CBO cost estimates: stage publication 61367's period-window claims

[PavelMakarchuk]

Summary

Stages CBO publication 61367 (P.L. 119-21 vs CBO's January 2025 baseline) as period-window claims in lane cbo-cost-estimates. The lane note said these PDFs were "pointers pending period-window staging"; window claims already work for JCT, TPC and PWBM, so this does the staging.

New adapter sources/cbo-pl119-21/ (raw files from Wayback captures; the two PDFs match the harvest manifest hashes):

• Distribution (61367-data.xlsx, the letter's Figures 1–2 data): 10 deciles × 4 channels + net effect × {2025 dollars per household, percent of baseline income} = 100 claims, annual averages over 2026–2034.
• SNAP supplement (61367-SNAP.pdf): 15 claims for sections 10102–10105, each read from its verbatim sentence (for example, −2.4 million participants in an average month over 2025–2034; −$41 billion direct spending over 2028–2034).

QC (any failure raises):

1. Workbook titles, units, headers and decile labels are verbatim.
2. In each decile, the four channels add to the net effect within rounding.
3. The letter's quoted deciles (lowest −$1,200 / −3.1%, 5th, 6th, highest +$13,600 / 2.7%) equal the workbook cells.
4. Section 10102's four groups add to its 2.4 million total.
5. Every SNAP statement occurs exactly once in the PDF text.

Modelling choices

• Baseline. CBO published after enactment, so its counterfactual is not the law in force. A new registry entry, cbo_january_2025_baseline, records this. Equivalence with pre_obbba_current_law / pre_obbba_law is not assumed. The harvest's TFP-cap row (2034, $213) moves to this baseline too, so its claim id changes. Nothing referenced the old id.
• Reforms. The distribution uses obbba_enacted_pl119_21, the same slug as PWBM's enacted-law rows. Each SNAP section is its own world ("as if implemented on its own"): pl119_21_snap_{work_requirements, sua_energy_assistance, internet_expenses, state_matching_funds}.
• Metrics. Two change siblings are added beside their levels: participant_count_change and average_monthly_benefit_change. The "reduce or eliminate benefits for 300,000 people" statement is an affected_count, not a participant change.
• Percent rows use value_kind: "percent", so the app shows −3.1%, not −310%.
• Not staged, with reasons in source.json: 61537 and 61536 (whole-bill deficit totals and a draft Senate text), 61505 and the Arrington–Guthrie letter (House text, not enacted), HSGAC (immigration fees), the letter's whole-population channel totals, and the uninsured workbook.
• No PE counterparts yet. They need the January-2025 world run over each window. The lane stays computed only through the existing registry row.

Test plan

• Two builds: same content hash (175dd415…); no drift after the build and the suite
• pytest tests/ — 1782 passed, 15 skipped (test_modal_uk_event deselected locally: it needs Python 3.12)
• New tests/test_cbo_pl119_21.py: the extract equals the adapter output; each QC check catches a broken cell, quote or sum; windows, baseline and values are checked
• ruff format --check; 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