[MaxGhenis]

Lands after #1398, because both change the NZ golden (tests/golden/nz_country_spec.json). #1398 landed at 19:00:55Z on 2026-10-29, and 3142d49 merges main (3088868) with the golden regenerated.

What changes

Max ruled d1981 on 2026-10-26 (option A). The listed non-wage donor money columns age from calendar 2024 to tax year 2026-27 by d1599's AWE convention, factor 1.079073 (1.077090 low, 1.081052 high), disclosed as interim. A hold-fixed run (factor 1) sits beside it as a pre-declared sensitivity.

• The disclosures. The four index-family disclosures in nz/money_normalization.json (mixed_income, capital_income, pensions, owner_housing_costs) said "provisional pending Max's ruling on non-wage ageing (H3)". Each now says "Max's ruling d1981 (2026-10-26) adopts this as an interim convention and pre-declares a hold-fixed sensitivity (factor 1), which no profile here carries yet (G8-w3)." The wording also moves to active voice ("ages by").
• The test. test_nz_money_normalization.py pins the ruling's sentence, including the clause that no profile carries the sensitivity yet, and refuses "provisional".
• The changelog. The unreleased fragment changelog.d/nz-money-normalization.added.md called non-wage ageing provisional. It now says ruling d1981 adopts AWE growth as the interim index and pre-declares the hold-fixed sensitivity, which no profile carries yet.
• The golden. It moves only money_normalization.json's hash and the spec fingerprint.
• No factor, column or rate changes. reference_aging.json's donor_cy2024_to_ty2027_{low,central,high} already carry operational factors 1077090, 1079073 and 1081052 over 10**6, the ruling's values.

Why the hold-fixed run isn't wired here

money_normalization.validate_money_normalization accepts an identity per index family, with no profiles, or a factor id per declared profile. It refuses an identity for one profile. transport/money_kernels.py reads an identity's reason from the family entry. So this resource can't carry a hold-fixed profile beside the AWE profiles as it stands. The independent review found two routes that touch neither module:

• A sibling declaration whose four families are all identities, run by its own money.normalize@1 node. The validator accepts it. It needs graph wiring, which G8-w3 does anyway.
• A fourth aging_index profile mapped to a unit factor. It needs a generator change, gives a receipt that labels a calendar-2024 schedule as target 2027, and overloads the shared aging_index dimension: reference_profiles would refuse the new profile until the IRD wage-band ageing gains one too.

Either way, the wiring belongs to G8-w3, which first runs NZ transport. The resource's status already says no NZ transport graph reads it yet. The disclosures say the sensitivity isn't carried yet, so they don't imply it is.

Verification

• With CI's import path (every packages/*/src plus the repo root), the NZ hub ran 12 engine-free files: test_nz_money_normalization, test_money_normalization, test_country_spec, test_nz_spec_package, test_spec_only_country_packages, test_spec_source_citations, test_transport_money_kernels, test_reference_profiles, test_calibration_columns, test_spec_engine_country_bundles, test_nz_reference_aging and test_reference_factors. 1,995 passed, 16 skipped, 0 failed.
• Mutant. With main's disclosures restored under the new test, test_every_family_uses_the_donor_factor_for_its_profile_with_a_disclosure fails. With this PR's disclosures, all 17 tests in the file pass.
• ruff check . and tools/ci_test_plan.py verify pass.
• No other file pins the old string, the old resource hash or the old fingerprint (git grep).

Review

• Independent review of e9c18d4 (Opus 5.5 in session, since every Codex lane was held; jobs/review-microcosm-1425): APPROVE_WITH_NITS. All five claims held.
  • It regenerated the golden on its own and got the same bytes. With main's resource, the test failed. git grep found no stale pins. Six files passed, 473 tests in all.
  • It checked each family's columns against its text and against the ruling's list, and it confirmed a hold_fixed selection raises an unknown-profile error.
  • Its nits were the "provisional" changelog fragment, the unpinned clause, ruff format and the overstated "needs a schema change". cb0619c fixes all four.
• Delta review of cb0619c (same reviewer): APPROVE_WITH_NITS. All ten in-memory mutants of the disclosures fail the test. Its two changelog nits (an over-long line; the missing "no profile carries it yet") are fixed in c031775.
• Merge with main after Reconcile regional Census family-type counts under d1932 #1398. Only the golden conflicted: Reconcile regional Census family-type counts under d1932 #1398 and this PR both move the NZ fingerprint. The hub regenerated it from the merged spec. Against main, it moves only the fingerprint and money_normalization.json's hash, 5e817f0b…, the same hash as before the merge.
  • The hub ran 16 engine-free files at the merge with CI's import path: the 12 above plus the 4 that Reconcile regional Census family-type counts under d1932 #1398 changed. 2,344 passed, 59 skipped, 0 failed.
  • Final delta review of the merge (same reviewer, at 1e226b2): APPROVE, no findings that need a change.
    • Only the four files of this PR differ from main, out of 3,638 paths.
    • Its own regeneration of the golden matches byte for byte.
    • It found no interaction with Reconcile regional Census family-type counts under d1932 #1398: no other code reads the disclosures, only the golden test reads the fingerprint, and this PR touches none of the files behind Reconcile regional Census family-type counts under d1932 #1398's re-pinned digests.
    • It ran 13 files: 2,086 passed, 43 skipped, 0 failed. Every skip names the unset NZ_CHRONICLE_ARTIFACT.
  • 3142d49 is 1e226b2 with only the merge commit's message rewritten, to add the trailer and drop git's leftover conflict list. Both commits carry tree 3ad28e014c10.
• CI at 3142d49: all ten checks pass in Tests run 40939280609 (engine-free, engine-uk and engine-us on 3.13 and 3.14, integration-uk, lint, select-countries, wheels).

Checklist

• Merge main after Reconcile regional Census family-type counts under d1932 #1398 and regenerate the golden.
• Independent review, at every head (see Review).
• CI green (see Review).

🤖 Generated with Claude Code

[MaxGhenis]

Merged by the NZ hub (session local_e407a527) at 3142d49 with --merge. Gates: gh pr checks exit 0; MERGEABLE; not draft; no CHANGES_REQUESTED review. Reviewed head: 3142d49. Independent review: in-session Opus 5.5 (Codex lanes held), APPROVE at e9c18d4 with deltas at cb0619c and 1e226b2, the last approving the main merge (tree 3ad28e01). The final head 3142d49 amends only the merge commit's message (same tree). Ruling d1981 (Max, 2026-10-26) supplies the disclosed convention.