Audit Louisiana's 2026 standard deduction behind two scored references

[MaxGhenis]

Draft: do not merge until Max rules on the cos decision named in the first comment. This PR changes no reference, exclusion, payload or published text.

What it audits

Two scored references in release dashboard-data-20260930 rest on a 2026 Louisiana standard deduction of $12,835. Both are state income tax before refundable credits:

• scenario_051: $820.35
• scenario_077: $305.14

No model of the 46 is exact on either (tolerance $1). The top five by exact match answer $830.40 and $315.18, using 2025's $12,500.

reference_audit/2026-10-05-louisiana/README.md sets out the sources, the options and the method.

• Statute. La. R.S. 47:294(B) multiplies the prior year's amount by the CPI-U increase "for the previous calendar year". December 2025 CPI-U was released 2026-01-13.

• Before the freeze. The Department of Revenue published $12,875 only as a provisional withholding and estimated-tax amount:

  • LAC 61:I.1501: emergency rule, Notice of Intent and Rule (LR 52:749);
  • RIB 26-005, R-1306 and R-1210;
  • the IT-540ESi (2026) worksheet.

  Each rule document says the return amount will differ. The Legislative Fiscal Office's note in the Notice of Intent reads the statute as December to December, about 2.68%, which gives $12,835.00. LDR's 2026-06-20 notice on the retirement exemption applies 2.7% under the same statutory words.

• After the freeze. RIB 26-019 (2026-09-28) sets the 2026 return amount at $12,838 ($12,500 × 1.027). That moves the two references by $0.09, so no exact-match result changes.

• Convention. $12,835 is policyengine-us's own explicit 2026 entry (Update Louisiana 2026 standard deduction policyengine-us#8411), computed from published CPI. It is neither a price-index projection nor an amount published after the freeze. The convention as written does not cover it, and the r18 projection screen could not see it.

Option	051	077	Models exact (051 / 077)
Keep, $12,835 (recommended)	820.35	305.14	0 / 0
LDR's official $12,838 (for comparison)	820.26	305.05	0 / 0
Pre-freeze provisional $12,875	819.15	303.94	0 / 1
2025's $12,500	830.40	315.19	21 / 19
Exclude	n/a	n/a	n/a

The recommendation is to keep both references, with documentation changes for the next release. The proposed card and paper sentence is in the README. That sentence also bears on Maryland's held 2026 return deduction (scenario_068), which is the decision's open part.

Contents

• reference_audit/2026-10-05-louisiana/:
  • the audit README;
  • the 1,984-output sweep under six candidate deductions (scripts/sweep_la_standard_deduction.py);
  • scoring of each option through policybench.cli analyze, the command the freeze runs. The unchanged copy reproduces the published payload's scoring (scripts/leaderboard_impact.py).
  • the alternative convention module fixes/latest_c_la_published_2026.py ($12,875, not recommended), its composition with latest_final, and a check that the composition reproduces the sweep on all 1,984 outputs;
  • a check that LDR's pre-freeze 2026 retirement exemption ($12,324) moves no output.
• tests/test_la_standard_deduction_audit.py checks the committed records without the engine:
  • the baseline reproduces every scored reference;
  • only the two outputs move;
  • the engine's tax equals 3% of Louisiana AGI less the deduction for every Louisiana household and candidate, and pairwise moves equal 3% of the deduction change;
  • the CLI's exact-match moves agree with an independent $1-tolerance count.

Invariants

• The baseline (policyengine-us 2.15.17 with latest_final) reproduces all 1,928 scored references to 1e-3.
• Each candidate changes only scenario_051 and scenario_077 state income tax.
• For every Louisiana household: tax = 3% × max(Louisiana AGI − deduction, 0), with the joint deduction at 200% of single.
• Between any two candidates where taxable income stays positive, the output moves by exactly −3% × Δdeduction × (1 or 2). Where taxable income stays zero, it does not move.
• The alternative module composed with latest_final agrees with the sweep's inline reform on all 1,984 outputs. This is a differential check between two implementations.

axiom: n/a: PolicyBench audit records and tests; no PolicyEngine policy change (the policyengine-us follow-up encoding RIB 26-019 carries its own axiom line)

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
policybench-site	Ready	Preview	Oct 5, 2026 2:19pm UTC

[MaxGhenis]

Max ruled 2026-10-05: keep both references at $12,835 and add the documentation sentence from the audit README in the next release (batched with d963, d974, d972 and d831).

[MaxGhenis]

Shipped in release dashboard-data-20261006 (PR #202, squash-merged as 9ce4ade8, now the latest data release and live on policybench.org). This PR's commits were folded into that release as a merge commit, so its content is on main; closing it rather than merging it a second time. The audit and the wording-only Louisiana case-note rewrites shipped. The proposed benchmark-card and paper sentence was held under d994. Max ruled on 2026-10-06 to keep the published-amounts convention as written, so that sentence is not adopted and Idaho 076, SNAP 008/038/109 and Maryland 068 keep their references. He also ruled to exclude scenario_051 and scenario_077 Louisiana state income tax. Those two exclusions ship in the next release (Claude Haiku 5.5 plus d1022 and d994), as reference_law_published_after_freeze records in reference_audit/2026-10-05-louisiana/proposed_exclusions.json.