Certify Build P US Populace as the certified default - #489
Merged
Conversation
BUILD_P_RELEASE_ID populace-us-2024-buildp-sparse-rmloss100-cae8640- 20260728T011454Z replaces populace-us-2024-buildo-sparse-rmloss100- 22bd902-20260722T232627Z; policyengine-us stays 1.764.6 (built_with_model_package), so pyproject and household snapshots are untouched - the #470 common case. Build P vs the incumbent on the SAME current register (O-1 rescored via score_us_fiscal_targets): within-10% 95.76% vs 78.91%. The three fixes between them, each merged behind cross-family review: populace#561 decomposes PUF E19200 by published TY2015 SOI 2.1 AGI-band component shares in the base builder (the mortgage critical target moves from +20.4% gate-failing to -0.00%, and freed gradient pulls medical-dental from +17.3% to +10.3%); the floor-aware SSI take-up prior with the attempt-6 delivered-weight basis lands the aged bands (18-64 +0.03%, 65+ +3.03%); populace#564 registry-excludes the four SOI 1.4 other-income rows whose mapped variable (E01200) is Form 4797 line-14 business-property gain/loss, not the line-21 concept - a mis-specification the previous defaults fit only by crushing truthful donor mass (full adjudication on populace#507). Known boundaries: SSI under-18 band inherits the known -82% class (same boundary the Build O verdict carried); medical-dental +10.3% inside all gates; 5,632 compiled targets after the #564 family drop plus per-run support registers (39 zero-support cells, 13 QRF-tail columns, every entry carrying its measured adjudication). TROs regenerated with credentials (the UK sidecar is the legitimate 4-line re-bind, not a degraded limited TRO); ACS local-area overlay preserved at buildo-acs-local-77e2061; 129 cert-suite tests green against the published tag; latest.json verified UNCHANGED (still the Build O default). Merging this PR is the default-flip decision and stays with the maintainer. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
MaxGhenis
marked this pull request as ready for review
July 28, 2026 15:36
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.
DO NOT MERGE until you intend to flip the default — merging this PR is the default-flip decision (latest.json is separately verified unchanged; the flip itself follows the Build M/O runbook on merge).
What this certifies
populace-us-2024-buildp-sparse-rmloss100-cae8640-20260728T011454Z(published immutable tonight, tag verified, root artifacts at the tag revision) replaces Build O-1 as the certified US default. policyengine-us stays 1.764.6 — the #470 common case, pyproject and snapshots untouched.Why it's strictly truer
Same-register comparison (O-1 rescored on the current truthful register): within-10% 95.76% vs 78.91%. The three merged fixes between them:
Known boundaries
SSI under-18 band inherits the known −82% class (same boundary the Build O verdict carried). 5,632 compiled targets after the #564 family drop; per-run registers carry 39 zero-support cells and 13 QRF-tail columns, every entry with its measured adjudication.
Cert mechanics
Certify script run against the published tag (manifest sha
dd949ba3…); TROs regenerated with credentials (UK sidecar = the legitimate 4-line re-bind, not a degraded limited TRO — the first tokenless run was caught and redone); ACS local-area overlay preserved atbuildo-acs-local-77e2061; 129 cert-suite tests green against the published tag.🤖 Generated with Claude Code