[MaxGhenis]

G8-d3: ingest 26 Stats NZ Table 7 all-sector FTE earnings levels (ordinary
QEMQ.SBSZ9A and total QEMQ.SBSZ9Z), Jun 2023–Jun 2026, plus seven Treasury
PREFU Table 4.1 C17:I17 annual ordinary-time earnings percentage changes.

Publisher qualification: PREFU D6 is Forecast4, but C22:D22 footnote 4 says
all measures except real/nominal GDP are actuals. Preserve Forecast4 verbatim;
2025 and 2026 earnings are observations, 2027–2031 source_projection. This
source-backed qualification supersedes the task's preliminary 2026 forecast
classification. All seven economic assumptions use Treasury's model_output
measurement-basis precedent; no survey instrument is invented.

Invariants and checks:

• Publisher bytes unchanged: SHA-256, exact byte length, source URL and
  fetched_at checked against staged receipts in the independent audit and
  pinned by each package's artifact tests.
  Three manifests carry source licence/attribution and canonical raw/nz keys
  generated by this worktree's chronicle.artifacts.build_r2_key. Remote R2
  upload and verification are explicitly pending hub; no upload claimed.
• Every fact is one literal publisher cell: independent tests pin all 26 QES
  and all 7 PREFU exact cells/values; selectors have scale 1, no divisor,
  rounding or ranges. No computed growth rates, averages, factors, aging,
  reconciliation, imputation, target activation or model bindings.
• Each QES quarter/series appears once: 2025 vintage rows13–16 own Jun2023–
  Mar2024; 2026 rows13–21 own Jun2024–Jun2026. Tests separately compare the
  ten overlapping publisher cells for exact agreement and reject a duplicate
  Jun2024 from the older vintage. Full 16-sheet QES used ranges preserved;
  other sectors/overtime/change blocks/worksheets emit no facts.
• QES periods use existing quarter:YYYY-Qn convention and calendar quarter
  coverage; pay-week definition is recorded only in notes and attributed to
  the hub-supplied methodology. Treasury periods use existing opening-year
  fiscal_year:2024–2030, with original years-ending-June2025–2031 labels.
  Off-year tests prove no build-year restamping for either publisher.
• Every record set visibly declares provenance_class. QES survey_aggregate
  names Quarterly Employment Survey (QES); Treasury model_output follows
  merged Treasury/OBR economic-publication convention. Both publishers keep
  explicit assertion values. Dedicated repository-wide provenance surface
  node passed (1 test); final result is in .hub-report.txt.
• Every dimension/value has a label, retaining series, component, FTE,
  all-sector/actual adjustment scope and Treasury column status/footnotes3,4.
  Explicit canonical equality constraints match source filters; consumer
  contract/raw-facts validators and suite agent acceptance check this.
• Guard regressions independently pin all 42 QES guard coordinates and all 22
  Treasury metadata/year/status/footnote coordinates plus 7 numeric guards.
  Expected-red scratch probes: QES 45 failures (42 guard mutations,2 P→R
  column shifts,1 duplicate-vintage YAML); Treasury 28 guard failures plus
  one B17 row-label failure and7 row-shift exact-cell failures. No publisher
  workbook bytes were mutated; mutants used copied YAML/in-memory cells.
• Independent ledger-source-fidelity PASS and ledger-boundary PASS reviews
  read all 33 cells, periods, receipt/storage fields and final constraint maps.
• validate-package CLI passes for each package at year2023; three separate
  single-source build_bundle calls run build-suite and pass all 12 acceptance
  checks with lineage 1.0. Source-cell counts 17427, 17298, 872. Optional external
  Axiom validation is skipped (one standard suite acceptance warning each);
  bundle warnings and duplicate keys are zero. Single-source output trees
  deleted immediately; full bundle/full suite never run. Targeted pytest,
  lint and surface-node results are recorded with counts in .hub-report.txt.

Final verification: 111 distinct targeted cases pass: QES 64 (57 + 7
corrected test-only guard-normalization reruns), PREFU 43 (42 + 1 corrected
test-only validator-call rerun), NZ roster/CLI 3 and provenance surface 1.
The required node is
tests/test_chronicle_source_package.py::test_every_source_package_record_set_declares_provenance_class
Every pytest invocation used -p no:cacheprovider --basetemp=.hub-scratch/pytest
-o tmp_path_retention_policy=failed, with basetemp deleted after completion.
Ruff check and changed-test format checks pass; git diff --check passes.
Detailed commands, expected-red evidence and final counts are in the report
and .hub-scratch/{qes-test-results.txt,treasury/test-results.txt}.

Exact measured bundle delta from base aa288c3,
at build year2023 (separate single-source builds; no whole bundle):

• stats-nz-qes-average-weekly-earnings-june-2025:8 facts,1 package,1 table,
  publisher stats_nz, country:NZ8, person8, four quarter periods2 each.
• stats-nz-qes-average-weekly-earnings-june-2026:18 facts,1 package,1 table,
  publisher stats_nz, country:NZ18, person18, nine quarter periods2 each.
• treasury-prefu26-economic-assumptions:7 facts,1 package,1 table,
  publisher treasury, country:NZ7, government7, seven fiscal periods1 each.
• Combined:+33 facts,+3 packages,+3 tables; +0 publisher/geography/period/
  entity keys,+0 warnings,+0 aggregate/semantic duplicates,+0 skipped sources.
  by_source:stats_nz+26,treasury+7; by_entity:person+26,government+7;
  by_geography:country:NZ+33 (the sole key; already present).
• by_period exact increments (all keys already exist):
  quarter:2023-Q2: +2
  quarter:2023-Q3: +2
  quarter:2023-Q4: +2
  quarter:2024-Q1: +2
  quarter:2024-Q2: +2
  quarter:2024-Q3: +2
  quarter:2024-Q4: +2
  quarter:2025-Q1: +2
  quarter:2025-Q2: +2
  quarter:2025-Q3: +2
  quarter:2025-Q4: +2
  quarter:2026-Q1: +2
  quarter:2026-Q2: +2
  fiscal_year:2024: +1
  fiscal_year:2025: +1
  fiscal_year:2026: +1
  fiscal_year:2027: +1
  fiscal_year:2028: +1
  fiscal_year:2029: +1
  fiscal_year:2030: +1
• New table keys/counts:
  stats_nz:Quarterly Employment Survey June 2025 quarter, Table 7: Average weekly earnings for FTEs, by sector =8
  stats_nz:Quarterly Employment Survey June 2026 quarter, Table 7: Average weekly earnings for FTEs, by sector =18
  treasury:PREFU 2026 Table 4.1 - Key economic assumptions for fiscal forecasts =7
• Default pins are base plus this delta only: facts1579296,packages408,
  publishers56,tables350,geographies102613,periods523,entities12,warnings122;
  stats_nz2041,treasury146,person521345,government3569,country:NZ6164.
  Full structured evidence: .hub-scratch/bundle-delta.json. Hub can re-pin
  additively if parallel NZ packages land first. No commit/push/GitHub action
  performed in this lane; this file supplies the proposed commit message.

Rebased by the NZ hub onto chronicle main acbedb5 (which includes #379, G8-d4).
The bundle pins are main's values plus this package's measured delta:
fact_count 1,579,313; source packages 408; stats_nz 2,058; treasury 146;
country:NZ 6,165; person 521,362.

Hub landing record

• Raw objects on R2: the hub published all three workbooks with this worktree's Chronicle code and streamed each back; SHA-256 matches the manifest:

  • raw/nz/stats_nz/qes_average_weekly_earnings_june_2025/2025/25053dcb…/quarterly-employment-survey-june-2025-quarter.xlsx
  • raw/nz/stats_nz/qes_average_weekly_earnings_june_2026/2026/1af2e7e3…/quarterly-employment-survey-june-2026-quarter.xlsx
  • raw/nz/treasury/treasury-prefu26-economic-assumptions/2026/e2486375…/prefu26-charts-data.xlsx

  The committed manifests are unchanged by the publish.

• Rebase: the build ran on aa288c3. The hub rebased onto main acbedb5, which carries Stats NZ regional resident population at 30 June 2023 (NZ, G8-d4) #379 (G8-d4), and set the bundle pins to main's values plus this package's measured delta.

• Re-runs on the rebased branch: test_chronicle_nz_qes_earnings.py 64 passed; test_chronicle_nz_prefu_earnings.py 43 passed; the NZ curated-sources bundle node passed. CI runs the full-bundle node.

• Why: these facts are evidence for the NZ hub's declared earnings-ageing factors (Max d1599). Chronicle stores the facts only; Microcosm computes the factors (G8-r2).

Facts-only source packages; no published number. Needs Max: no.

🤖 Generated with Claude Code

[MaxGhenis]

Merge audit (NZ hub): head 860212a. Independent reviews, GPT-6.1 Sol via Subfleet:

• 20261022-232331: APPROVE at b11a990. All 33 values match publisher cells; footnote 4 and opening-year fiscal handling correct; bundle pins = main + delta; 107 mutation probes red.
• 20261022-233515: APPROVE at dfbb17f (docs-only R2-verified wording).
• 20261023-015438: APPROVE at 860212a (NZ count pins 67→70, treasury 4→5; 74 tests).

The R2 raw objects were published and verified by SHA-256 read-back on 2026-10-23. CI: gh pr checks exit 0 (bundle, checks, surface). Merging with a merge commit, as #379.