[MaxGhenis]

Ready for review. The exact-ratio eligibility fix and the move to the Final 2027 Program Design are done, with tests, Axiom parity and a full-file impact run (see Impact). A GPT-6.1 Sol shell-backed review approved 93371bc7e8. The two commits since then, a merge of main and a test change for the federal 2027 table, are under review.

#10301 has landed. It merged to main as 81056c4acb, a squash of its head cc1af12cd1, which this branch already contained. 05ff1686f9 merges main into this branch, so the diff now shows only this PR's changes (see Coordination).

This PR makes two changes, one for each plan year:

1. PY2026: eligibility is decided on the exact income-to-FPL ratio. The old gate used the truncated fraction, which admitted incomes from 165.003% to 165.994% of FPL.
2. PY2027: the schedule and the limit move from the superseded Program Design V4 to the Final 2027 Program Design. That means one paying band from 2.8% to 5.28%, and eligibility "at or below 200 percent".

1. PY2026: exact-ratio eligibility

ca_premium_subsidy_eligible compared the truncated aca_magi_fraction (np.floor(100 * magi / fpg) / 100) with <= 1.65. Truncation maps every income from 165.00% to 165.99% of FPL to 1.65, so all of them passed. The adopted 2026 Program Design, (d)(7) limits an applicable return filer to household income "at or above 100 percent and at or below 165 percent of the federal poverty level".

Reproduced on main (bda23eac4c) for a single filer in CA in 2026, with the 2025 one-person FPG of $15,650:

Income	Exact % of FPL	aca_magi_fraction	Eligible on main	Eligible here
$25,822.50	165.000%	1.65	yes	yes
$25,823	165.003%	1.65	yes	no
$25,978	165.994%	1.65	yes	no
$25,979	166.000%	1.66	no	no

Fix: eligibility is now decided on the exact ratio max(aca_magi, 0) / tax_unit_fpg(period.last_year), divided in float64.

• Why float64: the model stores both values as float32. A float32 quotient can round a ratio a fraction of a cent above 165% of a fractional guideline down to 1.65.
• Precedent: this follows the California convention for the predecessor program. FTB 3849 (2020) Table 1-2 truncates for the rate lookup but decides band edges on the exact percentage. Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 applies the same convention to the 150% band.

2. PY2027: the Final 2027 Program Design

On main, PY2027 follows Program Design V4: it has a knot at 165% and "under 200 percent" eligibility. The Final 2027 Program Design, PDF p.2, replaces it.

The Final Design's (d)(4) table:

Household income	Initial	Final
100% up to 150%	0%	0%
Above 150% up to 200%	2.8%	5.28%

Its (d)(7) defines an applicable return filer by household income "at or above 100 percent and at or below 200 percent of the federal poverty level".

The 2027 Program Insight (published 9/2/2026, Table 1) agrees: "150% – At or Below 200%: 2.8% – 5.28%".

Changes:

• Parameters: thresholds [1, 1.5, 2.0], initial [0, 0.028], final [0, 0.0528], and fpl_limit 2.0, all re-cited away from V4 and Board Resolution 2026-22. in_effect's comment now names the Final Design.
• Limit: both years' limits are now inclusive, so this PR drops the fpl_limit_inclusive parameter it had added for V4's strict limit. The gate tests floor <= r <= limit.
• Formula: Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's exact-ratio band selection (side="left") puts exactly 200% at the top of the paying band, at 5.28%. The only change to ca_premium_subsidy_applicable_percentage.py is that its band ratio is now divided in float64, like the eligibility gate. A float32 quotient can round a ratio just above 150% down onto 1.5, and the 0% band would then catch it. Example: $24,600.00390625 over a supplied FPG of $16,400.001953125 is 1.50000006. Before this change it got 0%; now it gets the paying band's starting rate. This only matters for fractional FPGs; see invariant 3.

Applicable percentage at the lookup fraction, V4 (main) against the Final Design (this PR):

Lookup fraction	V4	Final	Note
1.51	2.738%	2.8496%
1.575	3.05%	3.172%	amount Case 16: subsidy $3,237.50 → $3,207.00
1.65	3.41%	3.544%	165% is no longer a knot
1.825	4.255%	4.412%
1.99	5.0517%	5.2304%
2.00 (exactly 200%)	ineligible	5.28%

The Final paying-band rate is above V4's at every lookup fraction from 1.50 to 2.00, since both are linear between knots and the gap is positive at each knot.

Who gains and who loses in PY2027, against main, with the published whole-dollar guidelines:

• Above 150% and below 200%: eligibility is unchanged, so the state subsidy can only fall or stay at $0.
• Exactly 200%: households newly qualify, at 5.28%.
• At and below 150%: no change. Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 (merged as 81056c4acb) already put exactly 150% in the Design's 0% band, choosing the band on the exact ratio with side="left", so main and this head both apply 0% there (for example, income $23,940 against an FPG of $15,960). Before Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301, main's side="right" put a truncated 1.50 in the paying band at 2.69%, and for FPGs of $262,145 or more its float32 fraction also truncated some ratios up to 1.50 from as much as 1.4e-7 below 150%. Those gains are Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's and are already on main.

How this was checked. This search ran before #10301 merged, against main with V4's 150% boundary, so its at-or-below-150% statements describe #10301's change, now on main. #10301 changed only the band choice at and just below 150%, so the statements for (150%, 200%) and for 200% hold against current main as well. I replicated main's and this head's formulas in numpy, on float32 inputs as the model stores them. The search covered every whole-dollar FPG from $15,000 to $1,000,000. For each FPG it took 530 incomes at 150% and at 200%: the 129 float32 values within ±64 ulps, plus the 401 whole-cent incomes within ±$2. That is 522 million points per edge. The independent review ran a superset: every float32 income within the larger of 64 ulps and $2 of 100%, 150%, 165% and 200%, 776 million points in all. It confirmed each statement below.

• (150%, 200%): no eligibility change and no rate decrease.
• At or below 150%: below an FPG of $262,145 (2^18 + 1), every rate decrease was at exactly 150%. From there up, main's float32 truncation also rounds some ratios just below 150% up to 1.50. No ratio below 200% truncated up to 2.00.
• Band choice: float32 and float64 chose the same band at every point. Since a float32 quotient is within half a float32 ulp of the exact ratio, only incomes within about one ulp of 150% of FPG could differ, and the search covers 64 ulps.

Fractional FPGs are not covered by these statements. tax_unit_fpg is first_person + additional_person × (size − 1) from whole-dollar guidelines, so no PE household has one. They can still be supplied. Sol's review found MAGI $240,703.03125 over FPG $120,351.5234375, an exact ratio just below 200%. Main's float32 aca_magi_fraction rounds it up to 2.00, so main rejects the household (limit 1.99). This head admits it, as (d)(7) requires, so its subsidy rises from $0. The matching case just above 150% is fixed by the float64 band ratio above.

Source conflict at 200% (candidate us-ca:premium-subsidy-200-fpl-limit). The OE27 enroller deck disagrees with itself. Its summary slide (PDF p.38) says the subsidy "applies only to enrollees up to under 200% FPL". Its worked example of $31,920 at exactly 200% FPL (PDF pp.37, 39–40) applies "5.28% (6.78% - 1.5%)". This PR follows the adopted Design's (d)(7), and the Insight agrees with it. The candidate is in source_conflicts/candidates.yaml (follows: statute). The deciding line in fpl_limit.yaml documents the conflict, and that file is in KNOWN_UNMARKED until the entry is registered.

Why an allowance instead of a marker. A marker must name an entry in the pinned register that lists its file (test_every_marker_names_a_registered_entry_that_lists_its_file). Marking here would mean registering in rulespec-us, landing that PR and bumping the pin, all inside this PR. The register pins shipped code ("Pin only commits on main", CONTRIBUTING.md#source-conflicts), so the US hub keeps register entries and markers out of the PR that makes the ruling. #10301's 150% candidate takes the same path, and so did #10315's guardian-fee candidate, which merged on 2026-10-24 (1dc89f047f) with its own KNOWN_UNMARKED line:

1. this PR adds the KNOWN_UNMARKED line and a candidate naming this PR and its files;
2. after merge, a rulespec-us PR registers the entry against the squash commit;
3. a PE-US pin bump adds # Source conflict us-ca:premium-subsidy-200-fpl-limit at the deciding line.

The guard forces the cleanup. A file with a valid marker is skipped by the count. So once the marker lands, test_conflict_sounding_comments_do_not_grow fails (its shrunk assertion) until the KNOWN_UNMARKED line is deleted. test_candidates_are_not_yet_registered likewise fails until the candidate is deleted.

Federal 2027 table: now on main, from another PR. When this PR was written, main's federal applicable percentages carried the 2026 table (Rev. Proc. 2025-25) into 2027. #10443 merged on 2026-10-25 as 15b2bdd2cf and added the 2027 table from Rev. Proc. 2026-26 §3.01:

Federal band	2026	2027 (main since #10443)
Under 133%	2.10%	2.15%
133–150%	3.14% → 4.19%	3.23% → 4.3%
150–200%	4.19% → 6.60%	4.3% → 6.78%
200–250%	6.60% → 8.44%	6.78% → 8.66%
250–300%	8.44% → 9.96%	8.66% → 10.22%
300–400%	9.96%	10.22%

The state subsidy is max(0, slcsp − aca_ptc − ca_pct × income). Wherever it is positive, each dollar of federal credit is a dollar less of state subsidy.

• The OE27 example: at 6.78%, the federal credit is $6,091.82 and the state subsidy $478.80, the deck's figures. With the 2026 table carried into 2027 (6.60% at 200%), they were $6,149.28 and $421.34.
• Case 19 now derives the federal credit from main's 2027 table, and pins it as an output (32004fd222). Before Add the 2027 ACA premium tax credit applicable percentage table #10443 landed, Case 19 supplied the official credit as an input, and this body said that whichever PR landed second would make this change. The federal work is tracked in Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442.
• Effect on this PR's impact run: aca_ptc is the same in the base and head runs, so the federal table cancels out of the head-minus-base change wherever the subsidy is positive in both runs. It does not cancel where one run's subsidy is clipped to 0, or for households that newly qualify at exactly 200%. The 2027 rerun against main with the 2027 table (see Impact) gives the same change to the cent.

Invariants

All of these are about this PR's head. They assume the non-income conditions hold (in effect, an APTC-eligible member, a filed return, residence in California). r is max(MAGI, 0) / FPG(prior year), computed exactly on the stored float32 values.

1. Band = (d)(7) exactly. The tax unit is eligible if and only if 1 <= r <= L, where L is 1.65 in PY2026 and 2.0 in PY2027. Both bounds are inclusive.
   • Why float64 is exact here: two float32 values whose ratio is near an edge differ from it by at least about 1e-9 relative, far above float64 rounding. 1.0 and 2.0 are exact.
   • Tested by: a Hypothesis differential test against a fractions.Fraction reading of the Designs (bounds transcribed from the PDFs, not read from the parameters).
2. Interval in income. For a fixed FPG, the eligible incomes are [FPG, L·FPG]. This follows from 1.
3. Rate shape (from Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's property tests, now run for PY2027 too).
   • The rate is 0 at or below exactly 150%.
   • Just above 150%, it starts at the band's initial rate (3.19% in 2026, 2.8% in 2027).
   • It never decreases as exact income rises.
   • Its interpolation is continuous except at 150%. The continuity checks are at 100% and 165% (2026) and at 100% and 200% (2027).
   • Band selection now divides in float64, so it uses the exact ratio for every float32 input, by the same argument as invariant 1. Before this head, Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's float32 division could misband a fractional FPG within about 1e-7 of 150%. Whole-dollar FPGs were never affected: the US hub's brute force over every whole-dollar FPG up to $200k found no misbanding, and the search above found none up to $1,000,000.
   • Tested by: test_float32_ratio_just_above_150_uses_upper_band, which feeds the example above as float32 for both years. Restoring the float32 division fails it in both years.

Differences from main. These are descriptive, not invariants. With the published whole-dollar guidelines:

• PY2026: eligibility differs only in the (1.65, 1.66) sliver, plus main's float32 misrounding of exact multiples of 1%. Example: the 2025 FPG for 72 people is $406,150, and MAGI of $674,209 is exactly 166%. Main computes 165.99% and admits the household.
• PY2027: the rate changes across (150%, 200%] as tabulated above, and exactly 200% gains eligibility.

main's float32 fraction can also round a ratio just below a 1% multiple up onto it.

• Whole-dollar FPGs up to $1,000,000: the search above found no such rounding onto 100% or 200%. Onto 165% it happens from an FPG of $15,002, but this head reads the same truncated fraction, so the rate is unchanged, and both gates admit the household.
• Fractional FPGs: it can also happen at 100%, so the floor can move either way. In every such case, this PR is the one that matches (d)(7) (found by the earlier Sol review).

Tests

ca_premium_subsidy_eligible.yaml (34 cases). In the PY2026 boundary rows, the incomes are derived from wages, using the actual FPG.

Plan year	Income	% of FPL	Truncated	Eligible
2026	$15,649	99.994%	0.99	false
2026	$15,650	100.000%	1.00	true
2026	$25,822.50	165.000%	1.65	true
2026	$25,823	165.003%	1.65	false
2026	$25,978	165.994%	1.65	false
2026	$25,979	166.000%	1.66	false
2027	$26,335	165.006%	1.65	true
2027	$31,919	199.994%	1.99	true
2027	$31,920 (OE27 example)	200.000%	2.00	true
2027	$31,921	200.006%	2.00	false
2027	$32,079	200.996%	2.00	false

Supplied-ratio cases cover 199%, 199.5%, 200% (eligible) and 200.5% (ineligible) in PY2027. Supplied cases set aca_magi and prior-year tax_unit_fpg instead of the fraction, which the gate no longer reads.

ca_premium_subsidy_applicable_percentage.yaml (25 cases). The PY2027 cases move to the Final Design: 1.575 → 3.172%, 1.65 → 3.544%, 1.825 → 4.412%, 1.99 → 5.2304% and 2.00 → 5.28%. New income-derived cases: $23,942 (150.0125%) → 2.8%, and $31,920 (exactly 200%) → 5.28%. #10301's $23,940 (exactly 150%) → 0% is kept.

ca_premium_subsidy.yaml (19 cases).

• Case 16 (PY2027, 157.5%): $3,237.50 → $3,207.00.
• New Case 19: the OE27 worked example at exactly 200%. Benchmark $688 a month, federal required contribution 6.78%, state rate 5.28%. The model gives $478.80 a year ($39.90 a month), and the deck shows "≈ $40 / monthly". The federal credit is derived from main's 2027 federal table (6.78% at exactly 200%) and pinned as an output at $6,091.824, as explained under "Federal 2027 table" above.
• Cases 4 and 11: their labels say exactly 165% and exactly 100%. They use $25,822.50 and $15,650 against a pinned FPG of $15,650, following Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's FPG pinning.

Property tests:

• test_ca_premium_subsidy_eligible_properties.py (4 tests): the eligibility differential test above, with PY2027 read as inclusive. New explicit examples: $31,919, $31,920, $31,921 and $32,079 at $15,960, and $32,001 at $16,000. Each example builds a fresh simulation.
• Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's test_ca_premium_subsidy_applicable_percentage.py: now parametrized over 2026 and 2027, plus test_float32_ratio_just_above_150_uses_upper_band (12 tests). It calls the formula directly, without building a simulation. Its _rates helper now takes an FPG and a dtype, so the new test can feed float32 values.

Mutation checks (local):

Mutant	YAML cases that fail	Property test
Strict limit (<)	4, 4b, 18, 18b	fails, both years
Gate on the truncated fraction	5b, 5c, 18a, 18d, 18e	fails, both years
Float32 division in the gate (earlier run)	—	fails, PY2026 explicit examples
Float32 division in the rate's band ratio	—	test_float32_ratio_just_above_150_uses_upper_band fails, both years

Local runs at 93371bc7e8, one file per process (macOS). At 32004fd222, which adds the merge of main and the Case 19 change, CI runs these files; see the checks.

• The four premium-subsidy YAML files: 86 passed (19, 25, 34 and 8).
• The two household health-benefit files: 41 passed (22 and 19). The largest peak was 5.1 GB, for household_health_benefits.yaml.
• The property tests: 4 for eligibility and 12 for the rate, all passing.
• test_source_conflicts.py: 103 passed.
• test_programs_registry.py: 765 passed, 10 skipped.

Impact

A real microsimulation on the default dataset, run by the US hub one run at a time under its heavy-job lock. Sums are weighted.

• 2027: main 16a5e6cb6d, which has the federal 2027 table (Add the 2027 ACA premium tax credit applicable percentage table #10443), against this head 32004fd222.
• 2026: the earlier run, main at 6a92d6a983 against 1238297ebf. It was not rerun, for the reasons below.

Output	2026 change	2027 change	2026 records changed (weighted)	2027 records changed (weighted)
ca_premium_subsidy	−$0.90	−$21.87M	1 (0.002)	113 (389,626)
assigned_ca_premium_subsidy	−$0.90	−$21.87M	1 (0.002)	113 (389,626)
healthcare_benefit_value	−$0.90	−$21.87M	1 (0.002)	110 (389,626)
household_net_income_including_health_benefits	−$0.90	−$21.86M	1 (0.002)	110 (389,626)
ca_premium_subsidy_eligible	−0.002	0	1 (0.002)	0
state_income_tax, income_tax, household_net_income, household_benefits	0	0	0	0

• 2027: the state subsidy falls $21.87M, from $292.20M to $270.34M (−7.5%), for 113 tax units (389,626 weighted). None rises. The largest single fall is $152.52 a year. No tax unit in the dataset changes eligibility, so none that meets the other conditions sits at exactly 200%. The 113 tax units are in 110 households, which is why the household-level rows count 110.
• The federal 2027 table moved the levels, not the change. The run before Add the 2027 ACA premium tax credit applicable percentage table #10443 (main 6a92d6a983 against 1238297ebf) had the same change to the cent, −$21,865,720.65, in the same 113 tax units, on levels of $269.35M and $247.48M. aca_ptc is the same in the base and head runs, so this is what the cancellation under "Federal 2027 table" predicts when every changed tax unit has a positive subsidy in both runs.
• 2026 (not rerun): one tax unit, at a weight of 0.002, sits in the (165%, 166%) sliver. It loses eligibility and $442.32, which is −$0.90 weighted. This repeats the earlier run at d9929e416a. The figure carries to this head:
  • 93371bc7e8 changed only the dtype of the band ratio, which changes no record's output (below).
  • 05ff1686f9 merges main. Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301, now on main, changed nothing in either year in its own run. Add the 2027 ACA premium tax credit applicable percentage table #10443 adds 2027 values only. None of main's other changes since 6a92d6a983 reaches an input of the 2026 California subsidy. They change other states and Puerto Rico, labels and reform-only parameters, federal items that don't enter AGI (the car-loan interest and QBI deductions, the child and dependent care credit, and the NIIT's own MAGI; aca_magi is AGI plus Medicaid's MAGI modifications), and behavioral-response code that returns 0 in a current-law simulation. Fix Social Security secondary earnings threshold reform #10380's change has no current-law effect, per its own impact run.
  • 32004fd222 changes a 2027 test case only.

Why the 2026 run also covers 93371bc7e8. The delta from 1238297ebf changes the model in one place: the dtype of the band ratio in ca_premium_subsidy_applicable_percentage. The rest of the delta is a test and a YAML comment.

• Only the 150% edge matters. Band selection can change only across 150%: clipping puts both sides of 100% and of the top limit in the same band.
• No whole-dollar FPG is affected. The search above found that float32 and float64 choose the same band for every whole-dollar FPG up to $1,000,000.
• Every FPG in the dataset is a whole dollar. tax_unit_fpg is first_person + additional_person × (size − 1), with an integer size and the whole-dollar 2025 and 2026 CONTIGUOUS_US, AK and HI parameters.

So every record's output at 93371bc7e8 equals its output at 1238297ebf.

Comparison files are on the US hub host: fixes/10305-impact/ for 2027, and fixes/10305-impact/prev/ for 2026 and the earlier 2027 run.

Coordination

• Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301: merged into this branch earlier at cc1af12cd1. Every conflict then was in ca_premium_subsidy.yaml Cases 4 and 11. They were resolved to this branch's whole-dollar boundary incomes, with Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's FPG pinning. A duplicate ca_premium_subsidy_eligible input in Case 6 was dropped.
• main merged after Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 landed (05ff1686f9, with main at 16a5e6cb6d). Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's head had not moved from cc1af12cd1. Five files conflicted, the five this body predicted:
  • ca_premium_subsidy.yaml, ca_premium_subsidy_applicable_percentage.yaml, test_ca_premium_subsidy_applicable_percentage.py and ca_premium_subsidy_applicable_percentage.py: main's versions equal cc1af12cd1's exactly, so this branch's side already contains them. This branch's side was taken.
  • source_conflicts/candidates.yaml: one hunk, this branch's new us-ca:premium-subsidy-200-fpl-limit candidate against nothing on main. It was kept. main's other candidate edits, including Exclude county agency staff from Minnesota MSA guardian fees #10315's Minnesota entry, merged without conflict.
  • The result: against main, the merged tree differs only in this PR's files.
• Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's body and rulespec-us#1832: both say this PR carries the Final 2027 schedule. At this head it does.
• Other truncated-fraction gates: Audit: ACA-related gates that compare the truncated aca_magi_fraction with an income limit #10329 audits every ACA-related gate that compares the truncated aca_magi_fraction with a limit, with a PR for each. Its table (row 9) lists this PR and Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 as covering the California gate. Treat income exactly at 400% FPL as premium tax credit eligible #10312, for the federal 400% ceiling, is one of its PRs.
• Federal 2027 table: added to main by Add the 2027 ACA premium tax credit applicable percentage table #10443 and tracked in Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442, as described under "Federal 2027 table" above.

axiom: TheAxiomFoundation/rulespec-us#1832 queued: the (d)(7) eligibility extension, now updated for the Final 2027 Design. It carries:

• the verbatim (d)(4) and (d)(7) text for both years;
• the required outputs and a review_finding;
• companion cases that match this PR's YAML.

The encoder path is blocked. The Covered California designs are not yet in the pinned corpus release, and signed generation needs Max's approval.

🤖 Generated with Claude Code

[MaxGhenis]

Hand-off to the US + core hub, at head 93371bc.

• Review: a GPT-6.1 Sol shell-backed review approved this head with no P1 or P2 findings.
  • Its earlier findings: the stale federal 2027 table and the KNOWN_UNMARKED allowance are accepted as splits, and the fractional-FPG wording is resolved.
  • Its three P3s are fixed in the description: the search-grid wording, the Treat income exactly at 400% FPL as premium tax credit eligible #10312/Audit: ACA-related gates that compare the truncated aca_magi_fraction with an income limit #10329 attribution, and tracking the federal follow-up in Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442.
  • It re-ran the premium-subsidy YAML files, both property-test files and the source-conflict guard. It confirmed the float32 regression test fails in both years without the cast.
  • It ran its own superset search, which confirmed every search claim in the description.
  • It confirmed that the impact run at 1238297ebf carries over exactly to this head.
• Impact: a full-file run (imp-10305-final, base 6a92d6a983 against 1238297ebf). 2026 changes by −$0.90 weighted. In 2027, ca_premium_subsidy falls $21.87M on 389,626 weighted tax units, and none rises. See the Impact section.
• Locally at this head, one file per process:
  • the premium-subsidy YAML files: 86 passed;
  • the household health-benefit files: 41 passed;
  • property tests: 16 passed;
  • test_source_conflicts.py: 103 passed;
  • test_programs_registry.py: passed.
• CI: it started when the PR left draft and is pending now. No run has finished at this head yet.
• Merge order: Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 lands first. It is still at cc1af12cd1 with changes requested, and that head is an ancestor of this one.
  1. If Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's head moves, merge its final head here.
  2. After it lands, merge main.
  3. The description's "After Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 lands" recipe covers the five expected conflicts. Take this branch's side of all five, but keep Exclude county agency staff from Minnesota MSA guardian fees #10315's Minnesota hunk in candidates.yaml. The reviewer reproduced the resulting tree byte for byte from a simulated squash.
• Follow-ups owed after merge:
  • a rulespec-us PR registering us-ca:premium-subsidy-200-fpl-limit against the squash commit;
  • then a PolicyEngine-US pin bump that adds its marker in fpl_limit.yaml, deletes the KNOWN_UNMARKED line and deletes the candidate. The guard forces the last two.
  • Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301's 150% candidate follows the same path.
• Elsewhere:
  • Federal 2027 table: Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442, which has its own session. Whichever of that PR and this one lands second derives amount Case 19's federal credit.
  • Axiom parity: [pe-parity] Encode California Premium Subsidy adopted-design boundary at 150% FPL TheAxiomFoundation/rulespec-us#1832 stays queued on the signed encoder.

[MaxGhenis]

#10443 (the 2027 applicable percentage table, Rev. Proc. 2026-26) merged first, as 15b2bdd. So this PR lands second and should change amount Case 19 in ca_premium_subsidy.yaml before it merges:

• Drop the supplied aca_ptc: 6_091.824 and let the case derive it. Main now gives 6.78% at 200% FPL in 2027, so $8,256 − 0.0678 × $31,920 = $6,091.82.
• Expect ca_premium_subsidy of $478.80, as now.
• Remove the comment that says the federal credit is supplied.
• Put "Fixes Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442" in this PR's body. Add the 2027 ACA premium tax credit applicable percentage table #10443 deliberately left Add the 2027 premium tax credit applicable percentages (Rev. Proc. 2026-26) #10442 open for this case.

The derived federal value is already pinned on main. aca_ptc.yaml has a case for a single filer at $31,920 with an $8,256 benchmark that derives the FPL fraction from income and gives aca_ptc 6,091.82. That case is set in Texas, so Case 19 still checks the California path.

🤖 Generated with Claude Code

[MaxGhenis]

US + core hub merge audit, us#10305 at 32004fd2227ea6a1af95bfa7a824992c0c030d0f (squash): Decide CA Premium Subsidy eligibility on the exact FPL ratio and adopt the Final 2027 design.

• Review: GPT-6.1 Sol, APPROVE at 32004fd (Subfleet job 20261026-022145-hub-us10305-delta-r6-g, ~/reviews/us-hub/reviews-gen/us-10305/review-delta-r6.md): no P1 or P2. It re-ran the main merge 05ff168 exactly (264 main source-conflict candidates and 51 allowances unchanged), re-derived Case 19 ($478.80), and checked the impact JSON against the body. Its P3, a stale paragraph claiming a gain at exactly 150% FPL against main (Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 is already on main), was fixed in the body on 2026-10-26. Combination with main: main moved past the CI base 16a5e6c by Model the IRS local sales tax worksheet with official locality rates #9879 (d08051c) only, which overlaps in tests/code_health/test_source_conflicts.py; on the merged tree of main 035a157 and this head, that file passed 103 of 103 (2026-10-26, hub).
• 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: Real default-dataset microsimulation, 2027 (fixes/10305-impact/compare.2027.json; main 16a5e6c against 32004fd): ca_premium_subsidy -$21,865,720.65 in 113 tax units (389,626 weighted), none rising, largest fall $152.52; healthcare_benefit_value -$21,865,807.69 and household net income including health benefits -$21,864,932.24 (110 tax units); eligibility, taxes, net income and other benefits unchanged. 2026 is carried over from the earlier run, for the reasons the body gives.
• Order: Lands after Follow the adopted California Premium Subsidy design at the 150% FPL boundary #10301 (81056c4) and Add the 2027 ACA premium tax credit applicable percentage table #10443 (15b2bdd), both on main. Next: the register chip in FOLLOWUPS.md (us-ca:premium-subsidy-150-fpl-boundary and -200-fpl-limit in rulespec-us).
• Axiom: [pe-parity] Encode California Premium Subsidy adopted-design boundary at 150% FPL TheAxiomFoundation/rulespec-us#1832 queued: the (d)(7) eligibility extension, now updated for the Final 2027 Design.