Repository navigation
Update Cascade Care Savings to the 2027 amounts, with the 200% FPL split - #11152
Conversation
From plan year 2027, Washington pays $35 a month per member with federal subsidies up to 200% of the federal poverty line and $20 above it, and $185 without federal subsidies (WAHBE memo of 2026-10-01, p.1; Final PY 2027 policy Section 5(1)(b)). The with-subsidy amount becomes a two-band scale on the income-to-poverty-line ratio. A new tax-unit variable holds that exact ratio, and both the 250% income limit and the 200% split read it. Per the project lead's 2026-11-18 ruling, a ratio at or below 2.00 gets $35 and any ratio above 2.00 gets $20. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #11152 +/- ##
=========================================
Coverage 100.00% 100.00%
=========================================
Files 1 4 +3
Lines 24 74 +50
=========================================
+ Hits 24 74 +50
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
hua7450
left a comment
There was a problem hiding this comment.
Thanks, David. The Washington 2027 amounts check out: $35 a month per member with federal subsidies up to 200% FPL, $20 above that up to 250%, and $185 without federal subsidies match the PMPM memo of 2026-10-01 (p. 1) and §5(1)(b) of the final PY2027 policy (p. 6), which is "Effective for Coverage Beginning January 1, 2027". The 250% limit (§4(1)(c)) and the household cap (§5(1)(d)) are unchanged. Paying $35 at exactly 200% follows the policy's "up to 200% FPL" and the CSR tier edge in 45 CFR 155.305(g)(2). 2026 results are unchanged, the scale's -.inf first threshold is needed and present, and the new ratio variable replaces Washington's inline division instead of adding a copy. Approving.
Optional:
- Follow-up issue for lawfully present members who lose the federal credit in 2027. The memo (p. 6) says these members "will move to non-federally subsidized group", and its p. 8 exhibit has that group growing from 6,550 to 13,400. In the model a refugee or asylee at about 180% FPL gets $0 in 2027, because Group 3 requires
UNDOCUMENTED. It is the same on main and you defer it in the PR body; an issue would keep it tracked (any fix also depends on issue 9553). The documentation atwa_cascade_care_savings_member_amount.pylines 46-49 still says these members "are not separately identifiable in the model", which stops being true in 2027. - A wage-driven 2027 pair in
integration.yaml. Every 2027 amount case suppliesis_aca_ptc_eligible,aca_magi,aca_fpg,slcspandaca_ptc, and the only 2027 integration scenario sits at 140% FPL, so the $20 band is never reached from real inputs. A single King County adult with wages of $25,000 (ratio 1.566, 12 × 35 = 420) and $35,000 (2.193, 12 × 20 = 240) next to Scenarios 13-14 would cover both bands. - Leaner tests. About 12 of the 36 new cases repeat a class another new case pins: member-amount Cases 11, 12, 18 and 19;
wa_cascade_care_savings.yamlCases 27-30 and 35-37; ratio Case 2. - Exhibit labels at exactly 200%. The Wakely exhibit in the memo (p. 8) prints "Fed Sub < 200% FPL $35" and "Fed Sub >= 200% FPL $20". The policy text governs, so the value is right; one sentence in the
with_federal_subsidy.yamlcomment would record the difference. - Section number.
wa_cascade_care_savings_eligible.pyline 33 cites "Section 4(2)" for the filing-behavior exclusions. In the PY2027 policy those are §4(3); §4(2) is now the plan-eligibility subsection. - Two assertions. Ratio Case 8 ("not a division by zero") passes with or without the zero-guideline guard. The
aca_magi_fraction: 2and2.5outputs run under a 0.01 margin, which is one truncation step; 0.001 would make them bind.
Series-wide (follow-up, not for this PR). wa_cascade_care_savings_fpl_ratio has nothing specific to Washington in it, and the same exact ratio is still inline in aca_ptc_income_eligible and eight other state premium files. One shared guarded variable under gov/aca/eligibility remains the follow-up.
- Member-amount documentation: lawfully present members who lose the
federal credit in 2027 move to the non-federally subsidized group (memo
p. 6), but Group 3 requires UNDOCUMENTED, so e.g. a refugee or asylee at
about 180% FPL gets $0 in 2027 (known limitation, issue 9553).
- Eligibility documentation: cite Section 4(3) of the PY2027 policy for the
filing-behavior exclusions.
- with_federal_subsidy.yaml: note that the Wakely exhibit (memo p. 8) labels
the bands "< 200%" and ">= 200%", but the policy text ("up to 200% FPL")
governs, so exactly 200% gets $35.
- Ratio Case 8 now asserts the member amount (12 * 20 = 240), which an
unguarded 0 / 0 (NaN, 0) fails; tighten Cases 28 and 34 margins to 0.001
so the aca_magi_fraction assertions bind.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks, Ziming. I made four of your optional changes in 32101b0:
Not done: the wage-driven integration pair, leaner tests, and the follow-up issue. The Cascade Care tests pass locally (118/118). |
What changes
Washington's Cascade Care Savings amounts change for plan year 2027:
The 250% income limit stays the same (policy §4(1)(c), p.3), and so does the household cap (§5(1)(d), p.7). The model has no end date for the program, so 2028 onward carries the 2027 amounts forward. The
in_effectcomment now records the funding risk: PY2027 ends the biennium, and WAHBE "has not yet secured funding for PY 2028" (memo p.3).How the model does it:
amount/with_federal_subsidyis now a two-bandsingle_amountscale on the income-to-poverty-line ratio. Its thresholds are-infand 2. Its amounts are 55/55 from 2026 and 35/20 from 2027. The formula reads it withcalc(ratio, right=True).amount/without_federal_subsidy: 185 from 2027-01-01 (was 250).wa_cascade_care_savings_fpl_ratio(YEAR,/1, WA only):max_(aca_magi, 0).astype(np.float64) / aca_fpg, without truncation. Whereaca_fpgis 0, it divides by 1 and returns an infinite ratio, asaca_magi_fractiondoes, which fails the income limit. The eligibility gate from Decide Washington Cascade Care Savings eligibility on the exact FPL ratio #10314 now reads this variable instead of its inline ratio, and the member amount reads it for the split. The 250% limit and the 200% split are therefore decided on the same ratio.fpl_limitandin_effectget their 2027 references (policy §4(1)(c), p.3; policy §2, p.1; memo p.3). Their values don't change.documentationstrings inwa_cascade_care_savings.pyandwa_cascade_care_savings_member_amount.pygave only the $55 and $250 amounts. They now give both years.Readings
Incomes above 200% but below 201% of FPL. The policy sets "one amount for customers up to 200% FPL and one amount for customers 201 – 250% FPL" (§5(1)(b)), and the memo has the same bands. Neither says which amount applies above 200% but below 201%. On 2026-11-18 the project lead ruled that this is decided on the exact ratio: a household at or below 2.00 gets $35, and any ratio above 2.00 gets $20. The parameter comment records the ruling. A whole-percent reading would have paid $35 up to 200.99%. Two cases pin the ruling at 200.5%: Case 29 in
wa_cascade_care_savings.yamland Case 11 inwa_cascade_care_savings_member_amount.yaml.The scale's edges. I read
SingleAmountTaxScale.calcin the installed policyengine-core (3.33.2) and checked that it is identical in the locked 3.36.0. It callsnumpy.digitize(base, [-inf] + thresholds + [inf], right=right)and pads the amounts with a 0 at each end. Withright=True, a ratio equal to a threshold falls in the band below it. So 2.00 gets $35 and 2.000001 gets $20. The first threshold is-infbecause a threshold of 0 would send a ratio of exactly 0 into the padding cell and pay $0. Member Case 13 (zero income, $35) tests this.Precision. The ratio is divided in float64 but stored as float32, like every float variable. For a whole-dollar guideline, the smallest float32 income above 2.00 or 2.50 times the guideline still rounds to a stored ratio above the threshold, so storing it as float32 doesn't move the limit or the split. The one-cent cases below check this.
Not in this PR: widening Group 3 to the lawfully present people who lose the federal credit in 2027 (audit §3.3 (6)), which is a later step. Plan choice isn't modeled, so the new Bronze eligibility and the carrier conditions (§4(2)) have no effect on the model.
Tests
Changed expected values (both have the same inputs: one Group 1 member at 140% FPL, SLCSP $6,000, credit $5,000, 2027):
integration.yamlScenario 9: 660 → 420. The base is 12 × $35 = 420, under the cap of 6,000 − 5,000 = 1,000.wa_cascade_care_savings.yamlCase 16: 660 → 420, for the same reason.Added (supplied
aca_magiandaca_fpgof $10,000 unless noted):wa_cascade_care_savings_fpl_ratio.yaml(new, 9 cases): 140%; exactly 200%; 200.5% (exact ratio 2.005, truncated fraction 2.00); 250.005%; negative income (0); guidelines derived from the model for 2026 ($15,650 → 2.5) and 2027 ($15,960 → 2.0); a zero guideline (tax unit ineligible, amount 0); a household outside Washington (0).wa_cascade_care_savings_member_amount.yaml, Cases 8–20. For 2027, Group 1 gets 420 at 140%, 420 at exactly 200%, 240 one cent above 200%, 240 at 200.5%, 240 at 201% and 420 at zero income. Group 3 gets 2,220 at 150% and 240%. In a mixed household at 220%, the amounts are 240 / 2,220 / 0, and a Group 2 member gets 2,220 beside a Group 1 member's 240. For 2026, Group 1 still gets 660 at exactly 200%, one cent above and 240%.wa_cascade_care_savings.yaml, Cases 26–37. For 2027: 300 when the cap binds; 420 at exactly 200%; 240 one cent above it, at 200.5% and at 201%. Two Group 1 members and one Group 3 member get 12 × (2 × 35 + 185) = 3,060 at 180% and 12 × (2 × 20 + 185) = 2,700 at 220%. Exactly 250% is eligible for 240, and 250.005% gets 0. Group 3 gets 2,220 at 150% and 240%. For 2026, Group 1 at 240% still gets 660.wa_cascade_care_savings_eligible.yaml, Cases 11–12: 2027 at the model's 2026 guideline. $39,900 (exactly 250%) is eligible. $39,900.01 is ineligible, although the truncated fraction is 2.50.I recomputed every audit §3.4 value from the memo and the policy, and all of them agree. The audit's Case 8 has the same inputs as the updated Case 16, so I didn't add it twice. Its Case 15 (refugee) belongs to the Group 3 change.
Tested on Modal: run
s2-wa-a, 159/159 passed (the eight YAML files: the six Cascade Care Savings files,healthcare_benefit_value.yamlandhousehold_health_benefits.yaml), plus 80/80 intest_reference_urls.py,test_parameter_yaml_keys.pyandtest_claimable_filer_ptc_invariants.py. In that run,test_source_conflicts.pyfailed only because the harness mountspolicyengine_us/without the repo-rootsource_conflicts/folder (10 FileNotFoundErrors), plus one case whose tokenizer behaviour differs on the image's Python 3.11. Runs2-wa-b, withsource_conflicts/mounted on Python 3.13: 251/251 passed.axiom: needed
🤖 Generated with Claude Code