[PavelMakarchuk]

Summary

Follow-up on diagnosis batch 3's open items. Three probes run on us-6.2.1; their scripts and outputs are in diagnosis/batch3/probes/. I also read the microcosm code at Build P's commit cae8640.

• F3, Iowa and Wisconsin SNAP → pe_gap, microcosm#647.
  • Eligibility is not the outlier: PE is 1.14–1.16× FNS, the median of the non-saturated States. Participants are the outlier.
  • BBCE-only units hold 40–47% of PE's participants. FNS Table A.2 puts only 21–26% of participants outside federal rules.
  • The State take-up fill draws one rate over the BBCE-expanded pool and calibrates to caseloads that include BBCE-only households, so 85–86% of federally eligible people carry the flag (FNS: 98–100%).
• A3, SSI couples → pe_gap, microcosm#1212.
  • SSI take-up is one draw per person at an age-band prior, with no couple dimension.
  • Couples participate at about the prior: 41.9%, against TRIM3's 22.7%.
  • The frame is 81% married at 65+ (#1212), so more recipients sit in couples.
• E2, Connecticut EITC → stays open, with the mechanism stated.
  • The claims hit the SOI target, so the gap is the eligible pool.
  • The flag-off third rests on 39 of 138 records.
  • The weight solve gives these records more weight than the flag-on ones (2,503 vs 2,030), whereas nationally they get about half (1,425 vs 2,633).
  • No microcosm issue covers this yet.

The release carries no survey anchors: spm_unit_snap and ssi_reported are zero for every record.

Classes are now 54 pe_gap, 2 concept_mismatch and 1 open (before: 50, 2 and 5). populations.json is regenerated for the five claims' diagnoses, and the lane note is updated.

Test plan

• Rebuild: no drift
• pytest tests/: all pass after the export regeneration (test_diagnosis_batch3 pins updated)
• bun test src

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated
policyengine-scorecard	Ready	Preview	Oct 15, 2026 6:33pm UTC