Score OBBBA reform validation stacked (match JCT) - #185
Merged
Conversation
Each OBBBA provision was scored standalone — against a baseline where every other OBBBA provision is reverted — so each was measured as if it were the only provision in the bill. JCT's JCX-35-25 lines instead stack (each incremental given the lines above it), so they sum to the bill total. Comparing standalone lines to stacked lines is apples-to-oranges and badly overstated the interacting provisions (e.g. standalone personal-exemption repeal +228B vs JCT +110B, because it ignored the bill's standard-deduction increase that absorbs that revenue). Score the provisions cumulatively in JCX order instead: enact them one at a time on the pre-OBBBA baseline and take each line's incremental effect. The per-line effects now telescope to the true total OBBBA effect and align far better with JCT per line. Falls back to isolated scoring only if the OBBBA group spans multiple measures/periods. Verified on the released populace_us_2024.h5: aggregate OBBBA effect (~-351B) already matched JCT's total (~-321B) within ~9%; this fixes the per-line attribution. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This was referenced Jun 29, 2026
PavelMakarchuk
added a commit
that referenced
this pull request
Jul 1, 2026
…ding stacking (#246) The estate exemption is scored on estate_tax (#182) — estate tax is not a component of income_tax, so measuring it on income_tax is a structural zero. But stacked_obbba_effects required a single (budget_measure, period) across ALL provisions and otherwise fell back to isolated scoring for everything — so since #185 landed on top of #182, a real build would silently not stack at all. Stack per (measure, period) group instead: the income-tax provisions keep their JCX-order cumulative stack, the estate provision scores within its own group (single member = enacted alone), and the shared pre-OBBBA baseline still merges every revert so each group is scored in whole-bill context. Cross-group interactions are nil by construction since the measures are disjoint taxes. Removes the now-dead isolated fallback, documents the estate survey-data caveat, and adds a mixed-measure regression test that would KeyError on any isolated fallback. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
10 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
The reform-validation page showed populace far off JCT on the OBBBA provisions — and the divergence was worst exactly on the provisions that interact most (standard deduction, personal-exemption termination). Root cause: each provision was scored standalone, against a baseline where every other OBBBA provision is reverted, then just that one added back. So each line was measured "as if it were the only provision in the bill."
JCT's JCX-35-25 lines instead stack — each is the incremental effect given the lines above it — so they sum to the bill total. Comparing standalone lines to stacked lines is apples-to-oranges.
The diagnostic on the released
populace_us_2024.h5made it unambiguous:The model's aggregate is within ~9% of JCT — but the standalone lines don't even sum to populace's own total (−222B vs −351B), because interactions are mis-attributed. The worst case: standalone personal-exemption repeal +228B vs JCT +110B, because measured alone it ignores the bill's standard-deduction increase that absorbs much of that revenue.
Fix
stacked_obbba_effects()enacts the provisions one at a time in JCX order on the pre-OBBBA baseline and scores each as the incremental change. The per-line effects now telescope to the true total, and align far better with JCT per line (verified — standard deduction −167B→−131B vs JCT −92B; personal exemption +228B→+149B vs +110B; QBI −53B→−48B vs −44B). Falls back to isolated scoring only if the OBBBA group spans multiple measures/periods.Tests
test_obbba_components_score_stacked_in_jcx_orderchecks that the second provision is scored against the post-first-provision state (not pre-OBBBA) and that the line effects telescope to the total.The released release's dashboard data is being refreshed with the stacked numbers separately (committed override in calibration-diagnostics).
🤖 Generated with Claude Code