Proposal: G5 source-of-record switch (WHO → WPP) — methodology v0.7.0

Status: ACCEPTED — signed off by Ben 2026-09-04 00:02 CEST ("yes i sign off"), pre-P1 shortcut; posted through 2026-09-17 regardless. Implemented as methodology v0.7.0. Proposed 2026-09-03 by the build loop. Refs: SPEC G5, DECISIONS 6, RP Part VII (licensing gate), docs/METHODOLOGY_CHANGE_PROCESS.md.

What changes

The settlement series' mortality input switches from the WHO GHO global abridged life table (2019 vintage, 19 band anchors) to the UN WPP 2024 complete life table (2023 vintage, single ages 0–100), and the stock estimator runs at single-age resolution instead of band midpoints. Population stays WPP-derived (it already reconciles person-for-person: N = 8,091,734,933 on both paths).

Why

Licensing, not taste. WHO GHO's data policy is non-commercial — docs/LICENSING.md row 3, VERIFIED-RESTRICTED — which is why the commercial licensing gate (tly/licensing_gate.py) blocks the current path by construction and every print carries the research-series notice. WPP 2024 is CC BY 3.0 IGO (CLEARED 2026-08-17): share and adapt, any purpose, commercial included, attribution required. G5 has recorded WPP as the intended licensed source of record since DECISIONS; A-16's closure (2026-09-03) removed the last gate in front of it.

The level change, computed and decomposed

Dual-run on committed snapshots (2026-08-17 WPP files; v0-original WHO inputs), Decimal-34, structure year 2023 on both sides:

• Table effect: +1.1639B — WHO 2019 → WPP 2023. WPP's World e0 for 2023 is 73.1694 vs the WHO 2019 table's 73.1234: the 2023 table reflects post-COVID recovery beyond the 2019 level. (The previously recorded same-year gap — WPP 2019 e0 72.6093 vs WHO 73.1234, snapshot note 2026-08-17 — is a different-estimators discrepancy, recorded per RP Part VI r4; switching vintage 2019→2023 and source in one governed step is exactly what that note anticipated.)
• Resolution effect: −0.0648B — single-age exactness vs band-midpoint interpolation, comfortably inside the error budget's ±0.5% banding term.
• Net: +1.0991B = +0.3033% — inside the budget's ±1.87% symmetric measurement interval; the series level moves, its stated uncertainty already covered it.

What does NOT change

• Archived prints. First-print-settles: every archived epoch stands as printed. The new version applies from the first epoch after acceptance, forward only.
• The v0 golden. AC-1.2 pins the RESTORED original values on the frozen data/snapshots/v0-original/ inputs — a historical anchor, not a live-series constraint. It stays green untouched.
• The keyless rule (G6). WPP files are keyless public downloads, already snapshotted and hash-manifested.

Implementation (the one-commit bump, after acceptance)

1. VERSION_POLICY_REGISTRY["v0.7.0"]: adds policy source_of_record: "UN WPP 2024 (CC BY 3.0 IGO): complete life table + single-age population, single-age estimator; WHO GHO demoted to triangulation" — prior policies unchanged. 2. build_settlement_print computes S/Ē from the WPP path (tly.wpp + tly.stock, the machinery the reconciliation and P6 tests already exercise at full scale). 3. Changelog entry; dual-run table published; licensing-gate mode check flips for the WPP path (WHO path remains blocked commercially). 4. Attribution line (UN suggested form) added to prints' provenance.

Comment window (the decision this document requests)

The change process offers two regimes and TODAY sits exactly on their boundary: the pre-P1 shortcut ("Ben's explicit sign-off, recorded in the proposal — expires at the first public print") versus the 14-day public window ("announced wherever prints are published"). The site went live 2026-09-03 serving the research-series prints.

Recommendation: treat the research series as pre-P1 for process purposes (no settlement consumers exist; the site is hours old), take Ben's recorded sign-off, AND leave this proposal published on the site for the 14 days anyway — belt and braces; the effective epoch lands whenever the sign-off does.

Sign-off (Ben): RECORDED — "yes i sign off", 2026-09-04 00:02 CEST (session 01Qgyk1CmBzRsFSdjjU4Yhda).