From 6abd9cc87168d237f8cae73ceec12865d0c65bd1 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Sat, 3 Oct 2026 22:27:38 -0400 Subject: [PATCH 1/9] Correct retired SNAP QC claims in the paper, README, FACTS and simulator FSBEN: Mathematica computes it for USDA from each edited case record with the QC Minimodel's benefit formula (tech doc, August 2026 posting). The paper and simulator no longer call it "the agency's own computational canon", "the agency's own benefit-calculation software" or "a closed model". The Step 12-13 reconciliation is described as the tech doc gives it; in Colorado FSBEN ends within $5 of BENFIX in 797 of 856 cases. Replay: the solver starts from the file's edited inputs and moves only the ELEMENT1 input in $3 steps. A match (246 of 283) is consistent with correct arithmetic on a wrong input; 230 moved an input and 16 moved nothing. A miss does not establish a computation error: 17 of the 37 moved nothing, and of 26 cases with a computational finding 13 reproduce. The "computation-side upper bound" reading and the disclosure-perturbation description are withdrawn, matching ANALYSIS.md (#92) and cause_shares.json (#94). The superseded "33 of 246" is reconciled in FACTS H8. Colorado's 0.03-point FY 2024 margin now carries its FY 2025 rate, 10.09%. FACTS C4 (tech-doc errata) is withdrawn: no source records the errata, so the abstract, introduction and oracle section no longer claim them. Re-rendered app/public/paper/web (revision 10) and bumped the wrapper's iframe version. tests/test_retired_claims.py ties every corrected figure to claims_audit.json, cause_shares.json and the replay rows and keeps the retired wording out of the living files. Co-Authored-By: Claude Opus 5.5 --- README.md | 5 +- app/public/index.html | 2 +- app/public/paper/index.html | 21 +- app/public/paper/web/index.html | 65 +++--- app/public/paper/web/index.pdf | Bin 209689 -> 211731 bytes paper/FACTS.md | 18 +- paper/index.qmd | 182 +++++++++------ tests/test_retired_claims.py | 402 ++++++++++++++++++++++++++++++++ 8 files changed, 573 insertions(+), 122 deletions(-) create mode 100644 tests/test_retired_claims.py diff --git a/README.md b/README.md index 4e81c6b..4f9b407 100644 --- a/README.md +++ b/README.md @@ -28,8 +28,9 @@ noise alone materially affects tier assignment. ## What v0.1 finds (FY 2024 file, all 53 jurisdictions) - **Tier assignment is noisy.** Several states' tiers are near coin flips - under QC sampling variation (Colorado: official rate 9.97%, 0.03 points - from the 15% boundary, with a ±0.9-point sampling SD). + under QC sampling variation (Colorado: official FY 2024 rate 9.97%, 0.03 + points below the 15% boundary, with a ±0.9-point sampling SD; its FY 2025 + rate, 10.09%, is above the boundary). - **Simplification options carry large expected values** where they can move a state across a boundary: about $609M/yr in combined expected state cost-share reduction nationally at 50% category-suppression effectiveness diff --git a/app/public/index.html b/app/public/index.html index f4720f9..535225f 100644 --- a/app/public/index.html +++ b/app/public/index.html @@ -130,7 +130,7 @@

Targeted review scenarios

diff --git a/app/public/paper/index.html b/app/public/paper/index.html index 8fc0947..12526af 100644 --- a/app/public/paper/index.html +++ b/app/public/paper/index.html @@ -76,23 +76,28 @@

Administrative quality control data as an oracle

failures, and the fiscal 2025 rates landing where the sampling-noise mechanism says realized years should: 18 of 53 jurisdictions changed cost-share tiers, only 7 of those changes beyond two-year sampling - noise. This page embeds the revision-7 manuscript (updated 2026-08-13), which - carries five rounds of adversarial review, archived unedited, and adds - the engine-adoption accounting bounds, the official-vs-file wedge - decomposition, and the pre-registered model-repair experiments. + noise. This page embeds revision 10 (2026-10-03), which corrects how + revision 9 described the QC file's computed benefit and the Colorado + error-case replay, adds Colorado's fiscal 2025 rate beside its fiscal + 2024 margin, and withdraws a claim of errata in the federal technical + documentation that no source records; the fact catalog lists each + change. The manuscript carries five rounds of adversarial review, + archived unedited, and adds the engine-adoption accounting bounds, the + official-vs-file wedge decomposition, and the pre-registered + model-repair experiments.

- Revision 9 · 2026-08-16 · Max Ghenis, PolicyEngine + Revision 10 · 2026-10-03 · Max Ghenis, PolicyEngine
-