Cross-check Bauer–Schanzenbach Appendix B against the case-level bootstrap (FY2023)

Twice-offered to Diane Schanzenbach (accepted in spirit — "happy to provide feedback", 8/27). The calibration handshake between the two teams' machinery.

1. Extract the published per-state FY2023 bin probabilities and standard errors from Bauer–Schanzenbach 2026 (Hamilton Project, "The SNAP state cost-shift policy leaves the program's existence to chance", Appendix B + Table 27 of the FY2023 QC Annual Report they source). Hash the acquisition per the issuance-file convention (analysis/issuance_fy2024.json source block is the model).
2. Replicate their construction exactly: Normal(point, official SE), their bins including the 13.33% carveout treatment.
3. Run our case-level bootstrap on the FY2023 raw file (already in the audited cache — see analysis/coding_consistency.json years) at the FY2023 official threshold, anchored on FY2023 officials.
4. Compare per state: bin probabilities (total-variation distance), SD scale (their design SEs vs our i.i.d. bootstrap — expect ours smaller; the earlier national comparison was ~0.8 vs ~1.1pp), tail-shape divergence. Report agreement AND divergence honestly — the earlier bootstrap-vs-normal check at FY2024 found shape effects second-order (median TV 0.005, max 0.046), so scale will likely dominate.

Output: artifact + memo + locks per repo conventions. This is the first deliverable of the Diane feedback loop and should be shareable with her directly.

Start with the source block in analysis/issuance_fy2024.json and the FY2023 year references in analysis/coding_consistency.json, then locate the audited FY2023 raw file. Follow the stated bin, bootstrap, and comparison steps. Done means producing the artifact, memo, and locks required by repository conventions, with agreement and divergence reported per state.