[MaxGhenis]

Belgian transport port (ruling d048), source package B2b: municipal personal-income-tax surcharge rates by commune.

What it adds

packages/spf_finances/municipal_tax_rate_ay2026 (alias spf-finances-municipal-tax-rate-ay2026): FPS Finance's five-page PDF "Municipal tax rates (%) for tax assessment year 2026".

• 565 facts, one per printed commune rate. Values are the printed percentages with decimal commas turned into points (7,5 → 7.5), no rounding. Zero is kept (Knokke-Heist 0,0).
• unit: percent, aggregation: rate, entity: government, provenance_class: administrative.
• Geography: the 565 Statbel NIS 2025 communes from statbel/population_structure_commune_2025.
  • Matching: 558 labels match a Statbel Dutch or French name exactly; 7 match after case folding, accent and punctuation removal. Both tiers are listed in the README and in commune_mapping_audit.csv.
  • Keys: all 565 geography keys already exist in the bundle (statbel-fiscal-income-2023-nis-2025), so the package adds no new geography keys.
  • Names: each fact carries SPF's printed name, as the harness requires of every publisher (chronicle#266). Nine differ from Statbel's: Bruxelles/Brussels, Waimes/Weismes and seven spelling or punctuation variants. The bundle reports these as conflicting_geography_name_across_packages warnings (chronicle/bundle.py). The audit keeps Statbel's name beside each.
  • Identity: the NIS code identifies the commune; the printed spelling is not a filter or constraint.
• Locators: PDF page, column block, and start, end and rate text lines for every cell. CSV guard cells pin each selector to its NIS code, publisher label and locator.

Period

The PDF's assessment year 2026 taxes income of 2025. The facts follow the existing Belgian tax packages (statbel/fiscal_income_distribution_2023, spf_finances/pit_2023, statbel/fiscal_income_2023_nis_2025):

• tax_year: 2025 (income year);
• shared constraints income_year == 2025 ("Publisher income year") and assessment_year == 2026 ("Publisher assessment year"), evidenced by matching CSV columns;
• legal vintage income_year_2025_assessment_year_2026;
• source_period_label keeps "for tax assessment year 2026" verbatim.

No 2027 assessment-year rates are inferred for 2026 income.

Invariants

• Bijection: 565 publisher labels map one-to-one onto the 565 NIS 2025 communes. An unresolved, ambiguous or duplicate label, or an incomplete match, stops the build.
• Fidelity: each fact's value equals its printed token. The tests re-parse the PDF with layout extraction, independently of the generator.
• Determinism: build_package.py regenerates the CSV, audit, manifest and source_package.yaml byte for byte.
• Lineage: every fact has source rows and cells; build-suite lineage coverage is 1.0 with 0 acceptance errors.
• Identity: respelling a commune never changes its fact keys. The test respells all 565 facts' names, filters and constraints as Statbel's name and casefolded, uppercased and accent-stripped forms. Moving a fact to another NIS code always changes its keys. The test fails on the 875dea6 shape. Hypothesis is not a Chronicle dependency, so this is an exhaustive check over a fixed set of respellings.

Checks

Check	Result
chronicle validate-package … --year 2025	valid, 0 warnings
chronicle build-suite … --year 2025	valid; 565 facts, 1,130 constraints (the two year constraints per fact), 0 acceptance errors, lineage 1.0
ruff check / ruff format --check on new files	pass
R2 ledger-raw: CSV 84fd7521…, PDF 051211d8…	uploaded, SHA-256 readback verified 2026-10-24
Focused tests (tests/test_chronicle_spf_finances_municipal_tax_rate.py)	all 11 test functions pass when called directly at 650620b (unchanged by the rebase); 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 (9323e9e)	+565 facts, one source, one source table and one built package; tax_year:2025 +565; government +565; commune:11002 9 → 10; nine commune geography-name warnings (warning_count 123 → 132); warning-id initials 2, 5, 6. Measured with Chronicle's bundle functions and confirmed against #387's CI bundle on the old base. The full bundle test, replayed against #385's CI bundle plus #386's skipped packages and this package, passes; main's unchanged test fails against it. CI runs the full bundle test on this head
Independent review (GPT-6.1 Sol)	875dea6: REQUEST CHANGES (geography names; spelling in fact identity), fixed in 863e8ec. It found 565/565 rates exact and all mappings correct. Re-review of 863e8ec: APPROVE, no findings. 650620b adds the identity test; a delta review will cover it with the bundle pins

Rebase

Rebased from 200bf6d onto main ba2509a, after #385 and #386. git range-diff shows all three commits unchanged, and the package, data, changelog and package test are byte-identical to 650620b. The bundle pins are the only new commit.

Shared code

Only the SOURCE_PACKAGE_ALIASES registration line in chronicle/source_package.py.

🤖 Generated with Claude Code

[MaxGhenis]

Delta review at 9323e9e (GPT-6.1 Sol, Subfleet run 20261024-041803-chronicle387-review-r3): APPROVE, no findings. ledger-source-fidelity: PASS. ledger-boundary: PASS.

• Rebase onto main ba2509a: git range-diff shows all three commits unchanged, and every package, data, changelog and test file has an identical blob to 650620b.
• 650620b's identity test: non-vacuous. Restoring 875dea6's printed-name filter and constraint makes it fail.
• Bundle pins (9323e9e): derived independently with Chronicle's bundle functions, and every changed pin matches. +565 facts, tax_year:2025 +565, government +565, commune:11002 9 → 10, nine commune geography-name warnings (123 → 132), and no geography-count change.
• Earlier reviews: 875dea6 REQUEST CHANGES, fixed in 863e8ec; 863e8ec APPROVE.

This merges as a merge commit once CI passes at 9323e9e. #388 and #389 are stacked on it and merge after it, each at its own tested head.