[MaxGhenis]

SNAP items where the model followed USDA or state practice against the statute. Under the statute-first rule (where a statute conflicts with a form, instructions or agency practice, the model follows the statute; forms fill gaps only where the statute is silent), each moves to the statute.

Changes

• ABAWD rules in Alaska and Hawaii, P.L. 119-21 §10102. The section has no later effective date, and USDA's implementation memorandum applies the amended exceptions from enactment, 2025-07-04. Alaska and Hawaii announced 2025-11-01, the end of the quality-control exclusion period. A good-faith exemption under 7 U.S.C. 2015(o)(7) needs a request and a federal determination; Alaska's starts 2025-11-01, and none covering July to October 2025 was found for either state. Both hr1_in_effect parameters now start 2025-07-04, like the federal one, so August 2025 is the first affected month. Alaska's waived areas and its November 2025 to October 2026 good-faith window are unchanged.
• Large-household cap, 7 U.S.C. 2012(u)(2)(I). "For a household of 9 persons or more, an additional 22 percent per person, which additional percentage shall not total more than 200 percent." The FY2026 table already applies the amended ratios (its $218 increment is 22% of the $994 4-person maximum, rounded down) but lists no cap; the model followed it until FY2027. The cap now applies from 2025-10-01, under the annual October 1 computation in 7 U.S.C. 2012(u)(3)(C) and FNS’s September 4, 2025 implementation memorandum, page 2. FY2026 caps use USDA’s method: the published 8-person maximum plus ⌊200% × the unrounded 4-person Thrifty Food Plan cost⌋, using costs before whole-dollar allotment rounding from the USDA cost workbook and allotment-table bounds where needed. The six resolved caps are $3,777 for the contiguous states/DC, $4,885 for Alaska urban, $6,228 for Alaska rural 1, $7,581 for Alaska rural 2, $6,418 for Hawaii, and $4,856 for the Virgin Islands. Guam remains unresolved: its published amounts bound the operational cost to [$1,465.4545…, $1,465.5462…), permitting a $5,567 or $5,568 cap. The branch retains $5,567 pending an agency cost source; it does not claim that the bounds fix this floor. The same method reproduces all seven FY2027 published 18+ maxima.

Tests

• Alaska and Hawaii: July 2025 not in effect, August 2025 in effect; a 57-year-old working one hour a week fails the modeled work requirement in Anchorage in October 2025 and Hawaii in September 2025.
• FY2026 caps in every region; October 2025 capped, September 2025 not; contiguous and Hawaii 17/18-person boundaries, including Hawaii’s $6,379/$6,418 in September 2026; all FY2027 published 18+ maxima.
• Source-conflict candidate updates, including the separate California implementation-date follow-up.

Invariants

• is_snap_abawd_hr1_in_effect for Alaska and Hawaii equals the federal toggle in every month (both parameters now share its 2025-07-04 date).
• snap_max_allotment never exceeds the region's cap from FY2026, and equals the uncapped amount below 18 members (the cap binds only at 18 or more).

Review

The r1 and r2 approvals were read-only reviews; neither ran tests. A shell-backed GPT-6.1 Sol review ran 107 cases and approved ee645cd346. The Sol merge-delta review ran 131 cases and approved 2bf1c16a.

Impact

Hub microsimulation on the default dataset, main 79249f0fea against head ee645cd346, under the heavy lock. Weighted sums:

Year	Variable	Change	Weighted households changed	Largest change
2025	snap	−$2.75M	4,601 (7 records)	$3,323
2025	household_net_income, household_benefits	−$2.75M	4,601	$3,323
2025	snap_max_allotment	−$53.28M	44,858 (91 records)	$3,441
2026	all of the above	0	0	—

The 2025 SNAP change is the Alaska and Hawaii time-limit timing. The −$2.75M for August–October 2025 reflects the model’s immediate-ineligibility simplification in meets_snap_abawd_work_requirements, which does not track the clock for three countable months in 36 months. Under 7 U.S.C. 2015(o)(2) and FNS screening, the statute alone would produce little or no benefit loss before November 2025. The FY2026 18+ cap moves snap_max_allotment for large households; the table reports the combined microsimulation against the earlier head, before the regional rounding correction in this update. No default-dataset household reaches the cap in 2026.

axiom: TheAxiomFoundation/rulespec-us#1830 queued (pe-parity; every provision needs encoding)

🤖 Generated with Claude Code

[MaxGhenis]

Hand-off to the US + core hub (statute-first backlog session).

• Head: ee645cd346c0fa555b243b00c1a138ff6686f74d
• Independent review: GPT-6.1 Sol, r2 APPROVE (three comment-only nits left) at this head. Reviews: ~/reviews/us-hub/source-conflicts/backlog/reviews/10299-r*.md.
  • Correction (2026-10-20): dispatched without -s workspace-write, so the lane was read-only: both rounds (r1, r2) were served by Opus 5.5 in plan mode with no shell, not by GPT-6.1 Sol; they read files and fetched sources but ran no diff or tests. No review executed the tests. Treat these as reading reviews, not the landing gate; the hub has dispatched a shell-backed GPT-6.1 Sol review of this PR.
• Passed locally: the targeted YAML files and Hypothesis invariants named in the PR body (single-file runs, mutation-checked).
• Pending: CI (queued in the Actions backlog); the impact run, after the route-a hold (2026-10-21 04:00Z); the Axiom parity issue. The parity audit found every provision needs encoding. The dispatch-ready issue body is drafted at ~/reviews/us-hub/source-conflicts/backlog/parity/draft-S-10299.md for filing under the standing parity go, which also sets the axiom: line here.
• Depends on: nothing.
• Follow-ups owed after merge: register the squash commit in rulespec-us known-source-conflicts.yaml (once rulespec-us#1825 lands), then a follow-up PE-US PR with the # Source conflict marker, the pin bump and the candidates.yaml deletion, per HUB-PROTOCOL.

[DTrim99]

PR Review: Follow the SNAP statute on ABAWD timing in Alaska and Hawaii and the FY2026 large-household cap

PR #10299 (author: MaxGhenis). Head 2bf1c16a0d. The branch is 4 commits ahead of main (two authored commits, a merge of main, and a comment-only fix) and 3 behind. First review.

Summary

The PR makes two statute-first SNAP changes.

1. ABAWD timing in Alaska and Hawaii. The PR moves gov.states.ak.dpa...abawd.hr1_in_effect and gov.states.hi.dhs...abawd.hr1_in_effect from 2025-11-01 to 2025-07-04, the same date as the federal toggle. The model reads monthly parameters on the first of the month, so August 2025 is the first affected month.
2. The FY2026 large-household cap. The PR moves the 18+ cap in snap_max_allotment from FY2027 (formula_2026_10_01) back to FY2026 (formula_2025_10_01). It adds FY2026 cap values for all seven regions, each the published 8-person maximum plus twice the published 4-person maximum.

Both changes are right on the law.

• ABAWD timing. Section 10102 has no effective-date clause. FNS's September 4, 2025 information memo says of the §10102 changes: "These changes were effective upon enactment" (p. 3). FNS's ABAWD-exceptions memo says the same thing: "effective upon enactment, July 4, 2025. State agencies must immediately screen for and apply the modified exception criteria".
• Why the states said November 1. That is the end of FNS's QC hold-harmless period, "120 days from the implementation date" (July 4 + 120 days = November 1).
• The good-faith exemption doesn't cover July to October. The only statutory route to a delay for Alaska or Hawaii is the 7 U.S.C. 2015(o)(7) exemption. It applies only if "the State agency submits to the Secretary a request" and "the Secretary determines" good faith (P.L. 119-21, 139 Stat. 82, PDF p. 12). FNS's September 4 memo still called the request guidance "forthcoming". Alaska's approved window starts 2025-11-01. Hawaii DHS's October 17, 2025 release gives only "effective November 1, 2025" and mentions no exemption.
• FNS agrees with the statute. I found no conflict between the statute and FNS guidance on either change. The only conflicts are with state announcements and with USDA's FY2026 table, which printed no cap.
• The cap. 7 U.S.C. 2012(u)(2)(I) caps the additional percentage at 200 percent (139 Stat. 81, PDF p. 11). FNS dates the household-adjustment provision "effective October 1, 2025" (September 4 memo, p. 2).
• FY2026 cap values. All seven match the formula the PR states, using USDA's FY2026 table. I re-read that table at p. 1. One method choice is worth a look: urban Alaska comes out $1 below USDA's own FY2027 method (Suggestion 1).

CI is red, but main caused it, not this PR. Both failing jobs come from the source-conflict ratchet. It flagged five FL and SD LIHEAP files, none of them in this PR. #10334 fixed that on main after this CI run started. On head merged with current main, the ratchet passes. Merging main and re-running CI should turn it green (see CI status).

Open items. There is one should-fix item: rulespec-us#1830 gives a cap formula that contradicts its own FY2027 companion cases. There are six suggestions.

Critical (must fix)

None.

Should address

1. rulespec-us#1830 tells the encoder to apply a cap formula to FY2027 that its own FY2027 cases contradict.

• Where: the PR body's axiom line, and rulespec-us#1830, section S2, "Required outputs".
• What's right:
  • The issue is OPEN and labelled pe-parity.
  • It names the modules. Two existing ones, us/regulations/7-cfr/273/24.yaml and us/policies/usda/snap/fy-2026-cola/maximum-allotments.yaml, are both present in rulespec-us. Two new targets are to be created: us/statutes/7/2012/u/2/I.yaml and us/policies/usda/snap/fy-2027-cola/maximum-allotments.yaml.
  • It quotes §10101 and §10102 verbatim, along with both FNS effective-date statements.
  • Its S1 cases and its FY2026 S2 cases match this head: 3,751 / 3,777 / 3,777, and 4,884, 6,228, 7,581, 5,567, 6,418, 4,856, plus the 6,596 control for 2025-09.
• What's wrong:
  • S2 says to "wire it into FY2026 and FY2027 regional maximum-allotment computations" with maximum = published_M8 + min((household_size-8)*published_increment, 2*published_M4).
  • Its FY2027 expectations, though, are USDA's printed maxima, and the formula does not give them in three regions:
    • Alaska urban: 2,352 + 2 × 1,306 = 4,964. The issue expects 4,965.
    • Alaska rural 2: 3,650 + 2 × 2,027 = 7,704. The issue expects 7,705.
    • Guam: 2,714 + 2 × 1,507 = 5,728. The issue expects 5,729.
  • So six FY2027 companion cases (sizes 18 and 30 in those regions) fail under the issue's own formula, and an encoder can't satisfy both. The PR body itself notes this $1 gap.
  • The issue also names no blocker. It says only "This read-only audit did not dispatch the signed encoder; no technical encoder failure is asserted". CLAUDE.md says to use queued "only when the signed encoder is blocked; record the blocker in the issue."
• Fix: comment on rulespec-us#1830, or edit its body, to say:
  • (a) The published-amount formula applies to FY2026 only, where USDA printed no cap. FY2027 uses USDA's printed 18+ maxima. Equivalently, FY2027 is M8 + ⌊200% × the unrounded 4-person cost⌋, which reproduces all seven FY2027 caps.
  • (b) If Suggestion 1 is adopted, the FY2026 Alaska urban expectation becomes 4,885.
  • (c) Add one line naming what blocks dispatch, for example a billed encoder run awaiting approval, or ingest of the FY2026 and FY2027 tables.

Suggestions

1. FY2026 cap basis: urban Alaska is $1 below USDA's own method.

• Where: parameters/gov/usda/snap/max_allotment.yaml, cap.AK_URBAN 2025-10-01: 4_884, and the comment block above it.
• The PR's formula mixes two bases. It adds the 8-person maximum, which USDA computes from the unrounded 4-person cost, to twice the rounded 4-person maximum.
  • For urban Alaska, 180% × $1,285 = $2,313, but USDA printed $2,314. So USDA applies the ratios to an unrounded cost, not to the printed $1,285.
  • The published FY2026 amounts pin that cost to [$1,285.56, $1,285.71). The lower bound comes from 2,314 / 1.8 and the upper bound from 1,530 / 1.19 for the 5-person $1,529.
• USDA's FY2027 method is the 8-person maximum + ⌊200% × the unrounded 4-person cost⌋. It reproduces all seven printed FY2027 caps, including the three where the PR's formula is $1 low.
• Applied to FY2026, that method gives:
  • Alaska urban: $2,314 + ⌊2,571.1–2,571.4⌋ = $4,885, where the PR has $4,884.
  • Contiguous, Alaska rural 1, Alaska rural 2, Hawaii and the Virgin Islands: the same as the PR.
  • Guam: indeterminate ($5,567 or $5,568), because the bounds straddle a whole dollar.
• Fix: either set AK_URBAN to 4_885 and update fy2027_cola.yaml ("FY2026 caps in the other regions", first output), or keep 4_884 and say in the comment that FY2026 deliberately reads "percentage of the maximum 4-person allotment" as the printed amount. Either is defensible. The effect is $1 a month for urban-Alaska households of 18 or more.

2. Cite FNS's date for the cap.

• §10101 has no effective-date clause of its own. Applying the PR's AK/HI reasoning ("no later effective date") by analogy would point to July 4, 2025, not October 1.
• The October 1, 2025 start is still right. It rests on (u)(3)(C)'s annual October 1 computation and on FNS's statement that the household-adjustment provision "is effective October 1, 2025" (September 4, 2025 memo, p. 2). rulespec-us#1830 already cites this.
• Fix: add that reference to max_allotment.yaml, and add one comment line to snap_max_allotment.formula_2025_10_01 saying why October 1, 2025 is the start.

3. Three backlog candidates in source_conflicts/candidates.yaml go stale on merge.

• Which entries: us-ak:snap-abawd-implementation-date, us-hi:snap-abawd-implementation-date and us:snap-large-household-cap-effective-date.
• What they say: each still has follows: conflicting-source, code: [], and a note that the code is "not yet on main".
• The author's plan: delete them after registering the conflicts once rulespec-us#1825 lands. That is a documented deferral.
• Fix: updating pe_us.follows to statute and listing the code paths in this PR would keep the public backlog accurate in the meantime. Note that Point the FL and SD LIHEAP conflict candidates at their code on main #10349, Add New Hampshire state supplementation for 2026 #10156 and Add Iowa state supplementation for 2026 #8017 also edit candidates.yaml.

4. California's 2026-06-01 delay has the same kind of basis as the AK/HI dates this PR removes.

• What ACL 25-93 says: "Due to the complexity of these policy changes and other factors beyond CDSS's control, the guidance provided via this letter is effective June 1, 2026. This is the earliest possible implementation date that reflects the time required to design and implement the necessary automation changes within the California Statewide Automated Welfare System" (p. 3).
• Why it matters: the PR keeps CA's delay in is_snap_abawd_hr1_in_effect and in the README. candidates.yaml has no CA ABAWD-timing entry.
• Fix: this is outside the PR's scope. Add a us-ca:snap-abawd-implementation-date candidate, or a README note on why CA is treated differently.

5. PR body.

• (a) Caveat the impact figure.
  • The 2025 SNAP change (−$2.75M, August to October 2025) comes from the model's documented immediate-ineligibility simplification in meets_snap_abawd_work_requirements: it doesn't track the 3-in-36 clock.
  • Under 7 U.S.C. 2015(o)(2), a newly covered ABAWD has three countable months, and FNS applies the new exceptions at screening. So the statute alone would produce little or no benefit loss before November 2025.
  • One sentence saying so would prevent misreading.
• (b) Fix the Review section. It still credits "GPT-6.1 Sol" with the r1 and r2 approvals. The author's 2026-10-20 comment corrects that: both were reading reviews served by another model, with no tests run.

6. Optional tests and naming.

• Hawaii person-level case. Add one for August to October 2025, for example a 57-year-old in 2025-09 working one hour a week, giving meets_snap_abawd_work_requirements: false. Hawaii's area waiver ended 2025-06-30, so this exercises the new date end to end.
• Noncontiguous 17/18 boundary. Add one, for example Hawaii at 2026-09 with 17 → 6,379 (3,040 + 9 × 371) and 18 → 6,418. The cap binds at 18 in every region: all seven 17-person amounts fall below their caps.
• Rename is_snap_abawd_hr1_in_effect.yaml Case 10 ("HI in effect at state adoption date") to match the new framing.

Verified correct

• Statute. I checked P.L. 119-21 against the enrolled text (139 Stat. 80–83):
  • §10102(a) replaces (o)(3), with the exception for those "under 18, or over 65", a dependent child under 14, and Indian, Urban Indian and California Indian exceptions.
  • §10102(b) adds the 1.5× noncontiguous unemployment criterion.
  • §10102(c) adds (o)(7): request plus determination, expiring no later than December 31, 2028.
  • §10101 adds (u)(2)(I), with the 200 percent limit.
  • Neither section has an effective-date clause.
• FNS guidance. The September 4, 2025 information memo says the §10102 changes were "effective upon enactment" (p. 3) and gives October 1, 2025 for the household adjustments (p. 2). It holds states harmless for QC "for 120 days from the implementation date", which ends November 1, 2025. I read it from a state-agency copy because usda.gov blocks automated fetches. The ABAWD-exceptions memo's "effective upon enactment, July 4, 2025" is corroborated by its search snippet and by rulespec-us#1830's quote. I couldn't fetch the PDF itself, so the #page=4 anchor isn't independently checked.
• AK/HI parameters. Both now read 2025-07-04: true, the same as gov.usda.snap.work_requirements.abawd.in_effect, so the two states equal the federal toggle in every month.
  • Alaska's waived areas are unchanged: the FY2025 waiver covers everything except Anchorage through 2025-10-31, and the FY2026 waiver plus the good-faith exemption run from 2025-11-01 to 2026-10-31.
  • Alaska's good-faith window (2025-11-01 to 2026-10-31, retained age minimum 56 from 2025-11-01) is untouched.
  • Hawaii's area waiver ended 2025-07-01.
• Formula.
  • is_snap_abawd_exempt switches age (65 under HR1, 55 before it) and the exception sets on is_snap_abawd_hr1_in_effect, so the date change flows through.
  • is_snap_abawd_hr1_in_effect uses select on state_code.
  • formula_2025_10_01 uses min_ with a snap_region_str-indexed parameter.
  • Both are vectorized, and the cap parameter has values for all seven regions from 2025-10-01.
• Callers. Nothing else references formula_2026_10_01 or max_allotment.cap, so the rename is safe.
• NH Add New Hampshire state supplementation for 2026 #10156 lesson. No state-gated variable is used as a parameter or breakdown key. state_code and snap_region_str aren't gated. The OK Add Oklahoma state supplementation for 2026 #10165, Integrate modeled state unemployment insurance into income flows #9558 and Document disjoint health premium inputs and correct consumer treatment #10046 lessons don't apply.
• FY2026 table (p. 1). The 4-person, 8-person and per-person amounts match the model in all regions: 994 / 1,789 / 218; 1,285 / 2,314 / 282; 1,639 / 2,950 / 360; 1,995 / 3,591 / 438; 1,465 / 2,637 / 322; 1,689 / 3,040 / 371; 1,278 / 2,300 / 281. The table has no cap row, as the PR says. The $218 increment is ⌊0.22 × 994⌋.
• PR-body claim about FY2027. "Matches in four regions and exceeds by $1 in three" is confirmed: Alaska urban, Alaska rural 2 and Guam.
• Hand-checked expectations (12):
  1. fy2027_cola.yaml, FY2026 contiguous, size 17: 1,789 + 9 × 218 = 3,751, under the 3,777 cap.
  2. Size 18: min(1,789 + 10 × 218 = 3,969, 3,777) = 3,777.
  3. Size 30: min(1,789 + 22 × 218 = 6,585, 3,777) = 3,777. The cap is 1,789 + 2 × 994.
  4. Alaska urban, size 30: min(2,314 + 22 × 282 = 8,518, 2,314 + 2 × 1,285 = 4,884) = 4,884.
  5. Alaska rural 1: 2,950 + 2 × 1,639 = 6,228. Alaska rural 2: 3,591 + 2 × 1,995 = 7,581.
  6. Guam: 2,637 + 2 × 1,465 = 5,567. Hawaii: 3,040 + 2 × 1,689 = 6,418. Virgin Islands: 2,300 + 2 × 1,278 = 4,856.
  7. 2025-10, size 18: 3,777. formula_2025_10_01 applies and the parameters read 2025-10-01.
  8. 2025-09, size 30: the base formula applies, with no cap: 1,756 + 22 × 220 = 6,596.
  9. is_snap_abawd_hr1_in_effect Cases 8 and 9 (HI, AK, 2025-08): true, because 2025-07-04 ≤ 2025-08-01.
  10. New HI and AK 2025-07 cases: false, because the parameter read at 2025-07-01 is before 2025-07-04. Cases 5 and 6 (2025-06) stay false.
  11. meets_snap_abawd_work_requirements Case 57 (Anchorage, age 57, 2025-10): HR1 is on, and age 57 is not over the HR1 bound of 65; before HR1 it would have been exempt at 55+. The good-faith window isn't open until 2025-11-01, and Anchorage was outside the FY2025 waiver. Work is 1 × 52 / 12 = 4.33 hours, below 80, so the result is false.
  12. In every region, the 17-person amount is below the FY2026 cap (Alaska urban 4,852, Alaska rural 1 6,190, Alaska rural 2 7,533, Guam 5,535, Hawaii 6,379, Virgin Islands 4,829). The cap binds only from 18, as the PR body says.
• References.
  • Each new reference is a single source with a single #page= anchor and no page text in the title.
  • The govinfo P.L. page 11 shows (u)(2)(I).
  • The FY2026 table is on page 1. Its URL returned a 404 once during this review, then 200 on retry.
  • The cap description still follows the house template.
• Comments vs the source-conflict guard. The PR's six scanned files have zero comment lines matching the guard's conflict-sounding pattern, at head and on main.
• Changelog and partner tests. changelog.d/statute-first-snap.fixed.md is top-level, and fixed fits. No partner test files are touched. No Python tests are added, so CI cost is unchanged.

Axiom line assessment

The line axiom: TheAxiomFoundation/rulespec-us#1830 queued (pe-parity; every provision needs encoding) has the right form.

• What's in place. rulespec-us#1830 is OPEN and labelled pe-parity. Its last update was 2026-10-20 03:52 UTC. That predates the merge commit and the comment-only last commit, but neither changes behavior. The issue gives module paths, corpus citations, verbatim law, required outputs and companion cases taken from the statute and USDA tables.
• Match with this head. S1 (the AK/HI amendment in force from 2025-07-04) and S2's FY2026 values match this head.
• What's not dispatch-ready. S2's formula contradicts its own FY2027 cases in three regions, and the issue records no blocker. See Should 1.

CI status

Per gh pr checks 10299, re-checked at the end of this review: 25 pass, 2 fail, none pending. All are in run 39535154699, which tested merge ref b5fb45aef8: this head on top of main at 7545b2c92f.

• Job 119015332185, "Full Suite - Rest (Python, shard-1 + variables)": test_source_conflicts.py::test_conflict_sounding_comments_do_not_grow failed. This is a real failure, not infrastructure. The assertion lists five files: fl_liheap_eligible.py, fl_liheap_income_limit.py, fl_liheap.py, FL LIHEAP payment/base_amount.yaml, and SD LIHEAP income/limits/table/in_effect.yaml. None is in this PR.
• Job 119036629410, "Full Suite - Rest (Python + variables)": this is the umbrella job. Its log ends "Rest shards did not all succeed: shard-1=failure, shard-2=success", so it only mirrors the shard failure.
• Cause: main, not this PR.
  • Count the FL and SD LIHEAP comments in the source-conflict ratchet baseline #10334 (fec98b87c8, merged 2026-10-21 13:24 UTC, after this run) added exactly those five files to KNOWN_UNMARKED. Its message says the test "fails on main from ce9377e".
  • On head merged with current main (git merge-tree tree 78544b6515), I ran only that test file, with conftest disabled. test_conflict_sounding_comments_do_not_grow passes.
  • Two other tests in the file failed locally: the register-pin sha256 check (CRLF line endings on Windows) and the tokenizer check (local Python 3.11 against CI's 3.14). Both pass in CI.
• Fix: merge current main into the branch, which brings in Count the FL and SD LIHEAP comments in the source-conflict ratchet baseline #10334, and re-run CI.

Branch status

• The branch is 4 commits ahead of main and 3 behind. The missing commits are Count the FL and SD LIHEAP comments in the source-conflict ratchet baseline #10334 and two version bumps.
• git merge-tree --write-tree PolicyEngine/main 2bf1c16a0d is clean (tree 78544b6515), with no conflicts.
• The merge commit 4727d9e5e0 was a clean merge of 7545b2c92f.
• Sibling note: open PRs Point the FL and SD LIHEAP conflict candidates at their code on main #10349, Add New Hampshire state supplementation for 2026 #10156 and Add Iowa state supplementation for 2026 #8017 touch source_conflicts/candidates.yaml. This PR doesn't, so there is no conflict unless Suggestion 3 is taken. No open sibling PR touches the SNAP files this PR changes.

Verdict: REQUEST_CHANGES

[DTrim99]

Both changes follow the law: OBBBA s.10102 applies on enactment in AK/HI, and the FY2026 cap for households of 18 or more matches FNS's table. The two red checks come from main (the FL/SD LIHEAP entries in the source-conflict ratchet, fixed by #10334), so merge main and re-run. One should-address item: rulespec-us#1830's S2 formula (M8 + 2xM4) disagrees by $1 with its own FY2027 cases in AK urban, AK rural 2 and Guam, and the issue doesn't record a blocker. Six suggestions. Full report in the comment above.

[DTrim99]

PR Review (round 2): Follow the SNAP statute on ABAWD timing in Alaska and Hawaii and the FY2026 large-household cap

PR #10299 (author: MaxGhenis). Head 32111093a4, last reviewed 2bf1c16a0d. Since then there is one clean merge of main (eb7fd3e2dc, no hand resolutions) and one author commit (32111093a4, "Address SNAP review on cap rounding and source conflicts"). The author didn't re-request review; this re-review is on our own initiative.

Summary

Every round-1 item is addressed.

• Should 1. The author posted a correction on rulespec-us#1830 (2026-10-22 08:55 UTC, after the head commit). It replaces the published-amount formula with USDA's method, M8 + min((n − 8) × increment, ⌊2 × unrounded C4⌋). That method reproduces all seven FY2027 printed caps, including the three where the old formula was $1 low. The correction also sets urban Alaska's FY2026 cap to 4,885, adds the new Hawaii cases and records a blocker. The issue is still OPEN and labelled pe-parity.
• Suggestion 1. Urban Alaska's FY2026 cap is now $4,885. The parameter comment records the bound intervals behind each region's cap. I re-derived every interval from USDA's FY2026 table and they all match. Guam is honestly documented as unresolved ($5,567 or $5,568), and $5,567 is retained.
• Suggestions 2 to 6.
  • The FNS September 4, 2025 memo is now cited, with a comment line on the formula.
  • The three stale candidates now read follows: statute and list the code.
  • A California candidate is added.
  • The PR body carries the impact caveat and the corrected Review section.
  • The Hawaii person-level case and the Hawaii 17/18 boundary are added, and Case 10 is renamed.

The cap method checks out against USDA's own numbers. I opened the cited Thrifty Food Plan workbook:

• Cell I1552 is $994.40 for June 2025 and cell I1904 is $1,023.20 for June 2026.
• Its ReadMe says costs are "rounded to the nearest 10 cents".
• Its only rows are "Mainland U.S.".

USDA's printed FY2027 Alaska rural 2 cap of $7,705 rules out a single ⌊3.8 × C4⌋ rounding, which would give $7,706. So the PR's M8 + ⌊2 × C4⌋ is USDA's method.

There are no critical or should-fix findings. There are four new suggestions, all about wording outside the model: the #1830 issue body, the changelog, the PR body's impact attribution, and one candidate note.

CI needs a re-run of one job. The ssa-usda baseline shard was cancelled when it hit GitHub's one-hour execution limit, and its log is unavailable. This looks like infrastructure, not this PR:

• The same shard took 16–19 minutes on this PR's three earlier runs.
• It took 20–26 minutes on four other PRs today.
• Quick Feedback ran all four changed SNAP test files at this head, and all 109 cases passed.

Prior findings

Finding	Status	Evidence
Should 1. rulespec-us#1830 S2 formula (M8 + 2×M4) contradicted its own FY2027 cases in urban Alaska, Alaska rural 2 and Guam; no blocker recorded	Fixed	See "Should 1 in detail" below the table. Residual wording is in new Suggestion 1.
Sugg 1. FY2026 urban Alaska $4,884 vs USDA's method $4,885; Guam indeterminate	Fixed	max_allotment.yaml cap.AK_URBAN 2025-10-01: 4_885. The comment lists each region's C4 interval (upper ends excluded), and all of them match my exact-fraction derivation from the FY2026 table. For example, urban Alaska is [1,285.5556 (8-person), 1,285.7143 (5-person)), so ⌊2C4⌋ = 2,571 and the cap is 2,314 + 2,571 = 4,885. Guam's [1,465.4545, 1,465.5462) straddles 1,465.50. The PR keeps $5,567 and says the bounds don't fix the floor. That is a reasoned, documented deferral.
Sugg 2. Cite FNS's October 1, 2025 date	Fixed	New reference "FNS SNAP Provisions of the One Big Beautiful Bill Act of 2025, September 4, 2025" (#page=2, where "This provision is effective October 1, 2025" appears). The Cornell link now covers (u)(2)(I) and (u)(3)(C). formula_2025_10_01 has a comment saying why it starts October 1, 2025.
Sugg 3. Three candidates.yaml entries stale on merge	Fixed	In us-ak: and us-hi:snap-abawd-implementation-date and us:snap-large-household-cap-effective-date: follows: statute, the code paths are listed, PR 10299 is added to prs, and the notes and examples are rewritten. The test_candidate_fields rules (FOLLOWS enum, repository paths, PR URL pattern) are met, and Rest shard-1 passed at this head.
Sugg 4. California's 2026-06-01 delay has no candidate	Fixed	New us-ca:snap-abawd-implementation-date: statute plus ACL 25-93 #page=3, follows: conflicting-source, state: open, with a note that the change is out of scope.
Sugg 5. PR body: (a) impact caveat; (b) Review section	Fixed	(a) The body now says the −$2.75M "reflects the model's immediate-ineligibility simplification … the statute alone would produce little or no benefit loss before November 2025." (b) The body now says "The r1 and r2 approvals were read-only reviews; neither ran tests."
Sugg 6. Hawaii person-level test, noncontiguous 17/18 boundary, Case 10 name	Fixed	meets_snap_abawd_work_requirements.yaml Case 60 (Hawaii, 2025-09, age 57). fy2027_cola.yaml "FY2026 Hawaii allotment reaches the cap at 18 members" ([6,379, 6,418]). Case 10 is renamed "HI remains in effect at its announced November 2025 implementation date".

Should 1 in detail. The correction comment on rulespec-us#1830:

• Method. It gives M8 + min((n − 8) × published_increment, floor(2 × unrounded_C4)) and asks the encoder to replace the old cases.
• FY2027 check. I re-derived C4 bounds from USDA's FY2027 sizes 1–8 and increment. All seven floors are determinate and equal the printed caps: 3,887 / 4,965 / 6,331 / 7,705 / 5,729 / 6,289 / 4,997.
• FY2026 values. Its FY2026 table matches this head, with urban Alaska at 4,885. It adds the Hawaii 2026-09 cases (6,379 / 6,418) and the September 2025 ABAWD case.
• Blocker. Section (c) records one: Guam's cost "straddles $1,465.50 … Signed-encoder dispatch waits on that source … No technical encoder failure is claimed."
• Issue state. OPEN, labelled pe-parity, updated 2026-10-22 08:55 UTC.

New findings

Critical (must fix)

None.

Should address

None.

Suggestions

1. rulespec-us#1830: fold the correction into the issue body.

• The problem. The fix lives in a comment, but the body's pasteable review_finding block still says maximum = published_M8 + min((household_size-8)*published_increment, 2*published_M4) and "AK Urban … ->4884". A dispatcher who pastes that block gets the superseded formula.
• Edit the body to replace the formula and the 4,884 case, as the comment already instructs.
• Name the state in the ABAWD case. The comment's "September 2025 ABAWD case: age 57 working one hour a week gives false" doesn't say Hawaii, but the state is the point of the case. Say Hawaii, and note that its area waiver ended 2025-06-30.
• Say what the noncontiguous inputs are. No agency C4 exists outside the contiguous states: the workbook has only "Mainland U.S." rows. The encoder's inputs there are the derived ⌊2C4⌋ values for FY2026: 2,571, 3,278, 3,990, 3,378 and 2,556 (Guam pending). For FY2027 they are USDA's printed caps.

2. Changelog wording.

• changelog.d/statute-first-snap.fixed.md ends "…correcting urban Alaska's FY2026 cap to $4,885."
• main has never had an FY2026 cap, so "correcting" suggests a released value that never existed.
• Suggested wording: "…which sets urban Alaska's FY2026 cap at $4,885."

3. PR body: say which change drives the 2025 snap_max_allotment row.

• What the body says now. It says "The FY2026 18+ cap moves snap_max_allotment for large households" and also "No default-dataset household reaches the cap in 2026".
• Why that's inconsistent. If no household reaches the FY2026 cap in January to September 2026, the same records almost certainly don't reach it in October to December 2025.
• The likely real cause. Members who fail the work requirement drop out of snap_unit_size through is_snap_excluded_member → meets_snap_work_requirements_person. So the 2025 row (−$53.28M, 91 records) most likely comes from the Alaska and Hawaii ABAWD timing, not the cap.
• Fix. One clause saying so, or a split run, would prevent misattribution. This is not verifiable without a microsimulation, which I didn't run.

4. One candidate note's wording (optional).

• The us-ca: note says 7 U.S.C. 2015(o)(7) "names only Alaska and Hawaii". In fact (o)(7)(A) defines a "noncontiguous State" as one that is not one of the contiguous 48 states or DC, and excludes Guam and the Virgin Islands. So only Alaska and Hawaii qualify, but the statute doesn't name them. "Covers only noncontiguous States, which excludes California" is exact.
• The "change is in flight" notes in the three updated entries go stale on merge. That is fine, given the author's plan to delete the entries once rulespec-us#1825 lands.

Verified

Hand-checked expectations, covering every new or changed one at this head:

1. fy2027_cola.yaml "FY2026 caps in the other regions", urban Alaska, size 30 (changed). min(2,314 + 22 × 282 = 8,518, 4,885) = 4,885.
2. Same case, the other five regions (unchanged, rechecked against the new comment):
   • Alaska rural 1: min(10,870, 6,228) = 6,228.
   • Alaska rural 2: min(13,227, 7,581) = 7,581.
   • Guam: min(9,721, 5,567) = 5,567.
   • Hawaii: min(11,202, 6,418) = 6,418.
   • Virgin Islands: min(8,482, 4,856) = 4,856.
3. New "FY2026 Hawaii allotment reaches the cap at 18 members", size 17: 3,040 + 9 × 371 = 6,379, below 6,418.
4. Same case, size 18: min(3,040 + 10 × 371 = 6,750, 6,418) = 6,418.
5. New meets_snap_abawd_work_requirements Case 60 (Hawaii, 2025-09, age 57, 1 hour a week, not disabled), result false:
   • HR1 is in effect, because the parameter reads true at 2025-09-01 ≥ 2025-07-04.
   • The age bracket exempts from 65 (2025-07-04: 65), so age 57 is not exempt.
   • Hawaii has no good-faith window, since is_snap_abawd_in_good_faith_exemption_window covers only Alaska.
   • There is no waived area: Hawaii's waivers ended 2025-06-30, before the 2025-07-03 snapshot.
   • Work is 1 × 52 / 12 = 4.33 hours, below 80.
   • The case discriminates: under the old 2025-11-01 date, the pre-HR1 age-55 bound would exempt the person and give true.
6. is_snap_abawd_hr1_in_effect Case 10 (Hawaii, 2025-11): only the name changed; the result is still true.
7. Contiguous FY2026 comment (values unchanged): C4 lies in [994.35, 994.4056), the workbook's 10-cent rounding intersected with the table bounds. So ⌊2C4⌋ = 1,988 and 1,789 + 1,988 = 3,777. Sizes 17/18/30 give [3,751, 3,777, 3,777].
8. The cap binds only from 18 at the new urban Alaska value: size 17 is 2,314 + 9 × 282 = 4,852, below 4,885.

Parameter comment intervals. Exact-fraction bounds from FY2026 sizes 1–8 (ratios 30/55/79/100/119/143/158/180 percent) and the 22 percent increment reproduce every interval in the comment:

Region	C4 interval	⌊2C4⌋
Urban Alaska	[1,285.5556, 1,285.7143)	2,571
Alaska rural 1	[1,639.2405, 1,639.4444)	3,278
Alaska rural 2	[1,995.1049, 1,995.4545), upper end from the increment	3,990
Hawaii	[1,689.0909, 1,689.2405)	3,378
Virgin Islands	[1,278.1818, 1,278.3217)	2,556
Guam	[1,465.4545, 1,465.5462)	2,930 or 2,931

The method is USDA's. For FY2027, M8 + ⌊2C4⌋ gives all seven printed caps. ⌊3.8 × C4⌋ gives 7,706 for Alaska rural 2, where C4 lies in [2,027.97, 2,028.00), against USDA's printed 7,705. So USDA rounds the 200 percent cap separately, as the PR does.

References.

• TFP workbook. It returned HTTP 200 and is a single source. Its Thrifty 21-Pres sheet I1552 = 994.4 and I1904 = 1023.2, the reference family's monthly cost for June 2025 and June 2026.
• FY2027 memo. Its two #page=4 / #page=5 entries are merged into one, titled "... memo #page 4-5", quoted, with the href opening page 4. That matches the CLAUDE.md reference rules.
• FY2026 table. The parenthetical page text is gone from its title.
• FNS memo. It has a single #page=2 entry.

Code.

• formula_2025_10_01 is unchanged apart from comments.
• No other test references AK_URBAN or an 18+ SNAP unit, so the 4,884 → 4,885 change has no other callers.
• The NH Add New Hampshire state supplementation for 2026 #10156, OK Add Oklahoma state supplementation for 2026 #10165, Integrate modeled state unemployment insurance into income flows #9558 and Document disjoint health premium inputs and correct consumer treatment #10046 lessons don't apply. No state-gated key variable is involved, and there is no income-source change.

Source-conflict ratchet. CI at this head ran after #10287 and #10334, on the merge ref with main at 86d3ae6c6d, and Rest shard-1 passed. A static scan of the six changed parameter and variable files with the guard's CONFLICT_COMMENT_RE finds 0 matching comment lines at head and on current main. None of those files is in KNOWN_UNMARKED. The only main change since then to the test file adds an unrelated IRS entry.

Merge commit. eb7fd3e2dc merged 86d3ae6c6d. git show --remerge-diff is empty, so there are no hand resolutions.

Changelog and partner tests. The changelog fragment is top-level with type fixed. No partner test files are touched. No Python tests are added.

Local runs. None, under the low-RAM rules. The arithmetic was done with exact fractions outside the model, and test results come from CI.

Axiom line assessment

The line is unchanged: axiom: TheAxiomFoundation/rulespec-us#1830 queued (pe-parity; every provision needs encoding).

• Valid queued line. Lower SSI resource test pass rate #1830 is OPEN and labelled pe-parity. Read together with its 2026-10-22 correction comment, it matches this head:
  • The method and FY2026 caps are identical to max_allotment.cap.
  • The FY2027 cases match USDA's printed maxima under the stated method.
  • The S1 ABAWD cases and the new Hawaii cases match the new tests.
• Blocker. It is recorded: no agency source gives Guam's four-person cost to decide $5,567 against $5,568.
• Remaining polish. The pasteable body block is stale (new Suggestion 1).

CI status

gh pr checks 10299, re-checked at the end of this review: 17 pass, 1 fail, 1 skipping (Contrib tests, path-filtered), none pending. All are in run 39745568824, which tested head 32111093a4 on the merge ref with main at 86d3ae6c6d. No re-run had been started.

The failing job is infrastructure, not a test failure.

• What happened. Job 119713234687, "Full Suite - Baseline (ssa-usda)", ended cancelled with the annotation "The job has exceeded the maximum execution time of 1h0m0s". Its test step never finished, and the job log returns BlobNotFound.
• Why it isn't this PR.
  • The same shard passed in 16–19 minutes on this PR's runs at ee645cd346 and 2bf1c16a0d.
  • It passed in 20–26 minutes on four other PR runs on 2026-10-22 against the same main.
  • Since the last green run, the PR changed one cap value, some comments, two small YAML cases and candidates.yaml.
  • Quick Feedback ran every changed SNAP test file at this head, and all passed: is_snap_abawd_hr1_in_effect.yaml 18, meets_snap_abawd_work_requirements.yaml 60, state_coverage.yaml 5, fy2027_cola.yaml 26.
• Action before merge. Re-run the ssa-usda job, or merge main, which re-triggers CI, and confirm it is green.

Branch status

• The branch is 6 commits ahead of main and 3 behind. The missing commits are Take the 2020 charitable deduction for non-itemizers above the line #10186, a version bump and Add Georgia LIHEAP regular heating assistance #9968.
• git merge-tree --write-tree PolicyEngine/main 32111093a4 is clean (tree cc91cc70a9).
• Sibling note: open PR Point the federal and AK-HI source-conflict candidates at their code on main #10350, "Point the federal and AK-HI source-conflict candidates at their code on main", also by MaxGhenis, edits the same three candidates.yaml entries. It describes main's current November 1 and FY2027 behavior, the opposite of this PR. git merge-tree of the two heads conflicts in source_conflicts/candidates.yaml. Whichever PR lands second needs to reconcile those entries, and Point the federal and AK-HI source-conflict candidates at their code on main #10350's text for them becomes wrong once this PR merges.
• Open PRs Add Arkansas LIHEAP regular heating assistance #9973, Add Iowa state supplementation for 2026 #8017, Limit Alabama, North Carolina and Hawaii mortgage interest to the amounts their laws allow #10274, Point the FL and SD LIHEAP conflict candidates at their code on main #10349, Exclude county agency staff from Minnesota MSA guardian fees #10315, Support state-specific other additions and subtractions for AL, AR, IA, IN, LA, MS, PA (Phase 3) (#10197) #10354, Point the NM-WY source-conflict candidates at their code on main #10352 and Point the IN-MT source-conflict candidates at their code on main #10351 also edit candidates.yaml, but they merge cleanly with this head.

Verdict: APPROVE

[DTrim99]

Round 2: Should 1 is fixed. rulespec-us#1830 now uses USDA's M8 + floor(2 x C4) method, which reproduces all seven FY2027 printed caps, sets AK urban FY2026 to 4,885 and records the Guam blocker. All six suggestions are fixed, and all 8 new or changed expectations hand-check. The red ssa-usda check was cancelled at the 1h runner limit (infrastructure), so please re-run it before merging. Four small suggestions. Full report in the comment above.

[MaxGhenis]

US + core hub merge audit, us#10299 at 32111093a46e936e783e6ff4d3f635270ed602df (squash): Follow the SNAP statute on ABAWD timing in Alaska and Hawaii and the FY2026 large-household cap.

• Review: GPT-6.1 Sol shell-backed merge-delta review approved 3211109 after running the SNAP ABAWD, FY2027 COLA and source-conflict tests (reviews-gen/us-10299/review-delta-david2.md); DTrim99 APPROVED; the cap method is logged as us:snap-large-household-cap-rounding in RULINGS.md.
• Gates checked live at merge:
  • gh pr checks exits 0;
  • the latest Pull request run at the head concluded success;
  • MERGEABLE;
  • not a draft;
  • no reviewer's latest review requests changes;
  • the semantic-overlap guard;
  • head pinned.
• Impact: 2025: snap -$2.75M (Alaska and Hawaii ABAWD timing), snap_max_allotment -$53.28M (FY2026 18+ cap); 2026: no change on the default dataset (in the body).
• Axiom: TheAxiomFoundation/rulespec-us#1830 queued (pe-parity; every provision needs encoding)