[hua7450]

Closes #10369

What and why

The Illinois Health Benefits for Workers with Disabilities (HBWD) premium chart in 89 Ill. Admin. Code 120.510(n) uses closed bands: 0-250, 251-500, … 4751-5000, then 5000+. The HFS premium chart prints the same bands. premium/earned.yaml and premium/unearned.yaml put the thresholds at 250, 500, … 5,000, and the default lower-inclusive lookup then charged income of exactly $250, $500, … $5,000 the next band's premium.

This PR moves each threshold to its band start (251, 501, … 4,751) and keeps the default lookup, following the model-development guidance for "251-500"-style tables. Because $5,000 sits in the 4751-5000 band, the open top band starts at $5,001. The premium amounts were already correct and do not change.

Monthly income	Chart	Before	After
Earned $250	$0	$6	$0
Earned $1,000	$19	$25	$19
Unearned $1,000	$56	$75	$56
Unearned $1,500	$94	$113	$94
Earned / unearned $5,000	$119 / $356	$125 / $375	$119 / $356

Only income in the first dollar above a multiple of $250 (for example $250.00 to $250.99) changes band. The regulation and the chart say nothing about cents. IDHS computes medical eligibility "in whole dollar amounts only" and drops cents (PM 15-04-03). Because every band starts at a whole dollar, income with cents already lands in the same band as the whole-dollar amount, so the PR adds no rounding step. No HBWD text applies PM 15-04-03 to the premium; the parameter comment names it only as the source of the whole-dollar reading.

Tests

Existing cases that pinned the old edges, and their new values:

File	Case	Before	After
il_hbwd_premium.yaml	Case 1, earned $2,000	$50	$44
	Case 2, earned $1,000, unearned $1,200	$100	$94
	Case 3, earned $3,000, unearned $2,000	$225	$200
	Case 6, $250 of each	$25	$0
il_hbwd_person.yaml	Case 1, earned $1,000, unearned $1,200	−$100	−$94
integration.yaml	Case 1, SSDI of $1,000 a month	$113	$94
	Case 6, SSDI of $1,500 a month	$132	$113

• il_hbwd_premium.yaml Case 4 now uses $5,001 of each income so that it still tests the top band; $5,000 falls in the 4751-5000 band. Its name no longer says "capped", because the chart sets no cap: $500 is its highest cell.
• New Case 8 pins that reading: $5,000 of each income pays $119 + $356 = $475.
• New Cases 9-12 put each income at every band start (251, 501, ... 5,001) and one dollar below it (250, 500, ... 5,000); expected values are the chart's first column and first row, so no case asserts one of the seven cells where a source differs from the sum. Case 13 pins cents ($250.99 stays in 0-250, $5,000.99 in 4751-5000). Case 14 is HFS 3792 Example 1 ($900 from work, $1,000 from Social Security): $75, which this PR fixes ($94 before).
• Comments in integration.yaml now show the exact countable income the assertions use (Cases 1, 4, 5 and 6), and the remaining "bracket" wording is now "band".
• Partner contract tests are unaffected. Their HBWD cases have about $677 and $2,158 of monthly countable earnings, inside a band.

The premium table

• Summing the first column (earned) and first row (unearned) reproduces 434 of the 441 cells in 120.510(n)(3). Six cells are $1 off the sum in both the regulation and the chart (for example, $600 earned and $300 unearned is $31 in the table and $32 in the model). That predates this PR; Illinois HBWD premium: encode the 21 x 21 premium table instead of adding earned and unearned components #10471 proposes encoding the full table.
• The seventh cell, earned 1251-1500 with unearned 501-750, is 79 in the regulation and 69 in the HFS chart; PolicyEngine gives 69. This PR adds it to source_conflicts/candidates.yaml as us-il:hbwd-premium-earned-1251-unearned-501-cell and does not change the cell. The record covers both texts from 2009 (the regulation effective February 1, 2009 already prints 79; the chart dates from June 2009) and adds the HFS 3792 brochure, which also prints 69.

Source conflict ruled: regulation (79); PE-US keeps 69 until #10471. The project lead for these Medicaid fixes ruled for the regulation's 79 on 2026-10-27 (ruling): under "Regulations are law" the regulation binds, and the written canons have no exception for an apparent typesetting error. PE-US gives 69 because its premium adds the earned and unearned components (31 + 38); #10471, which encodes the full table, will bring the cell to 79. The candidate is now state: ruled, follows: conflicting-source.

Test plan

• Thresholds checked row by row against the regulation chart and the HFS chart
• Changelog fragment in changelog.d/
• Source-conflict candidate added for the one cell where the regulation and the chart disagree
• Local review: band edges, chart rows and every changed expectation checked by hand
• CI passes (tests were not run locally)

axiom: TheAxiomFoundation/rulespec-us#1926 queued

🤖 Generated with Claude Code

[PavelMakarchuk]

Program Review

PR #10372, "Key Illinois HBWD premium bands at the chart's band starts" (hua7450). Round 1, full review.

IDs: C = critical, SA = should address, G = suggestion.

Paths: E = policyengine_us/parameters/gov/states/il/hfs/hbwd/premium/earned.yaml, U = .../hbwd/premium/unearned.yaml, T = policyengine_us/tests/policy/baseline/gov/states/il/hfs/hbwd/il_hbwd_premium.yaml, I = .../hbwd/integration.yaml (same test folder), R = source_conflicts/candidates.yaml. Line numbers refer to the reviewed head. PDF links use the file page number.

Source Documents

• Reviewed head SHA: bd587e4
• Mode: full
• Scope: 7 files, 136 insertions and 81 deletions against the merge base 9ee14cdb43. No Python changes.
• Regulation: 89 Ill. Adm. Code 120.510(n): Cornell LII (the parameter reference), the ILGA JCAR text (effective 2022-03-25) and its 2020 Wayback capture (effective 2009-02-01). All 441 cells of the (n)(3) table were parsed from each copy. The 2009 and 2022 tables are identical.
• PDF: HFS HBWD Premium Chart (one page, created June 2009), read as an image at 150 and 600 dpi. It is still the chart linked from the HFS premiums page.
• Web: HFS 3792 brochure (R-03-13, premium table and two worked examples); IDHS PM 15-04-03 (whole dollars).
• Repository and process: CONTRIBUTING.md:101-103; the current model-development guidance for band-start tables (thresholds 0, 3_001, 5_001 with the default .calc(x)); SingleAmountTaxScale.calc in policyengine-core (lower bound inclusive).
• Year: the table has one version from 2009 to today. The tests cover 2024-01, 2025-01 and 2026-01.

Branch Status

• 4 commits ahead and 19 behind main, MERGEABLE. No commit on main since the merge base touches HBWD or source_conflicts/.
• No other open PR touches il/hfs/hbwd or medicaid_working_disabled_buy_in*.
• 19 other open PRs edit R (for example #10240, #10364, #10365). None adds a us-il: id, so the risk is a textual merge conflict only.
• Issue #10471 (encode the full 21 x 21 table) builds on these band starts and on the 79-versus-69 ruling. Merge this PR first.

What the PR gets right

• Thresholds. All 40 new thresholds are the printed band starts, and all 42 amounts match the regulation and the chart. With the default lower-inclusive .calc(x), $250 is in 0-250, $251 in 251-500, $5,000 in 4751-5000 and $5,001 in 5000+. This follows the current guidance for "251-500"-style tables.
• Expected values. Every changed value equals the chart cell, not only the additive sum.
• PR body claims. The before/after table reproduces at the head and the merge base. The 434-of-441 claim is right (435 of 441 against the chart).
• The fix matches the agency. HFS 3792 Example 1 ($900 from work, $1,000 from Social Security) gives $75. The head gives $75. The merge base gave $94.
• Comments. The integration Case 1 and Case 6 breakdowns are exact. The $221.87 of withheld taxes is IL withholding $97.87 plus Social Security tax $124.00.
• Register entry. R:2167-2201 passes the register validator, and its keys match about 110 other candidates that have no statute.
• Housekeeping. The changelog fragment is correct. No partner contract test is edited, and the partner HBWD cases pass.

Critical (Must Fix)

None.

Should Address

SA1. No test sits on a band start (except $5,001), one dollar below most starts, or on an edge with cents.

• Where: T (append after T:134). The values under test are E:16-99 and U:16-99.
• Evidence: The PR's whole change is 40 thresholds. Across every test that reaches il_hbwd_premium (85 cases, including downstream and partner cases), earned income sits on an edge only at 250, 1,000, 2,000, 3,000, 5,000 and 5,001, and unearned income only at 250, 1,000, 1,500, 2,000, 5,000 and 5,001. Of the 80 one-dollar moves of a band start, 12 fail a test and 68 pass every test. For example, unearned 2,751 → 2,752 and earned 1,251 → 1,250 pass all 85 cases. Two encodings that break the PR body's whole-dollar reading also pass all 85: thresholds at band top + 0.0001 ("over $250") and np.round() before the lookup.
• Why this severity: missing edge cases for an already-tested variable, and the edges are the content of this PR.
• Note on cents: 120.510, the chart and HFS 3792 say nothing about cents. IDHS PM 15-04-03 is in the AABD medical-only income section, and no HBWD text applies it to the premium. Case 13 pins the reading that the PR body states. Band-start keying gives the whole-dollar result in any case.
• Fix: append Cases 9-13. They pass at the head, and each variant above fails at least one of them (for example, Case 11 person11 gives 188, not 206, when unearned 2,751 moves to 2,752). The expected values come only from the chart's first column and first row, so no case asserts one of the 7 cells where a source differs from the sum.

- name: Case 9, earned income at each band start (251, 501, ..., 5,001) pays that band's premium.
  period: 2026-01
  absolute_error_margin: 0.01
  input:
    people:
      person1: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 251, il_hbwd_countable_unearned_income: 0}
      person2: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 501, il_hbwd_countable_unearned_income: 0}
      person3: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 751, il_hbwd_countable_unearned_income: 0}
      person4: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 1_001, il_hbwd_countable_unearned_income: 0}
      person5: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 1_251, il_hbwd_countable_unearned_income: 0}
      person6: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 1_501, il_hbwd_countable_unearned_income: 0}
      person7: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 1_751, il_hbwd_countable_unearned_income: 0}
      person8: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 2_001, il_hbwd_countable_unearned_income: 0}
      person9: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 2_251, il_hbwd_countable_unearned_income: 0}
      person10: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 2_501, il_hbwd_countable_unearned_income: 0}
      person11: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 2_751, il_hbwd_countable_unearned_income: 0}
      person12: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 3_001, il_hbwd_countable_unearned_income: 0}
      person13: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 3_251, il_hbwd_countable_unearned_income: 0}
      person14: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 3_501, il_hbwd_countable_unearned_income: 0}
      person15: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 3_751, il_hbwd_countable_unearned_income: 0}
      person16: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 4_001, il_hbwd_countable_unearned_income: 0}
      person17: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 4_251, il_hbwd_countable_unearned_income: 0}
      person18: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 4_501, il_hbwd_countable_unearned_income: 0}
      person19: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 4_751, il_hbwd_countable_unearned_income: 0}
      person20: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 5_001, il_hbwd_countable_unearned_income: 0}
    households:
      household:
        members: [person1, person2, person3, person4, person5, person6, person7, person8, person9, person10,
          person11, person12, person13, person14, person15, person16, person17, person18, person19, person20]
        state_code: IL
  output:
    il_hbwd_premium: [6, 13, 19, 25, 31, 38, 44, 50, 56, 63, 69, 75, 81, 88, 94, 100, 106, 113, 119, 125]
    # First column of the chart, bands 251-500 through 5000+.

Cases 10-12 copy Case 9. Change only the name, the 20 income inputs (the other income stays 0) and the output:

• Case 10, earned income one dollar below each band start (250, 500, ..., 5,000) pays the lower band's premium. Earned 250, 500, ..., 5_000. Output [0, 6, 13, 19, 25, 31, 38, 44, 50, 56, 63, 69, 75, 81, 88, 94, 100, 106, 113, 119], comment # First column of the chart, bands 0-250 through 4751-5000.
• Case 11, unearned income at each band start (251, 501, ..., 5,001) pays that band's premium. Earned 0, unearned 251, 501, ..., 5_001. Output [19, 38, 56, 75, 94, 113, 131, 150, 169, 188, 206, 225, 244, 263, 281, 300, 319, 338, 356, 375], comment # First row of the chart, bands 251-500 through 5000+.
• Case 12, unearned income one dollar below each band start (250, 500, ..., 5,000) pays the lower band's premium. Earned 0, unearned 250, 500, ..., 5_000. Output [0, 19, 38, 56, 75, 94, 113, 131, 150, 169, 188, 206, 225, 244, 263, 281, 300, 319, 338, 356], comment # First row of the chart, bands 0-250 through 4751-5000.

- name: Case 13, income with cents stays in the band of its whole-dollar amount.
  period: 2026-01
  absolute_error_margin: 0.01
  input:
    people:
      person1: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 250.99, il_hbwd_countable_unearned_income: 0}
      person2: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 5_000.99, il_hbwd_countable_unearned_income: 0}
      person3: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 0, il_hbwd_countable_unearned_income: 250.99}
      person4: {il_hbwd_eligible: true, il_hbwd_countable_earned_income: 0, il_hbwd_countable_unearned_income: 5_000.99}
    households:
      household:
        members: [person1, person2, person3, person4]
        state_code: IL
  output:
    il_hbwd_premium: [0, 119, 0, 356]
    # $250.99 → 0-250 band; $5,000.99 → 4751-5000 band (IDHS PM 15-04-03 drops cents).

Suggestions

G1. Record the whole-dollar reading in the parameter comment.

• Where: E:11-12 and U:11-12.
• Evidence: only the PR body explains why no rounding step is needed.
• Note: whether PM 15-04-03 governs the HBWD premium is not settled (see SA1). The comment names the source without claiming an HBWD rule.
• Fix (after line 12 in both files):

    # Bands start at whole dollars, so income with cents stays in the band of its whole-dollar
    # amount, as IDHS PM 15-04-03 drops cents.

G2. Pin the agency's published example, which this PR fixes.

• Where: T, after Case 13.
• Evidence: HFS 3792 is the only place where the agency computes a premium in public. Example 1 sits on an edge that this PR moves. The head gives $75, and the merge base gave $94.
• Fix (passes at the head):

- name: Case 14, HFS 3792 Example 1, $900 from work and $1,000 from Social Security.
  period: 2026-01
  absolute_error_margin: 0.01
  input:
    state_code: IL
    il_hbwd_eligible: true
    il_hbwd_countable_earned_income: 900
    il_hbwd_countable_unearned_income: 1_000
  output:
    il_hbwd_premium: 75
    # HFS 3792 (R-03-13), Example 1: the monthly premium is $75.
    # Earned $900 → 751-1000 band → $19
    # Unearned $1,000 → 751-1000 band → $56
    # Total: $19 + $56 = $75

G3. Complete the source-conflict record before the ruling.

• Where: R:2170 (years.from: 2022), R:2175-2185 (sources and readings), R:2197-2201 (note).
• Evidence: (a) The 2009 text prints the same 79, and the chart with 69 is from June 2009, so the conflict dates from 2009. (b) HFS 3792 is a third source and prints $69. (c) The "$19 steps" argument holds along the row only. The column is stronger evidence: it reads 63, 79, 76 for earned 1001-1250, 1251-1500 and 1501-1750, so under 79 more earned income lowers the premium. (d) The note leaves out CONTRIBUTING.md:102 ("Regulations are law"), which favours 79. The agency tie-break (:103) applies only when the text does not settle the reading. The record must state both sides.
• Fix (passes the register validator). Replace the matching parts of the entry:

  years:
    from: 2009
    type: calendar

  - kind: form
    citation: HFS HBWD Premium Chart (one page, June 2009)
    url: https://hfs.illinois.gov/content/dam/soi/en/web/hfs/medicalprograms/hbwd/documents/hfsweb004.pdf
  - kind: guidance
    citation: HFS 3792 (R-03-13), Health Benefits for Workers with Disabilities brochure, premium table
    url: https://hfs.illinois.gov/medicalprograms/hbwd/hfs3792hbwd.html
  readings:
    statute: No statute is supplied; the regulation's table prints 79 for earned income 1251-1500 and unearned
      income 501-750, in the text effective February 1, 2009 (33 Ill. Reg. 1681) and unchanged by the amendment
      effective March 25, 2022 (46 Ill. Reg. 5705).
    other: The HFS premium chart (June 2009) and the HFS 3792 brochure print 69 for the same cell, which PE-US
      gives (31 + 38).

  note: 'PE-US gives the chart''s 69 because it adds an earned component (31) and an unearned component (38). The
    sources are a regulation and two agency publications, with no statute, so the statute-first rule does not settle
    the cell. CONTRIBUTING treats a regulation as binding ("Regulations are law"), which favours 79; the agency
    tie-break applies only where the text and the canons do not settle a reading. Against 79: the regulation''s row
    reads 31, 50, 79, 87, 106 (steps of about $19 elsewhere), and its column reads 63, 79, 76 for earned income
    1001-1250, 1251-1500 and 1501-1750, so under 79 more earned income would lower the premium; the agency''s chart
    and brochure print 69. The ruling decides whether 79 binds or is a scrivener''s error. The other 440 cells of the
    regulation and the chart agree. Issue 10471 proposes encoding the full table, which needs this ruling first.'

G4. Point the reference at (n)(3) and cite the chart.

• Where: E:13-14 and U:13-14.
• Evidence: the title cites § 120.510 (n), but the table is in (n)(3), as the new comment says. The PR relies on the HFS chart for the band labels, but the files cite only the regulation. The title is correct, only imprecise.
• Fix (both files):

    - title: 89 Ill. Admin. Code § 120.510(n)(3)
      href: https://www.law.cornell.edu/regulations/illinois/Ill-Admin-Code-tit-89-SS-120.510
    - title: Illinois HFS, Health Benefits for Workers with Disabilities Premium Chart
      href: https://hfs.illinois.gov/content/dam/soi/en/web/hfs/medicalprograms/hbwd/documents/hfsweb004.pdf#page=1

G5. Correct the stale integration comments next to the ones this PR rewrote.

• Where: I:115, I:119, I:136, I:160, I:162. These lines predate the PR.
• Evidence (probe at the head): Case 4 is $3,498.62 earned plus $400 unearned, $3,898.62 in total (comment "$4,089"). Case 5 is $4,378.95 (comment "$4,618"). Case 6 is exactly $849.95 and $2,349.95.
• Fix: I:115 becomes # Countable income: $3,498.62 earned + $400 unearned = $3,898.62. I:119 becomes three lines: # Earned $3,498.62 → 3251-3500 band → $81, # Unearned $400 → 251-500 band → $19, # Total: $81 + $19 = $100. I:136 becomes # Countable income after expenses and exemptions: $4,378.95, I:160 becomes # Earned after expenses and exemptions: $849.95, and I:162 becomes # Total: $2,349.95.

G6. Finish the change from "bracket" to "band" wording.

• Where: T:57, T:67-68, T:71, T:118-119 and il_hbwd_person.yaml:34 (same test folder). The PR changed every other case to "N-M band".
• Fix: - name: Case 5, no premium in the 0-250 band.; # Earned $200 → 0-250 band → $0 and # Unearned $0 → 0-250 band → $0; - name: Case 6, income of exactly $250 is in the 0-250 band.; # Earned $1,802.61 → 1751-2000 band → $44 and # Unearned $0 → 0-250 band → $0; # Both incomes in the 0-250 band → $0 premium.

Process note (not counted)

The body has axiom: TheAxiomFoundation/rulespec-us#1926 queued. Issue #1926 is open and covers the 120.510(n) premium tables for this PR. No action.

Investigated and cleared

• $5,000 in two bands. Both sources print "4751-5000" and "5000+". The PR puts $5,000 in 4751-5000, like every other band, and documents it at E:11-12 and in Case 8. A one-person unit cannot reach $5,000 of countable earned income under the 350% FPG test. A documented choice on a source ambiguity, with a small effect.
• "+0.0001" rule. It is for "over X" wording. Closed "251-500" ranges use band starts, as the PR does.
• Eligibility. No eligibility variable reads the premium, so no eligibility flips.
• Microsimulation. Premiums fall by one band step ($6-7 earned, $18-19 unearned) only for monthly income in [$250k, $250k + 1), mostly round unearned amounts such as $12,000 a year of SSDI. Small, with the expected sign.
• Negative income. It returns $0, the band-0 premium.
• CI "skipping" checks. They come from an earlier workflow run at the same SHA. A later run at that SHA ran every shard and passed.

Out of scope / follow-up

• The 79-versus-69 ruling. The PR body asks PavelMakarchuk to rule. It does not block this PR, because the PR does not change the cell. G3 completes the record first.
• The additive structure. Six cells are $1 off the sum in both sources. This predates the PR and is tracked in #10471.
• Start date. The values start 2022-04-01, but the table is in force since 2009-02-01. No numeric effect, because the first value applies to earlier periods.
• Cents in countable income. PM 15-04-03 drops cents per paycheck and per payment, but the model keeps cents in the total ($1,728.13 in integration Case 1). This is income computation, not band logic.

Validation Summary

Check	Result
CI	32 pass, 3 skipping, 0 fail
Local tests	HBWD folder 63/63; all 85 cases that reach the premium pass; register validator 103/103
Proposed fixes	Cases 9-14 pass at the head; all fixes together 91/91; validator passes after G3
Partner contract tests	None edited. IL HBWD partner cases 12/12; partner healthcare folder 70/70
Regulatory	Thresholds and amounts match 120.510(n)(3) and the chart. One registered conflict (79/69)
References	All links live. Title precision and chart citation (G4)
Code patterns	YAML only. Scale type, units, period and changelog correct
Tests	Changed values correct. Edge cases missing (SA1)
Mutation	80 start shifts: 12 caught. Two whole-dollar-breaking encodings survive. Cases 9-13 catch all
Microsimulation	Small premium drop at exact multiples of $250; no eligibility change
Branch	4 ahead / 19 behind main; MERGEABLE
Other reviews	None on the PR

Review Severity: REQUEST_CHANGES

One should-address item is open: SA1, the missing band-start, below-start and cents tests. The thresholds and amounts are correct.

Next Steps

To auto-fix issues: run the fix-pr workflow for this PR.

1. Append Cases 9-13 to T after line 134 (SA1).
2. Take the suggestions that you agree with. G5 and G6 change comments and names only.
3. Run make format, then run the HBWD test folder and tests/code_health/test_source_conflicts.py.
4. Leave the 79-versus-69 cell as it is until PavelMakarchuk rules. Then #10471 can continue.

[hua7450]

Thanks for the review. Round-1 changes are in a47acccc39, and d345ffec83 merges main (4c65c82191). Dispositions:

Item	Disposition
SA1 band starts, below-start and cents	Done: Cases 9-12 (each income at every band start 251 ... 5,001 and one dollar below, 250 ... 5,000), with expected values from the chart's first column and first row, and Case 13 (cents).
G1 whole-dollar comment	Done, in both parameter files.
G2 HFS 3792 Example 1	Done: Case 14 ($900 from work, $1,000 from Social Security: $75). I confirmed the example and the brochure's $69 cell on the live page.
G3 complete the conflict record	Done, with your text: years.from: 2009, the chart dated June 2009, the HFS 3792 brochure as a third source, both readings with the 2009 and 2022 texts, and a note that gives both sides (the binding regulation, and the row, column and agency evidence). follows: unknown and state: unreviewed are unchanged. The cell stays at 69 until your ruling.
G4 (n)(3) and the chart	Done, with one change: the chart href has no #page=1, because the PDF has a single page and our reference rule drops the fragment for one-page PDFs.
G5 stale integration comments	Done: Cases 4, 5 and 6 now give the exact amounts ($3,898.62 with the $81 + $19 breakdown, $4,378.95, $849.95 and $2,349.95).
G6 "bracket" to "band"	Done in il_hbwd_premium.yaml Cases 5, 6 and 7 and il_hbwd_person.yaml Case 3.

The body now covers the new cases and the fuller conflict record, and it repeats the request for your ruling on 79 versus 69. TheAxiomFoundation/rulespec-us#1926 now describes this head, with the brochure as a source and Example 1 as a companion case. No tests were run locally; every new expected value is a chart cell I checked by hand.

[hua7450]

Ruling on us-il:hbwd-premium-earned-1251-unearned-501-cell: the regulation's 79. I'm the project lead for these Medicaid fixes, so this settles it, and it replaces the sign-off request in the PR body.

• 89 Ill. Adm. Code 120.510(n)(3) prints 79 for earned income 1251-1500 with unearned income 501-750.
• The text has done so since February 1, 2009, and the 2022 amendment left it unchanged.
• Under "Regulations are law" the regulation binds. The written canons have no exception for an apparent typesetting error, so the HFS chart's and the HFS 3792 brochure's 69 doesn't override it.

The model keeps 69 for now. It adds an earned component and an unearned component (31 + 38), so it can't represent a single cell that departs from the sum. Issue 10471, which encodes the full 21 x 21 table, will bring this cell to 79.

The candidate will record this comment as the ruling, with PE-US marked as lagging until then. The register entry follows the merge.

[PavelMakarchuk]

Program Review

PR #10372, "Key Illinois HBWD premium bands at the chart's band starts" (hua7450). Round 2, incremental review of bd587e45b7..50d03d1e89.

IDs: C = critical, SA = should address, G = suggestion. New items continue the numbering after the highest prior ID in each class.

Paths: E = policyengine_us/parameters/gov/states/il/hfs/hbwd/premium/earned.yaml, U = .../hbwd/premium/unearned.yaml, T = policyengine_us/tests/policy/baseline/gov/states/il/hfs/hbwd/il_hbwd_premium.yaml, I = .../hbwd/integration.yaml (same test folder), R = source_conflicts/candidates.yaml. Line numbers refer to the reviewed head.

Source Documents

• Reviewed head SHA: 50d03d1
• Mode: incremental (round 2)
• Scope: two author commits and two merges of main. a47acccc39 adds Cases 9-14 (T:136-302), the whole-dollar comment and the chart reference (E:13-18, U:13-18), the exact integration comments (I:115, I:119-121, I:138, I:162-164), the "band" wording (T:57, T:67-68, T:71, T:118-119, il_hbwd_person.yaml:34) and the fuller conflict record. b3a2fbf23f records the 79-or-69 ruling in R:2167-2208. d345ffec83 and 50d03d1e89 merge main. Their only change in PR files is a new us-va: candidate in R from #10253.
• Regulation: 89 Ill. Adm. Code 120.510(n)(3). The live ILGA page timed out, so I used the Wayback captures of the 2009 text (33 Ill. Reg. 1681, effective February 1, 2009) and the 2022 text (46 Ill. Reg. 5705, effective March 25, 2022), plus Cornell LII (the parameter href). I parsed all 441 cells from each copy. The three tables are identical, and each prints 79 for earned 1251-1500 with unearned 501-750.
• Chart: HFS HBWD Premium Chart (live, one page, created June 12, 2009). It differs from the regulation only at that cell (69).
• Brochure: HFS 3792 (R-03-13). Example 1 gives $75. The partial table prints $69 at $1,251-1,500 and $501-750.
• Ruling and tracking: the author's ruling comment (3:20pm EDT on 2026-10-27) and issue #10471 with its comment of 3:21pm EDT.
• Repository: CONTRIBUTING.md:97-121 (source conflicts), the R header legend, policyengine_us/tests/code_health/test_source_conflicts.py, and the policyengine-claude PDF page rule (single-page PDFs are exempt).

Branch Status

• 8 commits ahead and 34 behind main (1976c4b600). CI at the head: 33 pass, 0 fail.
• GitHub showed mergeable: UNKNOWN at review time. A read-only git merge-tree against main shows that both sides changed R, with no conflict. Main's new hunks in R start at line 2527 and later, and the PR's hunk is at 2167-2208.
• Main's vendored register and pin are unchanged since the merge base and have no us-il:hbwd entry, so the candidate is not yet registered.
• No other open PR touches HBWD. No review or comment on the PR came after the ruling comment.

Round 1 items

ID	Status	Evidence
SA1	RESOLVED	Cases 9-13 at T:136-287 are the proposed cases. Their expected values match the first column and first row of the regulation (2009, 2022 and Cornell copies) and the chart, cell by cell. In a probe, every one of the 80 one-dollar band-start shifts fails at least one of Cases 9-12, and so does every +$1 change to the 42 amounts. The "over $X" encoding (band top + 0.0001), np.round and np.ceil fail Case 13. np.floor passes, which is correct, because it gives the same whole-dollar result. One real mutation through policyengine-core (unearned 2_751 to 2_752 in a package copy) fails Case 11 (1 failed, 13 passed). The tests work.
G1	RESOLVED	E:13-14 and U:13-14 carry the proposed text.
G2	RESOLVED	Case 14 at T:289-302. HFS 3792 Example 1 says that $900 from a job and $1,000 from Social Security give a $75 premium. Earned 751-1000 with unearned 751-1000 is 75 in the regulation, the chart and the brochure. The old thresholds (750 and 1,000 as lower bounds) gave 19 + 75 = 94, as the body says.
G3	RESOLVED	R:2167-2208 has years.from: 2009, the June 2009 date on the chart, HFS 3792 as a guidance source, and both readings with the 2009 and 2022 citations. All of these match the sources. The ruling commit b3a2fbf23f then replaced the two-sided note with the ruling. The row and column evidence against 79 is gone from the note, but the readings still give both texts. The register entry can carry that evidence as an alternative.
G4	RESOLVED	E:15-18, U:15-18: the title is now § 120.510(n)(3), and the chart is cited. The author left out the #page=1 that I proposed. That is correct: the chart has one page (pdfinfo: Pages 1), and the PDF page rule exempts single-page PDFs. I withdraw the #page=1 part of G4. Both hrefs return 200.
G5	RESOLVED	A probe at the head gives Case 1 earned $1,728.13 and unearned $1,000. Case 4 gives $3,498.62 + $400 = $3,898.62, a $100 premium. Case 5 gives $4,378.95, and Case 6 gives $849.95 and $2,349.95. I:115, I:119-121, I:138 and I:162-164 match these values. The main merges did not move them.
G6	RESOLVED	T:57, T:67-68, T:71, T:118-119 and il_hbwd_person.yaml:34 now say "band". "Bracket" stays only in the two parameter description lines and in il_hbwd_premium.py:22. G6 did not cover these lines, and "bracket" is the usual word for a scale there.

Round 1 process note (axiom: line): unchanged, TheAxiomFoundation/rulespec-us#1926 queued. Not counted.

79/69 ruling record

I did not rule on 79 against 69. I checked only that the PR records the author's ruling consistently.

• Ruling. hua7450 commented at 3:20pm EDT on 2026-10-27, as the project lead for these Medicaid fixes, and ruled for the regulation's 79 under "Regulations are law". The link in the body and in R:2202-2203 resolves to that comment.
• PR body. It says "Source conflict ruled: regulation (79); PE-US keeps 69 until #10471", gives the 31 + 38 reason, and says state: ruled, follows: conflicting-source. The request for Pavel's sign-off is gone from the body.
• Register record (R:2167-2208). It has state: ruled and pe_us.follows: conflicting-source. The header legend defines that value as "against the statute; ... calls for a model change", with the regulation in the statute reading here, so PE-US is marked as lagging. The note gives the date, the role (no personal name, as the register requires), the record URL, the basis, why the sum gives 69, and #10471 as the fix. The example still states $79 against $69. The validator passes (103/103).
• Code comments. E:9-10 and U:9-10 say that the sum "reproduces 434 of the 441 cells in 120.510(n)(3); see issue 10471 for the others". That counts against the regulation's table, so it agrees with the ruling, but it does not name the ruled cell or say that the model lags. See G7. A # Source conflict marker cannot be added yet: the validator fails a marker whose entry is not in the pinned register, and the author says the register entry follows the merge.
• Issue #10471. It is open. The body names the 79/69 cell and says that "the encoded table should follow its ruling". The 3:21pm EDT comment says that the table "should give 79" for that cell and that follows then changes to statute, or the candidate moves to the register. The issue covers the cell. Its proposed test list names the six $1-off cells and the corners, but not this cell. The comment covers it.
• No test asserts the cell. Cases 9-13 set one income to 0, so they use only the first row and column. Case 14 is the 751-1000 by 751-1000 cell. The round 1 sweep found none of the seven off-sum cells in the 85 cases that reach the premium.
• Verdict impact. The lag is documented in the body, the record and the issue, and #10471 tracks it, so it does not count toward the verdict.

New findings

Critical

None.

Should address

None.

Suggestions

G7. Name the ruled cell in the parameter comment.

• Where: E:9-10 and U:9-10.
• Evidence: the body and R state the ruling and the lag. The parameter files, which R lists under pe_us.code, say only "see issue 10471 for the others". A reader of the scale cannot see that one of the seven cells is a ruled lag and not a $1 rounding gap.
• Fix (after line 10 in both files). These lines do not match the validator's marker, conflict-comment or privacy patterns, so no KNOWN_UNMARKED change is needed:

    # One of the seven, earned 1251-1500 with unearned 501-750, is ruled 79, as the regulation prints
    # (candidate us-il:hbwd-premium-earned-1251-unearned-501-cell); this sum gives 69 until issue 10471.

Tests run

• policyengine-core test policyengine_us/tests/policy/baseline/gov/states/il/hfs/hbwd -c policyengine_us: 69 passed. Round 1 had 63, and the PR added 6.
• pytest policyengine_us/tests/code_health/test_source_conflicts.py: 103 passed.
• Hand check of Cases 9-14. I parsed the 2009, 2022 and Cornell copies of (n)(3) and the chart, and all four have the same first column and first row. Earned: 0, 6, 13, 19, 25, 31, 38, 44, 50, 56, 63, 69, 75, 81, 88, 94, 100, 106, 113, 119, 125. Unearned: 0, 19, 38, 56, 75, 94, 113, 131, 150, 169, 188, 206, 225, 244, 263, 281, 300, 319, 338, 356, 375. Case 9 is the earned list from item 2 on, and Case 10 is the earned list without its last item. Cases 11 and 12 are the same for the unearned list. Case 13: $250.99 is below the 251 start, so the band is 0-250 and the premium is $0. $5,000.99 is below the 5,001 start, so the band is 4751-5000 and the premium is $119 earned or $356 unearned. Case 14: 19 + 56 = 75, and the cell is 75 in all three sources. The sum matches 434 regulation cells and 435 chart cells, as the body says.
• Mutation probe of the scales against Cases 9-14: 80 of 80 start shifts and 42 of 42 amount changes fail at least one case. The "over $X", np.round and np.ceil variants fail Case 13.
• Real mutation in a package copy (deleted afterwards): unearned 2_751 changed to 2_752 fails Case 11.
• Probe of integration Cases 1, 4, 5 and 6 at the head (values in the G5 row).
• git merge-tree against main 1976c4b600: no conflict.

Verdict

APPROVE: SA1 and G1-G6 are resolved, and no critical or should-address item is open. The 79/69 lag is documented and tracked in #10471, so it does not block. G7 is optional. The 79/69 ruling is the author's, as project lead; this review checks only that the PR records it consistently.