Skip to content

Apply New Jersey's resource and income limit rules to the aged, blind and disabled Medicaid pathway - #10985

Merged
hua7450 merged 10 commits into
mainfrom
nj-medicaid-abd-limits
Nov 17, 2026
Merged

hua7450 merged 10 commits into
mainfrom
nj-medicaid-abd-limits

Conversation

@hua7450

@hua7450 hua7450 commented Nov 15, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

New Jersey's optional Medicaid pathway for aged, blind and disabled people (42 U.S.C. 1396a(m)) now follows two rules in N.J.A.C. 10:72-4:

  • Resources equal to the $4,000 individual or $6,000 couple limit qualify. Before this change, the model required resources below the limit.
  • The monthly income limit is 1/12 of the poverty guideline rounded down to the whole dollar. In 2026 the couple limit falls from $1,803.33 to $1,803 a month ($21,640 to $21,636 a year). In 2025 the individual limit falls from $1,304.17 to $1,304 a month ($15,650 to $15,648 a year); New Jersey's 2025 chart prints $1,305 (see Source conflict).

Partly addresses #10466 (NJ). The other states' resource operators remain open in that issue.

Legal basis

Rule Text Model
Resources at the limit qualify N.J.A.C. 10:72-4.5(b): "Eligibility for benefits does not exist in any month in which the countable resources of an aged, blind or disabled person exceeds the limits of $ 4,000 for an individual and $ 6,000 for a couple." 42 U.S.C. 1396a(m)(1)(C) also uses "do not exceed". NJ true in individual_limit_inclusive.yaml and couple_limit_inclusive.yaml (was false from 2015)
Monthly income limit rounds down N.J.A.C. 10:72-4.1(a): "The monthly income standard will be 1/12 of the annual poverty income guideline rounded down to the next whole dollar amount for household unit sizes of one and two for aged, blind, and disabled individuals." New rounds_monthly_limit_down.yaml, NJ true, every other state false
Income at the limit qualifies N.J.A.C. 10:72-4.1(d): income "must be equal to or less than the income limit". Unchanged: NJ false in requires_income_below_limit.yaml; this PR adds 4.1(d) as its reference
The level is capped at the poverty line 42 U.S.C. 1396a(m)(2)(A): the income level "may not exceed a percentage (not more than 100 percent) of the official poverty line". The rounded-down amount stays under the cap

Sources:

Source conflict

The Division of Medical Assistance and Health Services (DMAHS) publishes a chart of income standards each January. Its "New Jersey Care… Special Medicaid Programs – A, 100% FPL" column prints 1/12 of the guideline rounded up. The regulation rounds down, so the chart exceeds it by $1 in every year where the division leaves cents:

Year Unit Guideline ÷ 12 N.J.A.C. 10:72-4.1(a) (this PR) DMAHS chart, page 3
2018 1 / 2 1,011.67 / 1,371.67 1,011 / 1,371 1,012 / 1,372 (Medicaid Communication 18-01)
2019 1 / 2 1,040.83 / 1,409.17 1,040 / 1,409 1,041 / 1,410 (19-03)
2020 1 / 2 1,063.33 / 1,436.67 1,063 / 1,436 1,064 / 1,437 (20-02)
2021 1 / 2 1,073.33 / 1,451.67 1,073 / 1,451 1,074 / 1,452 (21-02)
2022 1 / 2 1,132.50 / 1,525.83 1,132 / 1,525 1,133 / 1,526 (22-02)
2023 1 / 2 1,215.00 / 1,643.33 1,215 / 1,643 1,215 / 1,644 (23-02)
2024 1 / 2 1,255.00 / 1,703.33 1,255 / 1,703 1,255 / 1,704 (24-02)
2025 1 / 2 1,304.17 / 1,762.50 1,304 / 1,762 1,305 / 1,763 (25-03)
2026 1 / 2 1,330.00 / 1,803.33 1,330 / 1,803 1,330 / 1,804 (26-03)

Every chart amount that differs from the regulation exceeds 100% of that year's guideline, the cap in 42 U.S.C. 1396a(m)(2)(A). The statute and the regulation agree in every year, so the statute-first rule decides the conflict: the model follows the regulation.

This PR records the conflict as us-nj:medicaid-abd-income-limit-rounding in source_conflicts/candidates.yaml (pe_us.follows: statute, state: ruled, years 2018-2026, both unit sizes). The Axiom corpus holds the 2026 chart; this PR captures page 3 of the 2018-2025 charts under source_conflicts/sources/us-nj/. The model applies New Jersey's 2018 values back to 2015; the 2015-2017 charts were not checked. Following CONTRIBUTING.md (Source conflicts, step 1), it adds no # Source conflict marker: the guard rejects a marker whose id is not yet in the pinned register. After merge, the rulespec-us entry pins this PR's merge commit, and a follow-up PR adds the marker to rounds_monthly_limit_down.yaml and refreshes the pin.

Changes

  • parameters/gov/hhs/medicaid/eligibility/categories/senior_or_disabled/income/limit/rounds_monthly_limit_down.yaml (new): per-state flag, NJ true from 2018-01-01, matching the start of the NJ income-limit share.
  • variables/gov/hhs/medicaid/eligibility/categories/medicaid_optional_senior_or_disabled_income_limit.py: on the poverty-guideline branch, flagged states get floor(share × guideline ÷ 12) × 12. Unflagged states keep share × guideline unchanged. Missouri's round-up branch reads the same monthly amount.
  • parameters/.../senior_or_disabled/assets/limit/individual_limit_inclusive.yaml and couple_limit_inclusive.yaml: NJ true, with N.J.A.C. 10:72-4.5(b) as a reference.
  • parameters/.../senior_or_disabled/income/limit/requires_income_below_limit.yaml: adds N.J.A.C. 10:72-4.1(d) as a reference. Values unchanged.
  • source_conflicts/candidates.yaml: the candidate entry above.
  • source_conflicts/sources/us-nj/dmahs-income-standards-<year>-p3.txt, 2018-2025: text captures of each chart's page 3.

The two other readers of the income limit compare income with <=: is_optional_senior_or_disabled_income_eligible (NJ is not flagged below-only) and is_209b_ssi_recipient_income_eligible_for_medicaid (NJ is not a 209(b) state). Both match 4.1(d).

Tests

New YAML cases, appended to the variable-named files:

File Case Expected Arithmetic
medicaid_optional_senior_or_disabled_income_limit.yaml NJ aged couple, 2026 21,636 each floor(1.00 × 21,640 ÷ 12) = floor(1,803.33) = 1,803; 1,803 × 12 = 21,636
medicaid_optional_senior_or_disabled_income_limit.yaml NJ aged single, 2025 15,648 floor(1.00 × 15,650 ÷ 12) = floor(1,304.17) = 1,304; 1,304 × 12 = 15,648
medicaid_optional_senior_or_disabled_income_limit.yaml NJ aged couple, 2021 17,412 each floor(1.00 × 17,420 ÷ 12) = floor(1,451.67) = 1,451; 1,451 × 12 = 17,412 (nearest-dollar rounding would give 17,424)
is_optional_senior_or_disabled_income_eligible.yaml Threshold: NJ couple at the 2026 limit true, true 10,818 + 10,818 = 21,636 ≤ 21,636
is_optional_senior_or_disabled_income_eligible.yaml Threshold: NJ couple one dollar over false, false 10,818 + 10,819 = 21,637 > 21,636 (under the unrounded 21,640)
is_optional_senior_or_disabled_asset_eligible.yaml Threshold: NJ individual at $4,000 true 4,000 ≤ 4,000
is_optional_senior_or_disabled_asset_eligible.yaml Threshold: NJ couple at $6,000 true, true 3,500 + 2,500 = 6,000 ≤ 6,000
is_optional_senior_or_disabled_asset_eligible.yaml Threshold: NJ individual one cent over false 4,000.01 > 4,000

Existing expectations that change:

File Case Variable Old New Reason
medicaid_optional_senior_or_disabled_uses_couple_standard.yaml Case 12 (NJ, 2026 couple) medicaid_optional_senior_or_disabled_income_limit 21,640 21,636 1,803 × 12; the couple's 29,760 stays above the limit, so eligibility is unchanged
tests/unit/test_medicaid_abd_ssi_rate_standard_properties.py test_unflagged_states_keep_their_share_of_the_poverty_guideline expected limit for NJ share × guideline floor(share × guideline ÷ 12) × 12 NJ stays covered for 2018-2026 under the rounding rule
tests/unit/test_medicaid_abd_ssi_rate_standard_properties.py test_state_standard_states_match_their_rules new assertion none the states flagged in rounds_monthly_limit_down equal ROUNDS_DOWN_STATES (["NJ"]) parameter-only check of the test's hand-written list; no new simulation

No partner contract test has a New Jersey case.

Review fixes

From DTrim99's review:

axiom: TheAxiomFoundation/rulespec-us#2262 queued

🤖 Generated with Claude Code

hua7450 and others added 2 commits November 15, 2026 13:52
… and disabled Medicaid pathway

N.J.A.C. 10:72-4.5(b) denies eligibility only when countable resources
exceed the $4,000 individual or $6,000 couple limit, so resources equal to
a limit now qualify in New Jersey.

N.J.A.C. 10:72-4.1(a) sets the monthly income standard at 1/12 of the
annual poverty guideline rounded down to the whole dollar. A new per-state
flag, rounds_monthly_limit_down (NJ true), applies the floor on the
poverty-guideline branch of medicaid_optional_senior_or_disabled_income_limit.
The 2026 couple limit becomes $1,803 a month ($21,636 a year).

Medicaid Communication 26-03 prints $1,804, above the 100 percent cap of
42 U.S.C. 1396a(m)(2)(A); the conflict is recorded as candidate
us-nj:medicaid-abd-income-limit-rounding.

Partly addresses #10466 (NJ).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… address

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hua7450 and others added 3 commits November 15, 2026 13:59
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Other states' false values are unverified defaults, and Missouri's round-up lives in the formula.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DTrim99

DTrim99 commented Nov 16, 2026

Copy link
Copy Markdown
Collaborator

PR Review: Apply New Jersey's resource and income limit rules to the aged, blind and disabled Medicaid pathway

Summary

The PR makes two changes for New Jersey's optional aged, blind and disabled group (42 U.S.C. 1396a(m)), and the law supports both:

  • Resources equal to the $4,000 / $6,000 limits now qualify.
  • The monthly income standard is 1/12 of the poverty guideline rounded down to the whole dollar.

The code is correct and vectorized, other states' results don't change, and every change is pinned by a test.

One SHOULD remains. The source-conflict record covers only the 2026 couple amount. New Jersey's own charts for 2023, 2024 and 2025 also round up and contradict values this PR changes, including its new 2025 single-person test.

Critical

None.

Should address

S1. The source-conflict candidate covers 2026 couples only, but the chart-versus-regulation conflict runs through every year checked (source_conflicts/candidates.yaml:8772-8814; PR body "Source conflict"; rulespec-us#2262 "Years").

The entry sets years: from 2026, to 2026 and asks only about "an aged, blind or disabled couple". Its note ends "Charts for other years were not checked." I checked three earlier DMAHS charts. In the "New Jersey Care... Special Medicaid Programs - A, 100% FPL" column, each prints 1/12 of the guideline rounded up. The regulation's floor, which this PR applies from 2018, differs in every year where the division leaves cents:

Year Unit Guideline ÷ 12 N.J.A.C. 10:72-4.1(a) floor = model at this head DMAHS chart
2023 1 14,580 ÷ 12 = 1,215.00 1,215 1,215 (agree)
2023 2 19,720 ÷ 12 = 1,643.33 1,643 1,644: MC 23-02, p. 3
2024 1 15,060 ÷ 12 = 1,255.00 1,255 1,255 (agree)
2024 2 20,440 ÷ 12 = 1,703.33 1,703 1,704: MC 24-02, p. 3
2025 1 15,650 ÷ 12 = 1,304.17 1,304 1,305: MC 25-03, p. 3
2025 2 21,150 ÷ 12 = 1,762.50 1,762 1,763: MC 25-03, p. 3
2026 1 15,960 ÷ 12 = 1,330.00 1,330 1,330 (agree)
2026 2 21,640 ÷ 12 = 1,803.33 1,803 1,804: MC 26-03, p. 3 (already recorded)

This leaves the record out of step with the code and tests in four places:

  • The 2025 test. The new case "New Jersey aged single person gets the 2025 individual standard rounded down to the whole dollar" expects 15,648, which is 1,304 × 12. The note lists it as a pinning case, but the agency's 2025 chart gives $1,305 (15,660 a year), and the entry's years and question exclude 2025 and single people.
  • The PR body. It says "In 2025 the individual limit falls from $1,304.17 to $1,304" without noting that the 2025 chart prints $1,305.
  • The Axiom issue. rulespec-us#2262 says "Only the 2026 chart was checked for the conflict". Its companion case 2 (2025, $1,304) records no alternative.
  • The register. Step 2 of CONTRIBUTING.md#source-conflicts needs the years. The register would show a 2026-only conflict while PE-US departs from the agency's published figure in 2023-2025 as well.

The ruling should carry over unchanged. Each chart amount is above 100% of that year's guideline, so 1396a(m)(2)(A) and the regulation point the same way every year.

Fix:

  • Widen years to from: 2018 (when the model's rounding starts) through 2026. If the 2018-2022 charts stay unchecked, say so in the note rather than starting at 2026. Use from: 2023 at the least, which the charts above cover.
  • Generalize the question to both unit sizes. For example: "Is the monthly standard 1/12 of the guideline rounded down, as N.J.A.C. 10:72-4.1(a) says, or rounded up, as each year's DMAHS income-standards chart prints?"
  • Add the 2023-2025 charts as guidance sources, with captures under source_conflicts/sources/ if the Axiom corpus lacks them.
  • Replace "Charts for other years were not checked."
  • Confirm that the 2026-11-15 ruling covers every year, and update the PR body and rulespec-us#2262 to match (its Years paragraph, and $1,305 as the alternative to companion case 2).

Suggestions

  1. Add a YAML case whose monthly fraction is above one half. Both new limit cases leave a fraction below .5 (1,803.33 and 1,304.17), so a nearest-dollar mutant (np.round in place of np.floor) passes every YAML case. Only the Python property test catches it: 2018 single is 12,140 ÷ 12 = 1,011.67, which floors to 1,011 but rounds to 1,012. A case such as NJ couple 2021 would close this: 17,420 ÷ 12 = 1,451.67, floor 1,451 × 12 = 17,412, while nearest gives 17,424. It would also give rulespec-us#2262 a companion row that separates floor from nearest-dollar rounding. Its case 5 separates floor only from the chart's round-up.
  2. Cross-check the property test's state list against the parameter. ROUNDS_DOWN_STATES (test_medicaid_abd_ssi_rate_standard_properties.py:118) is the only hand-written state list in that file that no parameter test checks. The docstring says the parameters are checked against those lists. A one-line assertion would do it, e.g. states_where(limit_parameters(year).rounds_monthly_limit_down) == ROUNDS_DOWN_STATES, added to a parameter-only test such as test_state_standard_states_match_their_rules (line 174). It needs no new simulation.
  3. Refresh the Axiom issue's head reference. rulespec-us#2262 names 32bc3f9102. The behavior hasn't changed since, but citing the current head keeps the case names checkable.

Verified correct

The law matches the code, by section:

  • N.J.A.C. 10:72-4.1(a): the monthly standard is "1/12 of the annual poverty income guideline rounded down to the next whole dollar amount for household unit sizes of one and two for aged, blind, and disabled individuals". The revision takes effect "with the first day of the year for which the poverty income guideline is promulgated", which matches the model's January 1 guideline dates.
  • 4.1(d) and (d)2: income "equal to or less than the income limit" qualifies, so NJ's false in requires_income_below_limit.yaml is right.
  • 4.5(b): eligibility fails when countable resources "exceeds the limits of $ 4,000 for an individual and $ 6,000 for a couple". 4.5(b)1 sends counting to the Medicaid Only rules, and N.J.A.C. 10:71-4.5(b)-(c) also disqualify only when resources "exceed" the limit.
  • 42 U.S.C. 1396a(m)(2)(A): the level is "not more than 100 percent" of the poverty line.
  • Years: 4.1 was last amended by R.2002 d.317 and 4.5 by R.2008 d.30, so both rules cover the full claimed span (inclusive resources from 2015, rounding from 2018).

MC 26-03, page 3: the NJ Care column prints $1,330 / $1,804, with resources of $4,000 / $6,000 and the footnote "Variances due to rounding may occur". This matches the candidate's citation and anchor.

Parameters:

  • Only one parameter is new (rounds_monthly_limit_down.yaml). The other three are existing files: NJ goes from false to true in individual_limit_inclusive.yaml and couple_limit_inclusive.yaml, and requires_income_below_limit.yaml gains a reference only.
  • All four break down by state_code with the same 51 keys as their siblings. The formula indexes them by the household's state_code_str, not by a state-gated int or enum variable, so the NH Add New Hampshire state supplementation for 2026 #10156 missing-key pattern doesn't arise.
  • Descriptions follow the folder's "..., if this is true." template.

Formula (medicaid_optional_senior_or_disabled_income_limit.py:62-70, 128):

  • For unflagged states, where returns the old limit_pct * unit_fpg.
  • Missouri's np.ceil branch reads the same monthly amount and is selected first.
  • NJ is not flagged as using the SSI rate or a state standard, and it covers non-SSI recipients, so it reaches the default branch.
  • where and np.floor work element by element. The property test builds one multi-state simulation per year and unit size, with NJ beside every unflagged state, for 2018-2026 at rtol 1e-6, and it passes in CI. Other states are unchanged.
  • NJ's share is 1.00 and the guidelines are whole dollars, so whole-dollar months (2023, 2024 and 2026 singles) divide exactly and can't floor down a dollar.

Callers: the income limit has two readers.

  • is_optional_senior_or_disabled_income_eligible uses <= unless a state is flagged below-only, and NJ isn't.
  • is_209b_ssi_recipient_income_eligible_for_medicaid also uses <=, and NJ is false in section_209b.yaml.

Mutation: every change is pinned.

  • Setting NJ's rounding flag to false fails both new limit cases, the one-dollar-over case and the property test.
  • Ceiling in place of floor fails the YAML cases; nearest-dollar rounding fails the property test (Suggestion 1).
  • Setting either inclusive flag back to false fails its at-limit case.

Hand checks (10):

  1. NJ couple 2026 limit: (15,960 + 5,680) × 1.00 = 21,640; ÷ 12 = 1,803.33; floor 1,803; × 12 = 21,636.
  2. NJ single 2025 limit: 15,650 ÷ 12 = 1,304.17; floor 1,304; × 12 = 15,648.
  3. Couple at the limit: 10,818 + 10,818 = 21,636 ≤ 21,636, so true, true.
  4. Couple one dollar over: 10,818 + 10,819 = 21,637 > 21,636, so false, false. The old 21,640 limit would have passed it, so the case pins the rounding. It also pins the couple standard: alone, each spouse is under 15,960.
  5. Individual resources 4,000 ≤ 4,000 with the inclusive flag: true. Under the old < this was false.
  6. Couple resources 3,500 + 2,500 = 6,000 ≤ 6,000: true, true.
  7. Individual resources of 4,000.01: float32 4,000.0100 rounds to 4,000.01, which is > 4,000, so false.
  8. Case 12 in ..._uses_couple_standard.yaml: (8,000 + 22,000 − 20 × 12) = 29,760, or 14,880 each. The limit is 1,803 × 12 = 21,636, and 29,760 > 21,636, so the case stays ineligible.
  9. Property test, NJ couple 2019: 12,490 + 4,420 = 16,910; ÷ 12 = 1,409.17; floor × 12 = 16,908.
  10. Property test, NJ single 2022: 13,590 ÷ 12 = 1,132.50; floor × 12 = 13,584.

Python test (CLAUDE.md "Test design and CI cost"): the PR changes one expectation in an existing parametrized test (18 cases). It adds no simulations, years or parameter-tree copies, so CI cost is unchanged. The change is needed because the existing independent reference would otherwise fail for NJ.

candidates.yaml entry:

  • It has every required field (id, program, question, years, three sources of valid kinds, readings, pe_us with follows: statute, tracked code paths and a valid PR URL, and state: ruled).
  • The example's arithmetic holds: 1,803.50 × 12 = 21,642.
  • It adds no marker, which is correct for step 1.
  • None of the new comments in the variable or parameter files reads as a conflict, so the ratchet counts don't change.

Housekeeping:

  • The changelog is the top-level changelog.d/nj-medicaid-abd-limits.fixed.md.
  • The tests sit in variable-named files.
  • References are one per source (Cornell section URLs).
  • No partner test file is touched.

Axiom line assessment

axiom: TheAxiomFoundation/rulespec-us#2262 queued is valid in form.

  • Status: OPEN, labelled pe-parity, created 2026-11-15.
  • Dispatch-ready: it has target module paths (us-nj/regulations/njac-10-72/10-72-4/1.yaml and .../5.yaml, beside the existing us/statutes/42/1396a/m.yaml) and corpus citations with their status. It quotes the law verbatim (4.1(a), (d), (d)2, 4.5(b) and 1396a(m)(2)(A)) and has a required-outputs table. Its eight companion tests come from the regulation and the HHS guidelines, not from PE output. It records its blocker: section-level corpus ingest of 10:72-4.1 and 4.5, a release pin, and signed-encoder approval.
  • Matches this head: it names 32bc3f9102. The later commits only add the PR link to the candidate, merge main and add a comment to the parameter file, so the eight cases and the behavior are the same.
  • Gap: the same as S1. Its Years paragraph covers only the 2026 chart, and companion case 2 needs the 2025 chart's $1,305 as the alternative.

CI status

gh pr checks 10985 at head 5aca3b7e6c, re-checked just before writing: 33 pass, 5 pending, 0 failing.

  • Passing: every Baseline shard, every Python Rest shard (including the property test), Microsimulation, PR validation (guard tests, lint, changelog) and Household API Partners.
  • Pending: Contrib states-shard-1/2/4 and other-shard-1/3. These run contributed reforms, which this PR doesn't touch.

Branch status

Verdict: REQUEST_CHANGES

@DTrim99 DTrim99 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Requesting changes: 0 critical, 1 should address (the rounding candidate us-nj:medicaid-abd-income-limit-rounding covers 2026 couples only, but the DMAHS charts round up in 2023-2025 too, including the 2025 single standard this PR tests at $1,304 vs the chart's $1,305), 3 suggestions. Full report in the comment above. The code matches N.J.A.C. 10:72-4.1 and 4.5.

hua7450 and others added 4 commits November 16, 2026 15:11
DTrim99's review (S1): the DMAHS income-standards charts round the
monthly standard up in every year checked, not only for 2026 couples.
All nine charts from 2018 to 2026 print 1/12 of the guideline rounded
up, $1 above N.J.A.C. 10:72-4.1(a) whenever the division leaves cents
(couples every year; one person in 2018-2022 and 2025, including the
2025 chart's $1,305 against the regulation's $1,304).

- years: 2018-2026; the question now covers both unit sizes.
- Cite Medicaid Communications 18-01, 19-03, 20-02, 21-02, 22-02, 23-02,
  24-02 and 25-03 as guidance sources beside 26-03, and capture page 3
  of the 2018-2025 charts under source_conflicts/sources/us-nj/ (the
  Axiom corpus holds only the 2026 chart).
- The note now says the statute-first rule decides the conflict:
  42 U.S.C. 1396a(m)(2)(A) and the regulation agree in every year, and
  each differing chart amount exceeds the 100 percent cap. It names the
  unchecked 2015-2017 charts and the new 2021 pinning case.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s one half

DTrim99's review (suggestion 1): both earlier New Jersey limit cases
leave a fraction below .5, so rounding to the nearest dollar would pass
them. 17,420 / 12 = 1,451.67 floors to 1,451 (17,412 a year), while the
nearest dollar gives 1,452 (17,424).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DTrim99's review (suggestion 2): assert in the parameter-only
test_state_standard_states_match_their_rules that the states flagged in
rounds_monthly_limit_down equal ROUNDS_DOWN_STATES. No new simulation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DTrim99

DTrim99 commented Nov 17, 2026

Copy link
Copy Markdown
Collaborator

PR Review (round 2): Apply New Jersey's resource and income limit rules to the aged, blind and disabled Medicaid pathway

Summary

The fix commits close the open SHOULD and take all three suggestions:

  • The source-conflict candidate now covers 2018-2026 and both unit sizes.
  • All nine DMAHS charts are cited, and page 3 of the 2018-2025 charts is captured.
  • A 2021 couple case pins rounding down against rounding to the nearest dollar.
  • The property test checks its rounding list against the parameter.
  • rulespec-us#2262 matches this head.

I opened all nine chart URLs and confirmed every quoted amount on page 3. The eight captures match those pages number for number.

The model code and parameters haven't changed since round 1. CI is green at the head (37/37).

There are no new critical or should findings, and two small suggestions about the candidate's wording. The branch now conflicts with main in one test file, because #10146 changed the same NJ case. The fix is small and is described under Branch.

Prior findings

Finding Status Evidence
S1. The source-conflict candidate covered only 2026 couples Fixed years is now 2018 to 2026, the year the model's NJ rounding flag and income share start. The question covers both unit sizes ("1/12 of the poverty guideline rounded down ... or rounded up, as each year's DMAHS income-standards chart prints"). Medicaid Communications 18-01, 19-03, 20-02, 21-02, 22-02, 23-02, 24-02, 25-03 and 26-03 are guidance sources. readings.other lists every differing chart amount. The note replaces "Charts for other years were not checked" with "All nine charts from 2018 to 2026 were checked" and states the remaining gap: the 2015-2017 charts, for years the model fills from its 2018 values. The PR body's Source conflict table and rulespec-us#2262 now carry the same years and amounts. See Verified for the chart and capture checks.
Suggestion 1: add a YAML case whose monthly fraction is above one half Taken New case "New Jersey aged couple gets the 2021 couple standard rounded down, not to the nearest dollar" expects 17,412. Rounding to the nearest dollar, rounding up or not rounding gives 17,424, 17,424 or 17,420, and each fails at the 0.01 margin.
Suggestion 2: check ROUNDS_DOWN_STATES against the parameter Taken test_state_standard_states_match_their_rules now asserts states_where(limit_parameters(year).rounds_monthly_limit_down) == ROUNDS_DOWN_STATES for 2018-2026. It is a parameter-only check on the shared module-level system, with no new simulation or model build, so it fits CLAUDE.md "Test design and CI cost". The assertion fails if a state is added to either the parameter or the list alone.
Suggestion 3: refresh the head reference in rulespec-us#2262 Taken The issue (updated 2026-11-16 20:17 UTC) cites 191ad3842b for its companion cases and its capture link.

New findings

Critical: none.

Should address: none.

Suggestions:

  1. Name the basis of the ruling with its date. The round-1 note said "Ruled by the project lead, 2026-11-15". The new note drops that line and says "The statute-first rule decides the conflict". That basis is sound and holds in every year. The statute (1396a(m)(2)(A)) and the regulation (4.1(a)) agree each year, and the chart is guidance, so CONTRIBUTING.md's statute-first and "regulations are law" principles settle it without a case-by-case ruling. Other ruled entries record the rule with its date, for example "Ruled under the 2026-10-19 statute-first rule", or link the project lead's ruling. Writing "under the 2026-10-20 statute-first rule", or keeping the 2026-11-15 ruling with a link, would keep the record traceable.
  2. 19-03 citation date. The candidate cites 19-03 as "(February 4, 2019)". The PDF at the cited URL is the reissue: page 1 reads "DATE: February 4, 2019 / REISSUED: March 5, 2019", and the chart is marked "Rev 3/2019". "(February 4, 2019; reissued March 5, 2019)" would match the captured text. The amounts are the same either way.

Verified

Chart sources. Each of the nine URLs returns a PDF (HTTP 200). On page 3, the "New Jersey Care… Special Medicaid Programs - A, 100% FPL**" monthly column prints:

Year MC One person Two people Guideline ÷ 12 (1 / 2) N.J.A.C. 10:72-4.1(a) = model
2018 18-01 1,012 1,372 12,140 → 1,011.67 / 16,460 → 1,371.67 1,011 / 1,371
2019 19-03 1,041 1,410 12,490 → 1,040.83 / 16,910 → 1,409.17 1,040 / 1,409
2020 20-02 1,064 1,437 12,760 → 1,063.33 / 17,240 → 1,436.67 1,063 / 1,436
2021 21-02 1,074 1,452 12,880 → 1,073.33 / 17,420 → 1,451.67 1,073 / 1,451
2022 22-02 1,133 1,526 13,590 → 1,132.50 / 18,310 → 1,525.83 1,132 / 1,525
2023 23-02 1,215 1,644 14,580 → 1,215.00 / 19,720 → 1,643.33 1,215 / 1,643
2024 24-02 1,255 1,704 15,060 → 1,255.00 / 20,440 → 1,703.33 1,255 / 1,703
2025 25-03 1,305 1,763 15,650 → 1,304.17 / 21,150 → 1,762.50 1,304 / 1,762
2026 26-03 1,330 1,804 15,960 → 1,330.00 / 21,640 → 1,803.33 1,330 / 1,803
  • The chart rounds up in every year. Every amount that differs is above 100% of that year's guideline, so the statute's cap and the regulation agree against the chart each year.
  • The guidelines match the model. The model's gov.hhs.fpg contiguous-US values give the same totals: first person plus additional person, e.g. 12,880 + 4,540 = 17,420 for 2021.
  • The citations match page 1. Each PDF's communication number and date agree with the candidate's citation, except for the 19-03 reissue noted in Suggestion 2.
  • Resources. Each chart shows NJ Care resources of $4,000 / $6,000, which matches the inclusive flags.

Captures (source_conflicts/sources/us-nj/dmahs-income-standards-<year>-p3.txt, 2018-2025):

  • The right pages. Each file is page 3 of the cited communication. Its first line is the Source: URL, then a citation and an extraction line, as the capture guide asks.
  • The text matches. Comparing every number token against a fresh pdftotext -layout of page 3 from the live PDFs finds no differences in any year.
  • Clean files. Each is 10-11 KB, LF-only, with no trailing spaces. The only non-ASCII character is the chart's own ellipsis, and the files hold no bot-check content.
  • Coverage. The 2026 chart is in the Axiom corpus (manifest us-nj-msp-income-standards-2026.yaml, cited in Kentucky Personal Income Tax Schedule #2262), so no capture is needed for it.

Candidate entry:

  • The example holds. $1,304.50 × 12 = $15,654, which is over $1,304 and the $1,304.17 cap but within $1,305.
  • The readings list matches the charts. "For two people ... $1,372, $1,410, $1,437, $1,452, $1,526, $1,644, $1,704, $1,763 and $1,804" and "for one person ... $1,012, $1,041, $1,064, $1,074 and $1,133 (2018 to 2022) and $1,305 (2025)" both match the table above. The one-person 2023, 2024 and 2026 amounts divide exactly and agree.
  • Structure. The required fields are present, the pe_us.code paths are tracked, the PR URL is valid and state: ruled. It adds no marker, which is correct for step 1.

Hand checks of new and changed expectations:

  1. 2021 NJ couple limit: (12,880 + 4,540) × 1.00 = 17,420; ÷ 12 = 1,451.67; floor 1,451; × 12 = 17,412 for each spouse. Both spouses are aged, so the couple standard applies.
  2. The same case under nearest-dollar rounding: 1,452 × 12 = 17,424, so the case rules out np.round.
  3. New assertion: rounds_monthly_limit_down has 51 state keys, all 2018-01-01: false except NJ true. That gives ["NJ"] for every year in YEARS (2018-2026), which equals ROUNDS_DOWN_STATES.
  4. PR-body table, 2022 single: 13,590 ÷ 12 = 1,132.50, floor 1,132 against the chart's 1,133.
  5. PR-body table, 2020 couple: 17,240 ÷ 12 = 1,436.67, floor 1,436 against the chart's 1,437.
  6. Kentucky Personal Income Tax Schedule #2262 case 9: 17,420 ÷ 12 = 1,451.67, so $1,451 ($17,412), with a chart alternative of $1,452 (21-02).
  7. Kentucky Personal Income Tax Schedule #2262 case 2: chart alternative $1,305 (25-03), confirmed on page 3.
  8. Kentucky Personal Income Tax Schedule #2262 case 3 points to ..._uses_couple_standard.yaml Case 7 (NJ, 2026, each spouse tested alone), which expects 15,960 = 1,330 × 12.

What else changed since round 1:

  • The three fix commits.
    • One YAML case, plus a comment adding the 2021 guideline.
    • One Python assertion.
    • The candidates.yaml entry and eight captures.
  • The main merge. One merge of main (03c74703b1) was clean, with no hand-resolved hunks.
  • No other changes. No variable, parameter or changelog file in the PR changed since 5aca3b7e6c.
  • No ratchet comments. None of the new comments reads as a source conflict ("rounded down, not to the nearest dollar"; "The rounding list used by the poverty-guideline test below"), so the Tie source-conflict markers to the register shared with Axiom #10287 counts don't move.
  • Partner tests. No partner test file is touched.

Axiom

axiom: TheAxiomFoundation/rulespec-us#2262 queued remains valid.

  • Status: OPEN, labelled pe-parity.
  • Dispatch-ready: it still has module paths, corpus citations with their status, verbatim 4.1(a), (d), (d)2, 4.5(b) and 1396a(m)(2)(A), a required-outputs table and its blockers.
  • Matches this head:
    • The Years paragraph says "All nine charts from 2018 to 2026 were checked ...; the 2015 to 2017 charts were not".
    • The Source conflict reading lists both unit sizes' chart amounts.
    • A new "Chart alternative" column gives $1,804, $1,305, $1,330 (agrees) and $1,452.
    • Companion case 9 is the 2021 couple, and the capture link points to 191ad3842b.

CI

gh pr checks 10985 at 191ad3842b, re-checked just before writing: 37 pass, 0 pending, 0 failing. This covers every Baseline and Contrib shard, all ten Python Rest shards (including the property test), Microsimulation, PR validation (lock, guard tests, lint, changelog), Package and compatibility, and Household API Partners.

Branch

Verdict: APPROVE

@DTrim99 DTrim99 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving (round 2): the prior should-address item is fixed (the rounding conflict now covers 2018-2026, both unit sizes, with all nine DMAHS charts cited and checked), and all three suggestions were taken. 2 optional suggestions. CI 37/37 green at 191ad38. Branch note: the branch now conflicts with main in medicaid_optional_senior_or_disabled_uses_couple_standard.yaml Case 12 (#10146); keep main's version and set the limit to [21_636, 21_636]. Full report in the comment above.

# Conflicts:
#	policyengine_us/tests/policy/baseline/gov/hhs/medicaid/eligibility/categories/medicaid_optional_senior_or_disabled_uses_couple_standard.yaml
#	source_conflicts/candidates.yaml
@hua7450
hua7450 merged commit 31721b3 into main Nov 17, 2026
38 checks passed
@hua7450
hua7450 deleted the nj-medicaid-abd-limits branch November 17, 2026 17:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants