Skip to content

Two-part ACS source-transfer calibration (battery pkg3): 13 checks greened - #742

Merged
MaxGhenis merged 205 commits into
mainfrom
battery-pkg3-two-part
Aug 23, 2026
Merged

Two-part ACS source-transfer calibration (battery pkg3): 13 checks greened#742
MaxGhenis merged 205 commits into
mainfrom
battery-pkg3-two-part

Conversation

@MaxGhenis

Copy link
Copy Markdown
Contributor

Two-part source-transfer calibration (battery burndown pkg3)

Implements the scoped ACS source-transfer two-part calibration from the battery adjudication (16-check package), then hardens it through three 1% verify-build rounds:

  1. Scoped ACS QRF binding regression (an unassigned target had leaked into calibration).
  2. Two carrier-capacity reduction fixes found by receipt validation and the whole-package audit: weeks_unemployed exceeded candidate capacity by one ULP; child-support by 42 ULPs. The strict receipt validator is untouched — both fixes are in the generating reductions.

1% verify-build evidence (committed: experiments/battery_burndown_pkg3_evidence.md)

Physical battery failure lines vs baseline1pct, same code path and seed: 127 → 114. 13 checks fully greened (child support expense+received, disability benefits, prior-year self-employment, weeks_unemployed quantile, workers' compensation, energy subsidy, adult care, unemployment compensation quantile, ssi quantile); ssi and weeks_unemployed incidence moved toward the band but remain red; zero regressions. The build's manifest is honestly non-ready (battery still red overall) — no artifact is promoted by this PR.

Authored by sol lanes over three rounds; verify builds, failure-line diff, and review by Claude.

🤖 Generated with Claude Code

MaxGhenis and others added 28 commits August 21, 2026 06:15
Checkpoint-resumed 1% verify build after the two carrier-capacity
reduction fixes (weeks_unemployed 1 ULP, child-support 42 ULP;
validator untouched). Physical failure lines 127 -> 114 vs
baseline1pct: 13 checks fully greened, ssi and weeks_unemployed
incidence moved toward the band but remain red, nothing regressed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…spec

Resolutions, per the merge charter:
- specs/us_imputation_lineage.yaml: accept main's deletion; lineage now
  derives from the authored US F0 bundle.
- The post-transfer calibration policy declaration moves to the authored
  spec: a closed regime_gated_qrf_model exact variant carrying the
  post_transfer_calibration_policy_v1 payload
  (spec_engine/schema/imputation.schema.json), authored fields in
  us/spec/imputation.yaml with their generator in
  tools/us_bundle_generation/imputation.py, and a ninth
  post_transfer_calibration stacked-authority component at binding
  version 11 (spec_engine/stacked_authority_semantics.py, us/spec/battery.yaml).
- test_imputation_lineage_spec.py combines main's producer-registry test
  with this branch's spec-matches-code policy-identity test at the
  generated-bundle boundary.
- tools/build_us_multispine_pool.py keeps main's F0 adapter path and this
  branch's omission of empty target_regimes.
- Anti-rot chain walked for the +92 authored fields: field-count/mode/effect
  literals, the imputation_models usage claim (4 -> 96), the pointer
  inventory sha, regenerated docs/evidence/spec-engine/us-f0-coverage.json,
  the US spec_sha256 pin, the BE/UK bundle pins, and the loader semantic-hash
  golden vector (mover verified by single-edit bisection: seed-attested
  kernel modules, a legitimate identity movement).
- Root journals unioned and historicized.

Verified: 339 tests across the nine affected spec-engine/pin/lineage/
gate-battery/contract files pass in one process, exit 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sole conflict: the UK bundle spec_sha256 pin in
test_spec_engine_country_bundles.py. Both parents' values were computed on
trees that lack the other side's envelope movers (this branch's
stacked-authority version-11 binding vs main's #733 UK sources/runtime
retarget), so the union pin was recomputed fresh via load_bundle("uk"):
bb7110699a9cb7a346ecd478f55b7a1c57bdc0c3f3283705b8a9b51830207193.

Verified on the union tree:
- load_bundle spec_sha256: BE bf022118... and US d3de6760... hold their
  committed pins; only UK moved.
- tools/spec_engine_coverage.py --check: 41471/41471 configuration fields,
  40/40 inventory checks, no drift.
- 339 tests across the nine affected spec-engine/pin/lineage/gate-battery/
  contract files pass in one process, exit 0 (loader golden vector and
  main's re-cut gate-battery digests included).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ipts

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis merged commit 057dd95 into main Aug 23, 2026
4 checks passed
MaxGhenis added a commit that referenced this pull request Aug 23, 2026
Journal-only conflicts (FINAL_REPORT.md, PROGRESS.md, _LANE-NOTES.md) resolved
by keeping this lane's journals; no code conflicts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant