Index IRS parameters to the 12-months-ending-August chained CPI-U, with CBO's tax-year projections

[MaxGhenis]

Draft / HOLD. The PR branch PolicyEngine:irs-index-tax-year-cpi is at 4c33655ae5 (the hub published the repair commits on 2026-10-08).

Local build: 4c33655ae5b3654f0624f02aed6d50f00908558c, against main e44d7c367327108858e06a22160b0ab5d7189c41, not pushed. CLI credentials/SSH agent are unavailable. Automatic approval rejected the body write because approval policy is never; no PR mutation occurred. Verified bundle/patch and correction remain local; merges preserve history.

Fixes #9105. Fixes #9608.

Index IRS parameters using twelve months ending August before the tax year, including ending August and excluding preceding August. Unobserved windows use distinct CBO tax-year projections. Preserve published IRS/calendar anchors; dependent/educator limits derive from statutory bases. Append verified July/August 2026 observations without refreshing historical chained vintages.

Separate VT/WI unchained indexes from federal chained CPI. WI uses statutory August bases, $10 half-up rounding, no-decrease floors, and the exact HOH crossover. Preserve published WI 2026 schedules and CA June CCPI/brackets/deductions/credits/renter ceilings. VT brackets remain preliminary; generated deduction/exemption amounts remain estimates. Preserve main's Louisiana/SSA/extensions, YCTC, role and NY CDCC changes.

Ordinary RI/UT/CA semantic tests retain original expected outputs with explicitly synthetic forecast inputs. Partner fixtures are untouched.

Fresh foreground single-file checks ran at 36a8a9705e86a8da9ce201a01f1d363854095365, Python 3.13.9/core 3.32.15. The complete model/data/test Git tree is identical here; the release merge changes changelogs/package version only. These are not new-head reruns:

• 76 Python passes: 35 state-rounding, 33 federal uprating, 8 Oregon Kids' Credit; one microsimulation test deselected and one existing Oregon warning.
• 137 YAML passes across fourteen files: IRS/WI/CA amounts and RI/CA/UT/MN semantic cases; all exits zero. Exact commands/results are saved locally.

HEAD-guarded exact-byte IO retains normal constructors/core processing without cached models/results; renter outputs matched between both IO configurations.

Fresh local-head partner inventory is INCOMPLETE/BLOCKED: 53/140 files, 209/597 cases, 4,689/8,006 assertions. Eight files contain 46 failed cases/95 outside-tolerance assertions; completed-file coverage/execution errors are zero. This includes 34 WIC assertions from appended seasonally adjusted CPI, distinct from IRS/CA effects. Published CA anchors change liability components; the $55,830 renter ceiling makes $55,218 eligible for $60. All 144 fixture hashes match main. Seven audit warnings concern one incomplete CSFP result, whose process may still be live; further runs await confirmation it has stopped. The remaining 87 files and full-precision evidence are saved locally. Missing results are not replaced by historical outputs. Complete inventory and the top-level three-question gate remain unresolved; no partner expectation edit is authorized.

Max HOLDs: missing-month completion/flat tails, post-CBO growth, future state proxies, CA proxy replacement/unpublished credits. Limitations: historical release vintages, VT inferred base/preliminary amounts, Oregon mixed NSA/seasonally adjusted CPI. No methodology ruling is recorded.

Exact-head independent source review APPROVE is mechanical only, with no new unresolved mechanical P1/P2; merge remains held. Aggregate impact is unmeasured; full suite/microsimulation/current green CI are absent. Reconcile #9601 CA expectations; serialize #9718 helper overlap. #9709/core #552 remain separate. Axiom parity remains incomplete: local IRS/state and RI drafts, no external issue/signed encoding.

axiom: 26 USC §1(f)(3)-(7); §63(c)(4)-(5); §62(d)(3) TheAxiomFoundation/rulespec-us#1400 queued | WI §71.05(22)(dp),(dt) TheAxiomFoundation/rulespec-us#1595 queued | WI §71.06(1r),(2)(k)/(L),(2e) TheAxiomFoundation/rulespec-us#1596 queued | 32 VSA §5811(21)(C)-(D) TheAxiomFoundation/rulespec-us#1597 queued | 32 VSA §5822(b)(2) TheAxiomFoundation/rulespec-us#1598 queued | CA RTC §17041(h) TheAxiomFoundation/rulespec-us#1599 queued | CA RTC §17073.5 TheAxiomFoundation/rulespec-us#1600 queued | CA RTC §17054 TheAxiomFoundation/rulespec-us#1601 queued | CA RTC §17053.5 TheAxiomFoundation/rulespec-us#1602 queued | WI §71.05(22)(dp) (published2026 HOH crossover only) encoded-correct (us-wi/policies/income_tax/2026_full_year_resident_core.yaml#standard_deduction_head_first_upper_boundary_plus_one)

Hub microsimulation impact (current head)

A real microsimulation on the default dataset, main at the merge base e44d7c3673 against this PR's head 4c33655ae5, for 2026 and 2027. The runs went one at a time under the shared heavy-job lock. Sums are weighted.

Output	2026 change	2027 change	2026 records changed (weighted)	2027 records changed (weighted)
state_income_tax	-738.94M	-921.05M	7,647 (20,156,114)	25,123 (48,387,502)
income_tax	+34.40M	-1.12bn	1,659 (4,844,508)	58,804 (118,346,325)
household_net_income	+719.82M	+2.22bn	10,650 (24,933,270)	49,963 (103,016,374)
household_benefits	+15.25M	+178.86M	4,407 (8,635,042)	6,502 (13,830,067)

Comparison files are in fixes/9621-impact/ on the hub host.