[MaxGhenis]

Stacked on #387; merge after it. Belgian transport port (ruling d048), source package B2f: a Belgian wage distribution from the social security office (RSZ/ONSS).

What it adds

packages/rsz/employment_daily_wage_class_2023_2026 (alias rsz-employment-daily-wage-class-2023-2026, 34,629 facts), from RSZ "Arbeidsmarktanalyse: gedetailleerde kwartaalgegevens":

• T17: daily-wage class × age × sex, 2023Q1–2026Q1, 5,616 facts. 2023 comes from the 2023–2024 workbook, 2024 onward from the 2024–2026 workbook.
• T7: status × sex × NACE 2025 × daily-wage class, 2024Q1–2026Q1, 29,013 facts.
• Measures:
  • jobs (arbeidsplaatsen): a quarter-end stock, entity person, role employee_job;
  • workers deduplicated by INSZ, carrying their principal job: a quarter-end stock, role employee;
  • FTE (VTE): volume over the whole quarter, role full_time_equivalent_employee.
  • Each measure has its own record set and coverage; table spec ids and record-id prefixes are shared.
• Wage classes: published as <100, 100 - 149, 150 - 199, 200 - 299, >= 300 and Onbekend-Inconnu, in EUR per day. Numeric endpoints are not encoded as constraints, because their inclusivity is not published.
• Daily wage (RSZ common criteria page): normal gross pay ÷ paid days for full-time workers, and hourly pay × 7.6 for part-time workers. It excludes double holiday pay, premiums, the 13th month and severance. It ranks pay rates, not annual earnings.

packages/rsz/annual_wages_key_figures_2020_2025 (alias rsz-annual-wages-key-figures-2020-2025, 252 facts):

• the "Kerncijfers-Chiffres Clés" sheet of "Lonen en bezoldigde prestaties", the October 2026 pre-publication release;
• worker type × sex × six measures, 2020–2025, in the published units (thousands of EUR, days, thousands of hours), plus its 36 typed total cells.

Licence: CC BY 4.0, from the captured RSZ open-data page. Attribution is in each manifest and README.

Precision, overlap and totals

• Every FTE and annual cell displays one decimal (#,##0.0); jobs and workers are integers. Values are the cached value rounded half-up at the displayed precision. The cached value and number format stay in each locator.
• 2024 overlap between the two quarterly workbooks: 1,728 T17 cells agree at two decimals. 285 FTE caches differ by float noise; 15 cross a half-tenth when rounded to one decimal. All are listed in workbook_audit.json, and the newer workbook supplies every 2024 fact.
• Quarterly totals: every Totaal row is a SUBTOTAL formula and is excluded. Tests check that the cached details sum to the cached subtotals. Rounded facts need not sum to a rounded subtotal.
• Annual totals: the 36 annual total cells are typed numbers, not formulas, so they are kept as published totals.

Invariants

• Fidelity: one fact per selected published numeric cell. Each value equals the cell's cached value at its displayed precision.
• No constructed totals: the facts contain no derived totals.
• Determinism: both generators reproduce their CSV, YAML and manifests byte for byte.
• Lineage: coverage is 1.0, with 0 acceptance errors.

Checks

Check	Result
validate-package, both packages	valid, 0 errors, 0 warnings
build-suite daily --year 2026	valid; 34,629 facts, 132,900 constraints, 0 acceptance errors, lineage 1.0
build-suite annual --year 2025	valid; 252 facts, 684 constraints, 0 acceptance errors, lineage 1.0
Spot check: 40 random daily rows (120 values plus printed dimensions) re-read from the workbooks	0 mismatches
ruff check / ruff format --check	pass
R2 ledger-raw, all 15 manifest files (daily CSV now b4f2f134…)	uploaded, SHA-256 readback verified 2026-10-24
Focused tests, annual (tests/test_source_rsz_annual_wages_key_figures.py)	7/7 pass when called directly at 7e447cb; the pytest run waits on the host reservation, and CI's Chronicle surface job runs the file
Focused tests, daily (tests/test_source_rsz_employment_daily_wage_class.py)	5/5 pass when called directly at 23c0bd5; the pytest run waits on the host reservation, and CI's Chronicle surface job runs the file
Bundle pins in tests/test_chronicle_bundle.py (099a4c1, on top of #387's)	+34,881 person facts at country:BE; two sources, two source tables and two built packages; quarter 2023-Q1 to 2026-Q1 and calendar years 2020 to 2025 incremented; no new geography, period or entity key; no warning. Measured with Chronicle's bundle functions and confirmed against #388's CI bundle on the old base. The full bundle test, replayed against #385's CI bundle plus #386's skipped packages, #387's package and these two, passes; #387's test fails against it. CI runs the full bundle test on this head
Independent review (GPT-6.1 Sol)	9da0e3b: REQUEST CHANGES for one broken annual test (consumer-contract validator given exported dicts), fixed in 7e447cb. Its independent full re-read of all 34,881 facts found 0 value, precision, dimension or locator mismatches, and 0 omissions or extras. It confirmed coverage, definitions, units, overlap and subtotal claims. 7e447cb: APPROVE, no findings. 23c0bd5 merges the T7 sex case variants (Man-homme/Vrouw-femme through 2025Q3) into one value per sex, as for Onbekend-Inconnu, so T7 series no longer split at 2025Q4. Re-review of 23c0bd5 pending

Source-only coverage:

• 34,881 facts in total;
• 0 new period keys: every quarter:2023-Q1…2026-Q1 and calendar_year:2020…2025 key already exists;
• 0 new geography keys: country:BE;
• no label conflicts.

Stacking and rebase

Rebased from 200bf6d onto #387's head 9323e9e, which is on main ba2509a. git range-diff shows all three commits unchanged. Every file is byte-identical to 23c0bd5 except the alias registry, which differs by exactly this PR's two RSZ aliases.

This PR and #389 are stacked so their CI runs on the trees that will land. They merge as merge commits in order: #387, then #388, then #389. Until #387 merges, this PR's diff on GitHub also shows #387's commits.

Shared code

Only the two SOURCE_PACKAGE_ALIASES registration lines in chronicle/source_package.py.

🤖 Generated with Claude Code

[MaxGhenis]

Delta review at 099a4c1 (GPT-6.1 Sol, Subfleet run 20261024-042210-chronicle388-review-r4): APPROVE, no findings. ledger-source-fidelity: PASS. ledger-boundary: PASS.

• Rebase onto Add SPF Finances municipal PIT surcharge rates, income year 2025 #387's head 9323e9e: git range-diff shows all three commits unchanged. All 30 non-registry files are byte-identical to the approved 23c0bd5, and the registry diff is exactly the two RSZ aliases.
• Bundle pins (099a4c1): derived independently at bundle year 2023, and every changed pin matches. +34,881 person facts, country:BE +34,881, two sources, two tables and two packages; quarter 2023-Q1 to 2026-Q1 and calendar years 2020 to 2025 incremented as pinned; no change in geography, period or entity cardinality. Zero warnings of any kind: all 19 packages emitting country:BE name it "Belgium", and the RSZ dimension ids and labels conflict with no other package.
• Earlier reviews: the first review REQUEST CHANGES, fixed in 7e447cb; r2 APPROVE; r3 at 23c0bd5 APPROVE.

This merges as a merge commit after #387, at 099a4c1, once CI passes.