and November because August and September contain no food support.

Witness at f04ff49222b0c281dece50bcce569dda19a3d06f, with displayed amounts rounded to cents:

continuing_food
2024-10: food 1; ISM 334.33; countable 314.33; SSI 628.67
2024-11: food 1; ISM 334.33; countable 314.33; SSI 628.67
2024-12: food 1; ISM 0.00; countable 0.00; SSI 943.00
2024: food 1; ISM 3677.67; countable 3457.67; SSI 7858.33
food_from_october
2024-10: food 1; ISM 0.00; countable 0.00; SSI 943.00
2024-11: food 1; ISM 0.00; countable 0.00; SSI 943.00

Read monthly countable income when allocating parental resources. SSA reallocates those resources in the first month a child's eligibility changes under POMS SI 01330.200 B.
The two-child regression allocates $4,000 to each child in January and November 2024, then $2,000 each in December when the food-supported child becomes income-eligible.

Validation and invariants

• CI: the latest Pull request run at this head hasn't finished.
  Pull request run 42159523214 uses final head f04ff49222b0c281dece50bcce569dda19a3d06f. All 33 changed test files passed: 457 YAML cases and 236 Python cases, totaling 693. The 57 unchanged IRMAA cases also passed. Attempt 1 stopped Python shard 8 after its runner received a shutdown signal. GitHub accepted a retry; attempt 2 remains in progress as of November 6 at 05:44 UTC.
• All twelve mutation checks failed their intended regressions, and the runner restored every original file. M1–M8 cover food-policy dates, PMV units, annual ISM averaging, cross-year food reads, deficit rounding, territory mapping, TANF abolition and concurrent food dating. Three added mutations replace Missouri's table with a product, remove its $40 increment, and remove TANF from Hawaii's 2017 list. The final mutation restores the annual-average resource reader; the December two-child case catches it.
• Final inputs fail on main: SNAP gross 3 failed/15 passed, net 3 failed/17 passed, FPG 2 failed/2 passed, and Missouri payments 9 failed/5 passed. Those files pass all 18, 20, 4 and 14 cases respectively in final-head CI. Hawaii's three pre-2021 households count $1,000 on main versus the required $1,100; all three pass in the 16-case final-head file. The main Hawaii comparison omits the new helper's output assertion because main lacks that variable.

The committed SSI properties check annual input conversion, sums of monthly amounts, monthly PMV bounds, income-month food reads, couples, shared-resource children, and each state supplement against independent January 2026 reference inputs.
We compare with independent January 2026 inputs; we do not use historical main outputs.
We compared four separate Texas households across all twelve 2024 months against main 040fe5686c956dfce960f7c55fa263687a5bf669.
SSI, countable income and ISM each differ in 23 of 48 person-months, using a one-cent tolerance. The household without support matches in every month.
For continuing support, only December changes: annual SSI rises from $7,544 to $7,858.33, and annual ISM falls from $4,012 to $3,677.67.
Main counts August-start and October-start food throughout 2024. This head counts August-start food in October and November, and excludes October-start food from every 2024 payment.
Broader population and supplement comparisons remain pending.

Missouri properties retain caretaker and dependent-parent checks against the printed row. SNAP regressions cover territory thresholds, one-dollar excesses and large households.

Microsimulation impact

We estimate that this branch raises household benefits by $1,246 and household net income by $1,183 in 2026 on the default dataset, by up to $4 per household. We compared main at the merge base e53b426082 with this branch at f04ff49222. We measured 2026 only, not 2021 or 2024.

• Missouri TANF, now paid from the regulation's whole-dollar maximum grants, falls by $19,685 in all for 8 units (33,428 weighted): 4 rise and 4 fall, by up to $10 a year each.
• Hawaii child care: countable income now includes Hawaii TANF (hi_ccap_tanf_income), which main left out. It rises by $29.5m for 5 units (2,673 weighted), by up to $13,435 each. The comparison didn't measure hi_ccap itself.
• SSI and SSI countable income change by less than $1 on every record: in total, SSI falls by $55 and countable income rises by $1,391. SNAP doesn't change.
• Federal and state income tax don't change.

Output	Main total	Change	Records changed (weighted)
mo_tanf	$109.8m	-$19,685	8 (33,428)
hi_ccap_countable_income	$71.10bn	+$29.5m	5 (2,673)
snap	$95.06bn	$0	0 (0)
ssi	$57.01bn	-$55	0 (0)
ssi_countable_income	$4,302.4bn	+$1,391	9,633 (16,739,838)
ssi_in_kind_support_and_maintenance	$0	$0	0 (0)
state_income_tax	$578.0bn	$0	0 (0)
income_tax	$2,595.1bn	$0	0 (0)
household_benefits	$2,060.0bn	+$1,246	2 (19,311)
household_net_income	$17,161.7bn	+$1,183	4 (23,001)

Not in this PR

• An otherwise eligible recipient first entitled in October 2024 with food-only support receives $943 under section 416.420(b)(1). The model treats that recipient as continuing.
• The model uses one monthly income figure for eligibility and payment, so October and November eligibility tests still count continuing food support.
• Other SSI income sources retain current-month accounting. We do not implement the first-month, second-month, assistance-income and COLA exceptions. The model holds neither eligibility history nor prior-year inputs for a one-year household.
• We leave population comparisons for 2021, 2024 and 2026 to the hub and ran no microsimulation in this lane.

Follow-ups

Read monthly ISM for California split SSI/CAPI couples

ca_state_supplement_person still reads ISM at period.this_year for split couples and averages the 2024 change across months.
Add a split-couple witness and read ISM for the same month.

Verify Guam and Virgin Islands BBCE gross poverty standards

tanf_non_cash_fpg still indexes the territories' zero placeholders for BBCE gross standards.
Confirm the governing territorial rule before extending the SNAP mapping to that reader.

Reconcile SSI period changes when overlapping PRs land

Open #10146 overlaps the exclusions helper, ISM, PMV, countable income and parental resource reader; #10639 overlaps the resource reader and its YAML.
Open #10141 overlaps the CAPI and Montana supplement YAML. Whichever PR lands second must reconcile the monthly definitions. We changed no sibling branch.

axiom: TheAxiomFoundation/rulespec-us#1444 queued, extended with a comment (MO TANF rounding; re-pin and encoder approval); TheAxiomFoundation/rulespec-us#1394 queued, extended with a comment (SNAP GU/VI FY2027 companions; signed encoding and HHS source admission); TheAxiomFoundation/rulespec-us#1757 queued, extended with a comment (HI CCAP TANF income; re-pin and encoder approval); TheAxiomFoundation/rulespec-us#1865 carried by #10254 (no HI TANF deduction change here); TheAxiomFoundation/rulespec-us#1866 queued (SSI food ISM; historical corpus ingestion, re-pin and encoder approval).

🤖 Generated with Claude Code

[PavelMakarchuk]

@MaxGhenis — heads-up on an overlap. #10254 (merged today as 52abee5, fixes #10208) limits the Hawaii TANF dependent care deduction to incapacitated adults (HAR 17-676-54.1(c)(4), (d)(4)), the same rule as this PR's "HI TANF care deduction covers incapacitated adults only" item, in the same files (hi_tanf_dependent_care_deduction.py, dependent_care/amount.yaml and their tests). An independent check found both give the same result through different inputs.

So the HI TANF part of this PR is now on main and will conflict. Could you drop it from this PR (keeping the MO TANF rounding, SNAP GU/VI, HI CCAP and SSI items), or rebase it onto main?

🤖 Posted by Claude Code

[MaxGhenis]

Picking this up for the US hub. I'll post a hand-off comment here when it's ready.

---

Generated by Claude Code

[DTrim99]

PR Review: Fix four statute-first benefit bugs: MO TANF rounding, SNAP GU/VI limits, HI CCAP TANF income, SSI food ISM date

Summary

I checked each fix against its own law before reading the code. Three of the four are right.

• SNAP Guam and the Virgin Islands. Matches 7 CFR 273.9(a)(1)(i) and (a)(2)(i), and every FY2027 expectation matches the FNS FY2027 COLA memo.
• SSI food ISM date. The monthly ISM chain and the September 30, 2024 switch match 89 FR 21199, and the annual and monthly arithmetic checks out.
• Hawaii CCAP. Counting Hawaii TANF from August 6, 2021 is right under HAR 17-798.3-10(b)(11).
• Missouri TANF. Flooring the grant follows 13 CSR 40-2.310(14), but the maximum it floors is wrong. The regulation's own table, repeated in DSS Manual Appendix B, sets the maximum grant at whole-dollar amounts, and for 7 of the 12 table sizes those are $1 above floor(34.526% × standard of need). Under this head a zero-income one-person unit gets $135 where the regulation says $136. The queued parity issue for this item says the same.

There are two smaller Hawaii items. The CCAP fix stops at August 6, 2021, although the predecessor rule counted Hawaii public assistance in the same words, and a new test pins that omission. The new TANF-abolition guard is also untested. CI is green (34/34), but the branch now conflicts with main in one CA CAPI test file.

Critical (must fix)

1. Missouri TANF: the grant is floored from 34.526% × standard of need instead of the regulation's maximum-grant table, so it is $1 low for unit sizes 1, 4, 5, 6, 7, 10 and 11.

Files: variables/gov/states/mo/dss/tanf/mo_tanf.py:28-29, which floors capped_benefit built on mo_tanf_maximum_benefit (= standard_of_need × 0.34526).

The law:

• 13 CSR 40-2.310(13)(C) (PDF p.35) says: "the grant amount will be the deficit determined in the Percentage of Need Test. The table below indicates the maximum grant amount by application of the Percentage of Need (34.526%) to the Standard of Need". Its table gives "34.526% of Need" amounts of 136, 234, 292, 342, 388, 431, 474, 514, 554, 595, 635 and 675 for sizes 1-12, then 715 … 1075 for sizes 13-22 ($40 per extra person).
• DSS Manual 0200.000.00 Appendix B, "From July 1, 1993 on", repeats the same row.
• (14) (PDF p.36) then lowers the payment to the whole dollar.

The binding maximum for one person is therefore $136, not floor(135.68718) = $135. Because the table amounts are whole dollars, the head gives $1 less in every case at these sizes:

Size	Standard of need × 0.34526	Head (floor)	Regulation table
1	135.68718	135	136
4	341.8074	341	342
5	387.72698	387	388
6	430.53922	430	431
7	473.69672	473	474
10	594.53772	594	595
11	634.93314	634	635

Sizes 2, 3, 8, 9 and 12-22 agree; for example 1956 × 0.34526 = 675.33 → 675, and 2420 × 0.34526 = 835.53 → 835 = table.

The new expectations pin the wrong amounts in 9 places:

• mo_tanf: 135 in integration.yaml (lines 423, 454, 487), mo_tanf_gross_unearned_income.yaml:267, mo_tanf_is_assistance_unit_member.yaml:416, non_parent_caretaker/integration.yaml:416 and non_parent_caretaker/mo_tanf_non_parent_caretaker.yaml:104. All should be 136.
• mo_tanf: 341 at mo_tanf_is_assistance_unit_member.yaml:537. Should be 342.
• integration.yaml:251 "Family of 5 with child support income" expects 12 × floor(387.727 − 300) = 1,044. The table gives 12 × (388 − 300) = 1,056.
• The two core property tests encode the same arithmetic: _payment(size) = np.floor(STANDARD_OF_NEED × PAYMENT_PERCENTAGE).

The companion parity request agrees. The #10332 extension comment on rulespec-us#1444 lists these rows as "135; published table says 136" and "341; published table says 342". It tells the encoder to "Preserve Appendix B's published table, including 136/342/388 for sizes 1/4/5 … rather than silently replacing it with PE's 135/341/387."

Fix:

• Encode the table as a parameter, e.g. gov.states.mo.dss.tanf.maximum_benefit.amount by size 1-12 plus a $40 increment per person above 12. Reference 13 CSR 40-2.310(13)(C) (PDF p.35) and Appendix B.
• Compute mo_tanf_maximum_benefit from it.
• Keep floor(max − countable income) and the $10 minimum.
• Update the 9 YAML expectations and the property tests' _payment.

The percentage-of-need eligibility test in (12) can stay on the computed product: the incomes where the two differ give a grant under $10 either way. If you read the percentage product as controlling over the table, that is a source-conflict choice to record per CONTRIBUTING.md#source-conflicts. However, the regulation and the manual both publish 136.

Should address

1. Hawaii CCAP: Hawaii TANF is still left out from 2017-08-01 to 2021-08-05, although the predecessor rule counts it in the same words, and Case 11 pins the omission.

The 2017-08-01 list in parameters/gov/states/hi/bessd/ccap/income/countable_income/sources.yaml:33-38 keeps the old list on the ground that "the edition in force from August 2017 was not established". hi_ccap_countable_income.yaml Case 11 (2021-01) then asserts 1,000, not 1,100, with the same reasoning.

The predecessor rule is established:

• HAR 17-798.2-10(b)(11) reads "Hawaii public assistance payments" (LII). Its history note reads "[Eff 03/08/08] … Comp 12/31/2017", so the same text was in force from 2008 until chapter 17-798.3 replaced it on 8/6/2021.
• The 2013 DHS copy already cited in this file (17-798.2-10(b), 2013 PDF) lists the same item.
• hi_ccap_countable_income.py itself says the source list is "identically 17-798.2-10(b) before August 6, 2021".

The circularity that used to justify leaving TANF out is now solved by hi_ccap_tanf_income, so nothing blocks the earlier period. rulespec-us#1757 also flags the January 2021 row: "must not become an unsupported exclusion".

Fix:

• Add hi_ccap_tanf_income to the 2017-08-01 list. The 2021-08-06 entry then duplicates it and can go.
• Change Case 11 to expect 1,100 (citing 17-798.2-10(b)(11)).
• Update the comment and the Incorrect MO adjusted gross income formula #1757 companion row.

2. Hawaii CCAP: the new TANF-abolition guard in hi_ccap_tanf_income is untested (mutation rule).

variables/gov/states/hi/bessd/ccap/hi_ccap_tanf_income.py:25-26 adds if parameters(period).gov.hhs.tanf.abolish_tanf: return 0. The guard is needed because hi_tanf does not check the switch, so without it CCAP would count a grant that tanf_if_takes_up zeroes. No test sets gov.hhs.tanf.abolish_tanf in the Hawaii tree (only gov/hhs/tanf/cash/tanf.yaml and integration.yaml do), so deleting the guard fails nothing.

Fix: add one case to hi_ccap_countable_income.yaml: period 2025-01, gov.hhs.tanf.abolish_tanf: true, hi_tanf: 100, earnings $12,000/year. Expect hi_ccap_tanf_income: 0 and hi_ccap_countable_income: 1_000. One override value adds only one cached system copy to the HI tree.

Suggestions

1. SNAP BBCE in Guam and the Virgin Islands. variables/gov/hhs/tanf/non_cash/tanf_non_cash_fpg.py still indexes GU/VI's zero guideline placeholders. As a result, Guam's BBCE gross limit (165%) and the Virgin Islands' (175% from 2024-10-01) are both $0, and BBCE never applies there. Its comment ("encodes the same guideline arithmetic … Keep the two in sync") is now stale, because this PR changes only snap_monthly_fpg_amounts. The body lists this as a follow-up. Please file an issue and update or remove the sync comment.
2. SNAP area list. SNAP_CONTIGUOUS_US_GUIDELINE_AREAS = ["GU", "VI"] is a policy list in Python. A list parameter under gov.usda.snap.income.limit, referencing 7 CFR 273.9(a)(1)(i), would fit the parameter conventions and be reformable.
3. Axiom line entry for Integration Test of CARE/FERA Shows SNAP Issue #1865. The PR no longer changes the Hawaii TANF deduction; it changes only dependent_care/amount.yaml metadata. Limit Hawaii, Vermont and Kentucky TANF dependent care deductions to the care their rules allow #10254, which made the change, already lists rulespec-us#1865 in its own axiom line. Integration Test of CARE/FERA Shows SNAP Issue #1865's body still describes hi_tanf_qualifying_incapacitated_adult_care_expenses, which this head deletes. Drop Integration Test of CARE/FERA Shows SNAP Issue #1865 from this line, or mark it as carried by Limit Hawaii, Vermont and Kentucky TANF dependent care deductions to the care their rules allow #10254.
4. SSI retrospective monthly accounting (20 CFR 416.420). This is documented as out of scope; October 2024 ssi reflects October income. A tracking issue, and a sentence in ssi's or ssi_countable_income's documentation, would keep the limitation visible.
5. Input semantics. A month-level input of ssi_countable_income or ssi_in_kind_support_and_maintenance, and now ssi_shelter_support_value and ssi_pmv_amount, is now a monthly amount. Year-keyed inputs are still split evenly. The changelog could say so for API users.
6. After merging main. Main's Compute the California state supplement per person #10601 ca_state_supplement_person reads ssi_in_kind_support_and_maintenance at period.this_year. With the monthly chain, that path averages 2024's nine months of food ISM over twelve for food-only couples. Reading the month would be more precise; this is not a blocker.

Verified correct

Hand-checked expectations

Case	Arithmetic	Result
MO size 2, no income (2026-01)	678 × 0.34526 = 234.08628 → 234; table 234	✓
MO 678, income 223.25 / 224.10	10.83628 → 10 (paid); 9.98628 → 9 → 0	✓
MO family of 3, $250 earnings (2024)	250 − (90 + 30 + 130/3) = 86.67; 292.09 − 86.67 = 205.42 → 205; × 12 = 2,460	✓
MO size 1, SSI parent excluded	floor(135.687) = 135 vs table 136	✗ (Critical 1)
MO family of 5, $300 child support	12 × floor(87.727) = 1,044 vs 12 × 88 = 1,056	✗ (Critical 1)
SNAP GU, 2 people, gross (2026-10)	1.3 × 21,640 / 12 = 2,344.33 → 2,345 (COLA memo: 2,345)	✓
SNAP VI, 2 people, net	21,640 / 12 = 1,803.33 → 1,804 (memo: 1,804); 1,805 fails	✓
SNAP GU, 10 people, gross	8 people 55,720 / 12 × 1.3 = 6,036.33 → 6,037; increment 5,680 / 12 × 1.3 = 615.33 → 616; 6,037 + 2 × 616 = 7,269 (memo: 6,037 / 616)	✓
SNAP VI, 9 people, net	4,643.33 → 4,644; 473.33 → 474; 5,118 (memo: 4,644 / 474)	✓
snap_fpg VI 3 / GU 2	27,320 / 12 = 2,276.67; 21,640 / 12 = 1,803.33	✓
HI CCAP Case 9	1,000 + 100 = 1,100	✓
HI CCAP copay witness	2,653.64 / (30,650 / 12) = 1.0389 → 1% band; 26.5364; min(500, 1,000) − 26.5364 = 473.4636	✓ (under the model's FPG-band schedule)
SSI 2024-09 food only	PMV 943/3 + 20 = 334.33; countable 314.33; SSI 943 − 314.33 = 628.67	✓
SSI 2024 annual (Case 9f)	ISM 9 × 334.333 = 3,009; countable 9 × 314.333 = 2,829; SSI 9 × 628.667 + 3 × 943 = 8,487	✓
SSI 2025 rebuttal	actual 100 < 342.33 → ISM 100, countable 80; actual 500 → ISM 342.33, countable 322.33	✓

SSI food ISM date logic.

• 89 FR 21199 (PDF p.1): "This final rule will be effective September 30, 2024".
• The amended 416.1130 on PDF p.11 adds "(Food is not included in the calculations of in-kind support and maintenance.)", so the #page 1,11 reference corroborates.
• food_counts switches on 2024-09-30. Monthly variables read it at the month start. ssi_pmv_applies also reads period.stop, so September 2024 counts food unless all of it came on or after the 30th, and October onward counts none. This agrees with SSA EM-24047.
• No day proration, which is right for a monthly PMV cap.
• The new branches are pinned: dropping ~late_food_excluded fails 9d, dropping the month-start test fails 9e, and keeping the old 2024-10-01 date fails 9d.
• With the MONTH definitions, core sums float variables over a year and reads bool STOCK variables at the last month. No caller reads ssi_pmv_applies annually.

SNAP law.

• 7 CFR 273.9(a)(1)(i) and (a)(2)(i) put Guam and the Virgin Islands on the 48-state/DC guidelines.
• 7 U.S.C. 2014(c) caps their standards at the 48-state level.
• The FNS FY2027 COLA memo (signed 2026-08-21) lists "48 States, D.C., Guam, Virgin Islands" in one column.
• The mapping affects income standards only; GU/VI allotments and deductions keep their own state group.

Hawaii CCAP.

• HAR 17-798.3-10(b)(11) reads "Hawaii public assistance payments prior to any reduction … due to overpayment recovery", [Eff 8/6/2021] (LII).
• 17-798.3-14 has no co-payment exemption for TANF/FTW families.
• hi_tanf × takes_up_tanf_if_eligible equals tanf for Hawaii households (tanf = tanf_if_takes_up gated by take-up) and avoids the multi-state cycle. Microsimulation CI passed.
• The HI TANF amount.yaml metadata reference is valid: HAR 17-676 PDF p.6 has "Full-time employed … thirty two hours a week", p.36 has (c)(4) $175/$165, and p.37 has (d)(4).

Re-keyed test inputs.

• All 71 year-keyed ssi_countable_income / ISM inputs in the CA, FL, IN, LA, MO and MT files match their case's year.
• The remaining bare month-period inputs at the head are zeros (LA OSS integration, NE AABD) or the PR's own monthly ISM cases.

Python property test. In CI it runs in the core-2 group of Rest Python shard 2: 17.3 s fixture setup and about 0.1 s of calls, within a roughly 29-minute group. It is collected once. Peak memory is not in the CI log; the local figure (2,656 MiB, including model import) is labelled as local in the hand-off.

Other checks.

• Vectorization is fine: np.floor, np.isin, and bool-array logic in ssi_pmv_applies.
• No new state-gated int or enum key (the NH Add New Hampshire state supplementation for 2026 #10156 pattern does not apply).
• The changelog fragment is top-level and typed fixed.
• Tests are in variable-named files.
• No partner-contract files are touched.
• The source-conflict ratchet is satisfied: the two removed KNOWN_UNMARKED entries go with their comments, and test_source_conflicts passed at the head.

Axiom line assessment

The line follows the required format, and all five issues are OPEN and labelled pe-parity. Each records its blocker (re-pin, encoder approval, HHS 2026 admission, historical CFR ingestion), and each extension cites fixed head d318d73b78. The code for these provisions has not changed since, apart from the CCAP 2021 list now reading military_retirement_income.

Issue	Status
rulespec-us#1394 (SNAP GU/VI)	Dispatch-ready. Verbatim 273.9(a) text and FY2027 companion rows from the COLA memo. Matches the head.
rulespec-us#1866 (SSI food ISM)	Dispatch-ready. Dated CFR text, companion rows and a 416.420 counterexample. Matches the head.
rulespec-us#1444 (MO TANF rounding)	The extension says sizes 1/4/5 should follow the published table (136/342/388), while the head returns 135/341/387. It matches once Critical 1 is fixed.
rulespec-us#1757 (HI CCAP)	Matches the post-2021 behavior. Its January 2021 row needs updating with Should 1.
rulespec-us#1865 (HI TANF adult care)	Now carried by #10254 (see Suggestion 3).

CI status

All 34 checks pass at d93c0dd3c2 (run 41425286966), re-checked just before writing. That includes every Baseline and Contrib shard, all seven Rest Python shards, Microsimulation and Household API Partners. None are pending or failing.

Branch status

• 11 commits ahead (including a merge of main at 97b6d5135c) and 122 behind main.
• git merge-tree --write-tree PolicyEngine/main d93c0dd3c2 reports one content conflict, and GitHub shows the PR as conflicting:
  • The conflict is in tests/policy/baseline/gov/states/ca/cdss/capi/ca_capi_person.yaml, which main's Compute the California state supplement per person #10601 rewrote.
  • Main's version has bare ssi_countable_income: 4_800 and 0 at period 2026-01 in Cases 6 and 11.
  • When resolving, key them to {2026: ...} as this PR did elsewhere. Otherwise $4,800 becomes a monthly amount.
• ssi_countable_income.py auto-merges correctly: main's Count income of aged, blind, or disabled people who fail SSI's noncitizen test #10240 is_ssi_aged_blind_disabled gate works with the monthly formula.
• test_source_conflicts.py auto-merges correctly, with no overlap with the two removed entries.
• Main's other SSI changes since the merge base are compatible: railroad benefits in unearned sources, the federal_supplement_payment helper, and the spousal-deeming defined_for.
• None of main's new tests set food support.

Verdict: REQUEST_CHANGES

[DTrim99]

Requesting changes: 1 critical (the MO TANF grant floors 34.526% of the standard of need, but 13 CSR 40-2.310(13)(C) and DSS Appendix B set whole-dollar maximum grants, so the head is $1 low at sizes 1, 4, 5, 6, 7, 10 and 11; use a maximum-grant table), 2 should (HI CCAP still omits Hawaii TANF from 2017-08-01 to 2021-08-05, which HAR 17-798.2-10(b)(11) counts; the abolish_tanf guard is unpinned). SNAP GU/VI and the SSI food ISM date are right. Full report in the comment above.

[MaxGhenis]

Hub review in progress

Superseded by the final review: #10332 (comment)

---

Generated by Claude Code

[MaxGhenis]

Ready for hub review
VERDICT: REQUEST_CHANGES
2ed172e
METHODOLOGY_CHOICE: yes

I reviewed ed37709fbd, then 2ed172e6d6. The head moved twice during this review, past the brief's d93c0dd3c2.

At 2ed172e6d6: two commits (f65a67d0e5, 2ed172e6d6) fix item 1 below, essentially as proposed there.

• ssi_pmv_applies reads the September-30 flag only in the transition month.
• HI CCAP Cases 13 and 14 key their inputs to 2021.
• The read-pin property test now expects the flag only for September 2024 income.

Run at 2ed172e6d6: hi_ccap_countable_income.yaml 14 passed, test_irmaa_first_modeled_year.py 57 passed, test_ssi_monthly_ism_properties.py 64 passed. Items 2-14 are unchanged at 2ed172e6d6; I reran the item 2 and item 10 witnesses there with the same results. CI at 2ed172e6d6 (run 42140083442) was still in progress when I posted.

What changed since d93c0dd3c2:

• a40e4afcf2 merges main again. The PR now merges cleanly with current main (4279ce2e63): git merge-tree --write-tree --name-only pr-10332 origin/main lists no conflicted files, and GitHub reports blocked, not dirty.
• Five new commits change behaviour:
  • b6dc7b7ed4 dates the SSI food exclusion by the income month two months back (20 CFR 416.420(a)). Food-only ISM now counts through the November 2024 payment month instead of September.
  • 35ae134c01 adds a float tolerance before the Missouri TANF floor.
  • 0882398b61 adds HI CCAP Cases 13 and 14 at the 2021 boundary.
  • ae83afa6c3 and ed37709fbd add property and reference checks, and change ssi_resources_deemed_from_ineligible_parent to compare monthly countable income.
• The body and the hand-off were not updated for these commits, and nobody has answered the 2026-11-05 review that requested changes.

CI at d93c0dd3c2 passed (run 41425286966, 34/34). CI at ed37709fbd fails (run 42127473564: 24 passed, 3 failed, Contrib skipped).

P1

1. CI was red at ed37709fbd: two new HI CCAP cases and main's IRMAA test failed (fixed and verified at 2ed172e6d6)

Kept for the record; no action is left at 2ed172e6d6.

(a) HI CCAP Cases 13 and 14 return 0.

• File: policyengine_us/tests/policy/baseline/gov/states/hi/bessd/ccap/hi_ccap_countable_income.yaml, the last two cases.
• Both use a month period (2021-08, 2021-09) but enter year-defined inputs bare. The runner does not apply a bare year-defined input at a non-January month, so the household is never in Hawaii and has no earnings.
• CI (Baseline states-shard-4) and my run agree:

  policyengine-core test .../hi_ccap_countable_income.yaml -c policyengine_us
    Test 'Case 13, August 2021 income excludes Hawaii TANF.':
      hi_ccap_countable_income@2021-08: [0.] differs from 1000.0
    Test 'Case 14, September 2021 income includes Hawaii TANF.':
      hi_ccap_countable_income@2021-09: [0.] differs from 1100.0
  2 failed, 12 passed
  

• The code is right once the inputs arrive. With every input keyed to its year, the same household gives 1,000 for 2021-08 and 1,100 for 2021-09.
• Fix: in both cases, key the year-defined inputs to 2021:

      parent:
        age:
          2021: 30
        employment_income_before_lsr:
          2021: 12_000
      child:
        age:
          2021: 5
  ...
        takes_up_tanf_if_eligible:
          2021: true
  ...
        state_code:
          2021: HI

  With this change on a scratch copy of the head, the file passes 14/14.

(b) Main's test_irmaa_first_modeled_year.py fails 6 cases for 2015 (Rest Python shard 6).

• The test asserts that computing 2015 caches nothing before 2015.
• ssi_pmv_applies.py:35-37 now reads ssi_food_from_others_only_on_or_after_food_exclusion at income_period in every month, which is 2014-11 and 2014-12 for January and February 2015.
• CI log: At index 1 diff: ('ssi_food_from_others_only_on_or_after_food_exclusion', '2014-11') != ('tax_exempt_interest_income', '2013').
• Fix: the flag matters only in the month the exclusion starts, so read it only then. In ssi_pmv_applies.py, replace

          food_only_after_exclusion = person(
              "ssi_food_from_others_only_on_or_after_food_exclusion", income_period
          )
          late_food_excluded = exclusion_starts_this_month & food_only_after_exclusion
          receives_food = person("ssi_receives_food_from_others", period)

  with

          receives_food = person("ssi_receives_food_from_others", period)
          # Read the date flag only in the month the exclusion starts, so no
          # earlier month is read or cached.
          late_food_excluded = (
              person(
                  "ssi_food_from_others_only_on_or_after_food_exclusion",
                  income_period,
              )
              if exclusion_starts_this_month
              else np.zeros_like(receives_food, dtype=bool)
          )

• I ran this replacement, together with the HI CCAP keys above, on a scratch tree:
  • test_irmaa_first_modeled_year.py: 57 passed.
  • hi_ccap_countable_income.yaml: 14 passed.
  • ssi_in_kind_support_and_maintenance.yaml: 9 passed.
  • ssi_living_arrangement_integration.yaml: 25 passed.
  • test_ssi_monthly_ism_properties.py: 63 passed, 1 failed. The failure is test_pmv_reads_income_month_dates_and_payment_month_support, whose lines 479-481 assert that the flag is read in every month. That assertion pins the old read pattern rather than a result, so it needs the same condition: assert the flag read only when (year, month) == (2024, 11).

2. SSI: the food date comes from the income month, but food receipt from the payment month, which gives results neither reading supports

• File: policyengine_us/variables/gov/ssa/ssi/eligibility/income/ism/ssi_pmv_applies.py:29-39.
• food_counts is read at period.offset(-2, "month"), the §416.420(a) income month. ssi_receives_food_from_others, which is MONTH-defined, is still read at period, the payment month.
• When food receipt changes within 2024, the payment counts the payment month's food under the income month's rule.
• Witness: TX, age 70, ssi_receives_food_from_others: {"2024-10": True}, so food from October on (inputs carry forward). Run at the head and at d93c0dd3c2:

  pr  food only in 2024-10 | 2024-08: ism 0.00 ssi 943.00; 2024-09: ism 0.00 ssi 943.00; 2024-10: ism 334.33 ssi 628.67; 2024-11: ism 334.33 ssi 628.67; 2024-12: ism 0.00 ssi 943.00
  old food only in 2024-10 | ... 2024-10: ism 0.00 ssi 943.00; 2024-11: ism 0.00 ssi 943.00; ...
  pr  food only in 2024-08 | 2024-08: ism 334.33 ...; 2024-09: 334.33; 2024-10: 334.33; 2024-11: 334.33; 2024-12: ism 0.00 ssi 943.00
  

• Under concurrent budgeting, October and November food is already excluded, so SSI is 943. Under §416.420(a), the October and November payments use August and September income, which had no food, so SSI is also 943. The head pays 628.67 in both months.
• For food that runs from August only, the head counts four months (August-November). Concurrent budgeting counts August and September; retrospective budgeting counts October and November.
• The new tests can't see this. Every SSI case holds food receipt constant across months, so "Case 9b, a continuing recipient counts August food for the October 2024 payment" passes although the code reads October's food. The property test pins the payment-month read deliberately (line 482: assert set(person_reads.values()) == {str(payment)}, with a mock that returns False for any other month).
• The PR's reason, per the test's comment, is that a one-year household has no earlier-year inputs. That only affects the January and February payment months, whose income months fall in the prior year. Food counts in every month before September 2024, so nothing is lost there.
• Mutant M4 (read receipt at income_period) fails test_months_match_when_nothing_changes_within_the_year[2023] for exactly that cross-year reason, plus the read-pin test and 8 YAML cases whose bare month inputs don't reach August.
• Fix: read receipt in the income month when it falls in the payment year:

          support_period = (
              income_period if income_period.start.year == period.start.year else period
          )
          receives_food = person("ssi_receives_food_from_others", support_period)

  Key the food inputs in the 2024 YAML cases to the year (ssi_receives_food_from_others: {2024: true}). Then add a case with food from 2024-10 only that expects 943 in October and November.
• The alternative is to drop §416.420 dating and return to d93c0dd3c2's concurrent dating. Concurrent dating is what the model does for every other income source; mutant M8 shows only Cases 9b, 9b2, 9d and 9f pin the difference.
• Either way, this is a methodology choice, so record it in the body.

3. SNAP: none of the 10 new Guam or Virgin Islands cases runs as a territory household

• Files:
  • policyengine_us/tests/policy/baseline/gov/usda/snap/eligibility/meets_snap_gross_income_test.yaml (4 new cases);
  • .../meets_snap_net_income_test.yaml (4);
  • .../income/snap_fpg.yaml (2).
• Every new case has period: 2026-10 and a bare state_code: GU or state_code: VI. state_code is year-defined, so the bare input doesn't apply at a non-January month, and the household stays in the default state, California.

  probe: household state_code {"2026-10": "GU"} vs {"2026": "GU"}
  main  bare  -> state_code_str@2026 CA  snap_fpg@2026-10 856.6667
  main  keyed -> state_code_str@2026 GU  snap_fpg@2026-10 0.0
  pr    bare  -> state_code_str@2026 CA  snap_fpg@2026-10 856.6667
  pr    keyed -> state_code_str@2026 GU  snap_fpg@2026-10 856.6667
  

• So the code fix works, but the tests don't test it:
  • At main, all three PR test files pass (gross 18/18, net 20/20; the PR's snap_fpg.yaml 4/4).
  • Mutant M6 (SNAP_CONTIGUOUS_US_GUIDELINE_AREAS = []) passes every SNAP file.
  • The body's witness "main: 0 and 0, so every GU/VI household failed both tests" is true of the model but not shown by any committed test.
• Fix: key each territory state_code to its year:

        state_code:
          2026: GU

  With that change in all 10 cases, the three files pass at the head (18, 20 and 4 passed).
  The same keyed files fail at main: gross 3 failed ('Guam two-person household passes the FY 2027 gross income test', 'Virgin Islands two-person household at the FY 2027 gross limit passes', 'Ten-person Guam household at the FY 2027 gross limit passes'), net 3 failed, and snap_fpg 2 failed. So the keyed tests reproduce the defect.

4. The body and hand-off describe a previous head

A reader of the body would rely on each of these, and each is false at ed37709fbd:

• §416.420. The "Not done" section says 20 CFR 416.420 retrospective accounting is not modelled and "October 2024 payments here reflect October income". The head now applies §416.420(a) dating to the food rule (item 2).
• SSI witness.
  • The body says "October 2024 PMV off, ISM 0, SSI 943" and "Year 2024 ISM 3,009 = 9 × 334.33 (main 4,012)".
  • At the head, October 2024 SSI is 628.67 and the 2024 ISM is 3,677.67 (11 × 334.33). The run above prints year ISM 3677.67 year ssi 7858.33.
  • Against main, a continuing food-only recipient now differs only in the December 2024 payment month, plus November if the September-30 flag is set.
• Differential and property-test claims.
  • The body says differences are "all in October-December 2024" and that in 2024 food-only support changes "only from October". Those numbers come from the earlier head and main 97b6d5135c.
  • The property test now asserts the change starts in December (test_2024_food_income_changes_payments_from_december).
  • The differential was not rerun.
• HI CCAP years. The body says "CCAP change from 2021-08-06". The code first counts Hawaii TANF in the September 2021 month, and Case 13 pins August 2021 as excluded. The changelog was updated ("from the September 2021 calculation"); the body was not.
• Follow-ups that the head now does.
  • "mo_tanf floors without a float tolerance": the head adds one (mo_tanf.py:28-39).
  • "ssi_resources_deemed_from_ineligible_parent reads at period.this_year / 12": the head now reads the month.
• CI. The hand-off's CI section describes a green run at d93c0dd3c2. ed37709fbd was red (item 1); CI at 2ed172e6d6 had not finished when I posted.

Fix: rewrite the SSI section, witness, "Not done", follow-ups and differential for the head. State the §416.420 dating as a methodology choice, as in item 2.

P2

5. The body has no impact: line

• gh api repos/PolicyEngine/policyengine-us/pulls/10332 --jq .body | grep -c '^impact:' prints 0.
• From the runs above, a line that matches the head would be:

  impact: mo_tanf falls by the fractional part of every grant in every year (under $1/month per unit); HI CCAP counts Hawaii TANF from the September 2021 month (higher copays and lower subsidies for TANF families, and some lose income eligibility); SNAP income limits for GU and VI households go from $0 to the contiguous-state limits (territories are not in the CPS); SSI and the readers of SSI countable income rise for food-only own-household recipients in the December 2024 payment month only (main counts food through 2024 and none from 2025). Hub microsim: MO TANF all years, HI CCAP 2021+, SSI 2024.
  

6. Missouri maximum grant: the earlier review's Critical 1 is unanswered

• The 2026-11-05 review says 13 CSR 40-2.310(13)(C) publishes whole-dollar maximum grants (136, 234, 292, 342, 388, 431, 474, 514, 554, 595, 635, 675 for sizes 1-12). On that reading, flooring 34.526% × standard of need pays $1 too little at sizes 1, 4, 5, 6, 7, 10 and 11. For example, size 1 is floor(393 × 0.34526) = floor(135.68718) = 135, where the quoted table says 136.
• I could not reach the regulation from this session, so I can't confirm the table.
• I did check its internal consistency: the quoted row equals the product rounded to the nearest dollar at all 12 sizes (my computation: 135.69→136, 234.09→234, 292.09→292, 341.81→342, 387.73→388, 430.54→431, 473.70→474, 514.09→514, 554.49→554, 594.54→595, 634.93→635, 675.33→675).
• If the table is the binding maximum, this PR moves size 1 from main's $135.69 to $135, away from $136, and pins 135 or 341 in 8 YAML places and in the property tests' _payment.
• The PR carries five commits after that review with no reply. The hub should check the source and resolve this before merge.

7. HI CCAP: Hawaii TANF stays excluded from 2017-08-01 to the September 2021 month

• sources.yaml:36-40 says "The predecessor 17-798.2-10(b)(11) lists it too, but the edition in force from August 2017 was not established."
• The same file already cites the 2013 DHS copy of HAR 17-798.2-10(b), and hi_ccap_countable_income.py describes the list as identical to 17-798.2-10(b) before August 6, 2021.
• The circularity that once justified leaving TANF out is now solved by hi_ccap_tanf_income. The earlier review asked for hi_ccap_tanf_income on the 2017 list; that is unanswered.
• I could not fetch the HAR history notes. If the hub keeps the 2021 start, the August 2021 exclusion in Case 13 is a methodology choice: the rule took effect August 6, and the monthly formula reads August 1. Record it as one.

8. HI CCAP: the TANF-abolition guard is untested

• hi_ccap_tanf_income.py:25-26. Mutant M7 (if parameters(period).gov.hhs.tanf.abolish_tanf: → if False:) fails no new case in hi_ccap_countable_income.yaml. Only Cases 13 and 14 fail, and they also fail at the unmutated head.
• Fix: add a case at 2025-01 with gov.hhs.tanf.abolish_tanf: true, hi_tanf: 100 and $12,000 of yearly earnings. Expect hi_ccap_tanf_income: 0 and hi_ccap_countable_income: 1_000.

9. The axiom: line still claims rulespec-us#1865

• The PR no longer changes the HI TANF deduction; it changes only dependent_care/amount.yaml metadata.
• Limit Hawaii, Vermont and Kentucky TANF dependent care deductions to the care their rules allow #10254, merged 2026-10-27, made the change and lists TheAxiomFoundation/rulespec-us#1865 queued (HI HAR 17-676-54.1, per-adult caps; ...) in its own axiom line.
• Fix: replace this PR's Integration Test of CARE/FERA Shows SNAP Issue #1865 entry with TheAxiomFoundation/rulespec-us#1865 carried by #10254 (no HI TANF deduction change here).
• I could not read the rulespec-us issues from this session.

P3

10. The Missouri float tolerance is smaller than float32 spacing above $128

• mo_tanf.py:28-39 snaps to a whole dollar only within atol=0.00001. float32 values are spaced 1.5e-5 apart in [128, 256) and 3.1e-5 apart in [256, 512), so some stored maxima carry more error than the tolerance.
• Witness: mo_tanf_standard_of_need: 1_123 and mo_tanf_countable_income: 287.72698 at 2026-01 give an exact deficit of $100. The model returns max 387.7269592285156 mo_tanf 99.0.
• A brute-force search over the 12 sizes in standard_of_need/amount.yaml and every whole-dollar deficit found 1,532 deficits that floor $1 low, at sizes 5, 8 and 11.
• These need incomes with five-decimal fractions, so real cents-valued incomes are not affected. But the comment's claim that the snap "preserves" whole-dollar deficits holds only below $128.
• Fix: round to a fixed precision before flooring, for example np.floor(np.round(capped_benefit, 4)). Then replace the test pinned to 106.99996 with one whose fraction survives rounding to 4 places. Alternatively, reword the comment to state the limit.

11. ssi_shelter_support_value changes meaning without a note

• The variable is now monthly; the label changed from "Actual annual value" to "Actual monthly value".
• The PR rewrote the test input from 1,200 to 100 (ssi_living_arrangement_integration.yaml Cases 8 and 13).
• A caller that sets it for a single month now supplies a monthly amount. Year-keyed inputs still split evenly: {2025: 1200} reads 100 in 2025-01 and 2025-07 (run above).
• The hand-off's partner flag names only ssi_countable_income and ssi_in_kind_support_and_maintenance. Add ssi_shelter_support_value and ssi_pmv_amount, and say so in the changelog.

12. A garbled comment in food_counts.yaml

• Lines 3-6 end "SSA values that support using the payment month's benefit rate." That is not a sentence.
• Suggested: # The PMV uses the payment month's federal benefit rate.

13. Follow-up: ca_state_supplement_person reads ISM at period.this_year

• ca_state_supplement_person.py:86-91. For split couples (a spouse eligible for CAPI), own countable income now averages 2024's 11 counted months over 12 in every 2024 month, while ssi_countable_income uses the month itself.
• The path and the read are identical on main, where ISM was annual. The PR makes the annual value a sum of months.
• I did not build a witness household for this path.

14. Overlap with open PRs

• Limit SSI spousal-deeming child allocations to the couple's co-resident children #10146 changes _apply_ssi_exclusions.py, ssi_in_kind_support_and_maintenance.py, ssi_pmv_amount.py, ssi_countable_income.py and ssi_resources_deemed_from_ineligible_parent.py.
• Round summed resources to whole cents before exact-limit asset tests #10639 changes ssi_resources_deemed_from_ineligible_parent.py and its YAML.
• Apply 416.1163(e)(2) to deemed SSI and identify SSI couples by marriage, not tax roles #10141 changes ca_capi_person.yaml.
• Whichever lands second has to absorb the MONTH definitions.

Checked and sound

• Merge state. No conflicts with current main; see the top.

• Scope. No file under .github/ or policyengine_us/tests/policy/baseline/partners/ (git diff --name-only e53b426082 pr-10332).

• Year-keyed inputs split evenly across months for the monthly variables (one simulation, TX, age 70):

  ssi_countable_income year-keyed 6000 -> 2025-01 500.0 2025-07 500.0
  ssi_in_kind_support_and_maintenance year-keyed 1200 -> 2025-01 100.0 2025-07 100.0
  ssi_shelter_support_value year-keyed 1200 -> 2025-01 100.0 2025-07 100.0 | ISM 2025-03 100.0
  

• Re-keyed test inputs.

  • I scanned every YAML test with a month period for bare inputs of ssi_countable_income, ssi_in_kind_support_and_maintenance, ssi_shelter_support_value, ssi_pmv_amount or ssi_pmv_applies. There are 43 bare inputs; 4 are nonzero amounts, and all 4 are monthly by intent: shelter values 100 and 500, and PMV 334.33 in the ISM file.
  • In the eight re-keyed IN, FL, MT, LA, CA CAPI and MO SAB files, every changed line is an ssi_countable_income or ISM input key, or a comment (git diff -U0). No expected value changed, and all eight files pass at the head and at main.

• Readers of the monthly SSI chain. I grepped every reader of these variables in policyengine_us/variables:

  • federal: uncapped_ssi, _ssi_state_supplement, ssi_resources_deemed_from_ineligible_parent;
  • Medicaid: is_medically_needy_for_medicaid, is_optional_senior_or_disabled_for_medicaid, medicaid_optional_senior_or_disabled_countable_income;
  • states: SC, CA (state supplement and CAPI), OK, LA, IA, FL, MD, ID, KY, CO (supplement and OAP), TX, NC, NY, VA and IN;
  • users of _apply_ssi_exclusions: MSP, CA WDP, OH, MN and others, which are unchanged in meaning (annual in, annual out).

  Each reads at its own period. Where months are constant, monthly and annual reads agree with main. The only averaging read is item 13.

• The defects reproduce at main (the PR's test files run against main's code):

  • MO TANF: mo_tanf.yaml 6 failed, integration.yaml 17 failed, and the other changed MO files also fail.
  • HI CCAP: hi_ccap_countable_income.yaml 5 failed, hi_ccap_copay.yaml 1 failed.
  • SSI: the ISM file 1 failed, ssi_living_arrangement_integration.yaml 7 failed, the deemed-resources file 1 failed.
  • SNAP: not reproduced by the committed tests (item 3); reproduced by the probe above (Guam snap_fpg is 0 at main).
  • For SSI against main, my probe gives food-only SSI of 628.67 in every month of 2024 at main (ISM 4,012). The head gives 628.67 January-November and 943 in December.

• Hand values, recomputed before reading the code.

  • SSI 2024: PMV = 943 / 3 + 20 = 334.333. Countable income = 314.333. SSI = 628.667. Annual: ISM 11 × 334.333 = 3,677.67; countable 3,457.67; SSI 11 × 628.667 + 943 = 7,858.33. These match Case 9f and my run.
  • SNAP FY2027, using the 2026 contiguous-state guideline parameters at 2026-10-01 ($15,960 + $5,680):
    • two people: net ceil(21,640 / 12) = 1,804, gross ceil(1.3 × 21,640 / 12) = 2,345;
    • ten-person gross: 6,037 + 2 × 616 = 7,269;
    • nine-person net: 4,644 + 474 = 5,118.
  • MO, size 2: 678 × 0.34526 = 234.08628 → 234. The deficits 10.836 → 10 and 9.986 → 0.
  • HI CCAP copay: 2,653.64 / (30,650 / 12) = 1.0390 → the 1% band; copay 26.5364; subsidy 473.4636.

• HI CCAP 2021 sources. The 2021-08-06 list is exactly main's 2017-08-01 list, in the same order, plus hi_ccap_tanf_income (checked with a YAML load). That includes military_retirement_income, which main had already moved the 2017 list to. I could not fetch the 2021 rule's text to check that item against it.

• Main's HI TANF version. Nothing in the tree reads hi_tanf_qualifying_incapacitated_adult_care_expenses (grep over .py, .yaml and .md). The PR still changes hi/dhs/tanf/deductions/dependent_care/amount.yaml, but only its description, label, reference and comments; no value changes.

• Tests at the head, one file per process.

  • At ed37709fbd, 26 of the 27 changed YAML files pass, and hi_ccap_countable_income.yaml fails 2 cases (item 1). At 2ed172e6d6 that file passes 14/14.
  • The three SNAP files are among the 26 (18, 20 and 4 passed), but they pass vacuously (item 3).
  • CI at the head passes every Python shard except shard 6 (item 1), including test_ssi_monthly_ism_properties.py, the two MO property tests and test_source_conflicts.py.

• Mutants, each run in its own process on a scratch worktree, restored afterwards (git status --porcelain empty):

  Mutant	Killed by
  M1 food date read at 1 January (the hand-off's)	ssi_living_arrangement_integration Cases 9b3, 9d2, 9f; property tests test_2024_food_income_changes_payments_from_december, test_continuing_food_support_counts_eleven_income_months and 2 more; ISM file 1 case
  M2 PMV × 12 (the hand-off's)	3 ISM cases and 13 ssi_living_arrangement_integration cases
  M3 annual ISM in monthly countable income (the hand-off's)	13 ssi_living_arrangement_integration cases
  M4 food receipt read in the income month (mine)	read-pin property test, test_months_match_..._within_the_year[2023], 8 integration cases (item 2)
  M5 Missouri floor without the tolerance	mo_tanf.yaml "A whole-dollar deficit survives float32 subtraction."
  M6 GU/VI mapping removed (mine)	survives all three SNAP files (item 3)
  M7 HI CCAP abolition guard disabled (mine)	survives (item 8)
  M8 concurrent food dating, as at d93c0dd3c2 (mine)	Cases 9b, 9b2, 9d, 9f

• Changelog. The fragment is top-level and typed fixed. Its text matches the head, apart from "Apply the SSI food exclusion by income month", which item 2 qualifies.

What I couldn't check

• Every legal source. The session's proxy refused or failed to resolve govinfo.gov, ecfr.gov, federalregister.gov, uscode.house.gov, sos.mo.gov, law.cornell.edu, humanservices.hawaii.gov, capitol.hawaii.gov, aspe.hhs.gov and ssa.gov. That leaves these unconfirmed:

  • 89 FR 21199's effective date and amended §416.1130;
  • how SSA applies the food exclusion under §416.420, and so which payment month first rises;
  • 7 CFR 273.9(a);
  • the 2026 HHS guidelines (I used the repository's parameters);
  • 13 CSR 40-2.310(13)(C) and (14);
  • HAR 17-798.3-10(b) and 17-798.2-10(b).

  I quote no law above from memory. The legal text I cite is the PR's, the earlier review's or the repository's, and I label it that way.

• The rulespec-us parity issues, which are outside this session's repository access.

• Population impact and microsimulation. I ran none, per the brief.

• Peak memory and timing of the property test in CI. Locally it took about 60 s of test time per run.

• Whether the main-versus-head differential from the hand-off reproduces at this head. It was a local script and was not committed.

---

Generated by Claude Code