Sim-paper section + app forecast panels: persistence, backtest, election-correct bills

The assembly vehicle. Diane Schanzenbach: "an excellent point for you all to make in a paper or tool" (8/27); Max to the group (8/26): "we'll be building this for release as a tool and/or paper."

Paper half — new section in paper/index.qmd (sim paper):

• The persistence decomposition (#83: σα²=1.91pp², σu²=2.44pp² ρ=0.72, sampling 0.54pp²; I²=0.97 cross-sectionally; the Fay–Herriot/Rao–Yu framing).
• The distributional backtest (#87) with the steel-manned comparison: naive single-SE forward read covers 48–64% of nominal 90%; fair-official static calibrated at ~1.56pp; persistence model same calibration ~20% sharper, best CRPS (0.878 vs 0.901).
• The election-correct bill translation (FY2029 median modal-bucket 0.84→0.59; national bill SDs).
• Cite Bauer–Schanzenbach 2026 (Hamilton Project) — including their point that pre-OBBBA practice sanctioned on the CI lower bound.
• Incorporate regime-robustness + cross-check results (issues filed alongside this one) when they land.
• Standard ship pattern: render → wrapper revision bump → referee round → PR → deploy → tag (see paper-migrations revisions for the pattern).

Tool half — app/public/ forward panels:

• FY2026 election odds + FY2027 band currently run sampling-only with the crude τ=1.07 toggle. Upgrade to the persistence forecast: Kalman-style conditioning on the FY2025 official as a noisy observation (design discussed 8/24-ish; sol's #83 review blocker 2 documented why the current exposure module deliberately does location-translation only — the upgrade does the conditioning properly with the official sampling SD as observation noise).
• Keep sampling-only as the disclosed-lower-bound toggle.
• Hash-pinned payload + test locks per interventions_data.json conventions; ASSET_V bump; deploy + live-verify + tag.

Also riding along: the Campbell's-law scoring harness for the pre-registered OBBBA boundary design (analysis/PREREGISTRATION_OBBBA_BOUNDARY.md) — build the scorer now so it runs the day FY2026 rates publish (June 2027); outcomes 5 (cause-mix) and 6 ($58 bunching) are the measurement-response discriminators promised to Diane in the 8/27 reply.

Migrations-paper program (separate but adjacent): venue pick pending = Max (Data & Policy recommended, memo in ~/.cache/axiom-oracles/snap-fy27/journal-targeting-memo.md); replication export ready via scripts/export_replication_package.py --ref <tag> --verify.

Start by reading paper/index.qmd, app/public/, analysis/PREREGISTRATION_OBBBA_BOUNDARY.md, and the related #83 and #87 issues. Trace the existing render, payload-lock, asset-version, and deployment checks before splitting the paper, forecast, and scoring-harness work. Done means the paper renders, tests and hash locks pass, the live panels are verified, and the release is tagged.