[MaxGhenis]

Implement d1886 Census household-size and tenure allocation through the shared
allocated_residual_by_ratio operation and one resolve_reference_profiles API.
Keep regional residuals local, preserve independent national diagnostics, add
MAR and small-household profiles plus V4, and prepare all 136 size indicators.

Verification below identifies every invariant and its checking tests. Final
execution counts and named-mutant evidence follow.
All test IDs below are relative to:
packages/microcosm-build/tests/engine_free/shared/

Design section 6 invariants and their checks:

1. Regional closure: both size profiles and allocated tenure close to each
   region's published Census household total, then that total times its ERP
   ratio, using fsum with rel_tol=1e-12 and abs_tol=1e-7. Retain fractional
   targets and each region's own computed residual.

   • test_nz_target_references.py::test_inventory_asserts_exact_regional_residual_table
   • test_nz_target_references.py::test_real_artifact_regional_closure_shares_and_sensitivity
   • test_nz_target_references.py::test_scaled_tenure_values_are_fractional_exact_bridges_and_national_diagnostics_are_separate
   • test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once

2. MAR shares: allocated cross-products preserve stated shares, including
   zero-valued leaves.

   • test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once
   • test_nz_target_references.py::test_real_artifact_regional_closure_shares_and_sensitivity

3. Sensitivity recipients: only stated sizes 1 and 2 receive additions, in
   their stated proportions; total size counts stay fixed. Tenure remains
   proportional across all three source atoms in both residual profiles.

   • test_ledger_allocated_residual.py::test_sensitivity_exact_recipients_and_increment_ratio
   • test_nz_target_references.py::test_residual_profiles_have_exact_recipient_rosters_and_preserve_other_targets
   • test_nz_target_references.py::test_real_artifact_regional_closure_shares_and_sensitivity

4. Sensitivity comparison: against MAR, sizes 3-8 lose their MAR additions;
   against stated counts under the same ERP scaling, they remain unchanged.

   • test_ledger_allocated_residual.py::test_sensitivity_exact_recipients_and_increment_ratio
   • test_nz_target_references.py::test_sensitivity_removes_mar_additions_from_sizes_three_to_eight

5. ERP commutation: A(qx,qT,E)=qA(x,T,E) in exact arithmetic; apply the paired
   growth ratio once. In floating point an independently scaled total can leave
   a rounding residual of either sign; a negative one still refuses, never
   clips (explicit examples at counts 1, 2 and q=0.55, 0.58). The
   exact-rational oracle also checks generated output-group partitions.

   • test_ledger_allocated_residual.py::test_exact_rational_oracle_for_partitions_recipients_groups_and_growth
   • test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once
   • test_nz_target_references.py::test_all_17_erp_ratios_match_the_reviewed_source_table
   • test_nz_target_references.py::test_real_artifact_regional_closure_shares_and_sensitivity

6. Grouping: allocating tenure atoms and then grouping source categories 1+3
   equals allocating the owned group directly; category 3 stays in the exact
   atomic partition. Facts reused in growth roles are counted once per role.

   • test_ledger_allocated_residual.py::test_grouping_preserves_the_third_atomic_member
   • test_ledger_allocated_residual.py::test_growth_roles_may_reuse_total_or_atom_without_double_counting
   • test_ledger_allocated_residual.py::test_exact_rational_oracle_for_partitions_recipients_groups_and_growth

7. Region separation: keep all 17 Census regions, preserve geography 99 and
   its explicit national-growth proxy, and make unrelated regions inert.
   Never substitute national residuals for local residuals or equate 99 to
   territorial authority 999. The regional size/tenure residual sums remain
   61,713/429; independent national diagnostics remain 61,725/426.

   • test_nz_target_references.py::test_inventory_has_exact_contract_counts_and_named_parking_reasons
   • test_nz_target_references.py::test_inventory_asserts_exact_regional_residual_table
   • test_nz_target_references.py::test_published_regional_totals_and_national_totals_remain_independent_diagnostics
   • test_nz_target_references.py::test_region_99_is_a_separate_national_growth_proxy_and_ta_999_has_no_erp
   • test_ledger_allocated_residual.py::test_explicit_proxy_preserves_output_geography
   • test_ledger_allocated_residual.py::test_prefilter_keeps_every_auxiliary_selector_and_unrelated_rows_are_inert

8. Source preservation: selectors contain no observations; source facts and
   raw published totals/leaf sums/discrepancies remain unchanged. Inventory
   receipts explicitly classify residuals and allocated outputs as computed.

   • test_ledger_allocated_residual.py::test_receipt_labels_keys_and_preserves_base_geography_and_raw_facts
   • test_nz_target_references.py::test_source_discrepancies_remain_unreconciled_and_no_totals_enter_the_fit
   • test_reference_profiles.py::test_observed_values_cannot_enter_effective_references
   • test_reference_profiles.py::test_unselected_profiles_cannot_carry_observed_values

9. Preparation: validate household/member identities and memberships, refuse
   duplicates, missing identities, orphans and empty households, and derive
   usual-resident household roster counts. The 136 region-by-size indicators
   partition each valid household exactly once; size 8 includes 9+ members.
   Prepared columns cannot overwrite the validated roster count.

   • test_calibration_columns.py::test_household_size_indicators_partition_all_regions_and_include_nine_people
   • test_calibration_columns.py::test_household_roster_refuses_invalid_membership
   • test_calibration_columns.py::test_prepared_columns_cannot_overwrite_a_validated_roster_count
   • test_calibration_columns.py::test_all_active_calibration_bindings_enter_the_small_frame_matrix

10. Profile identity: references, intervals, preparation and problem bindings
    use one selection identity and exact effective-document hashes. Preserve
    full raw-resource identity while projecting effective documents to the
    compiler's strict country/schema_version/hierarchy/target_references keys.
    Bind the prepared column inventory, values and identity axis as well.

    • test_reference_profiles.py::test_resolution_is_pure_and_binds_effective_documents
    • test_reference_profiles.py::test_effective_document_projects_resource_headers_and_binds_full_base
    • test_nz_target_references.py::test_full_nz_resource_resolves_directly_to_the_strict_compiler_document
    • test_transport_target_kernels.py::test_profile_compilation_checks_actual_reference_document_hash
    • test_transport_target_kernels.py::test_preparation_and_resolution_evidence_bind_the_problem
    • test_transport_target_kernels.py::test_problem_refuses_stale_interval_or_preparation_profile
    • test_transport_target_kernels.py::test_decode_resolved_surface_requires_the_resolution_receipt
    • test_transport_target_kernels.py::test_decode_surface_refuses_stale_resolution_evidence
    • test_transport_target_kernels.py::test_preparation_requires_its_effective_interval_companion
    • test_transport_composed_graph.py::test_preflight_and_composition_use_the_same_resolved_reference_view
    • test_transport_composed_graph.py::test_preflight_checks_the_selected_variant_reference_status

Additional section 6 properties and refusal contracts:

• Hypothesis compares nonnegative counts, recipient subsets, output-group
  partitions and valid growth ratios with an exact Fraction oracle. It covers
  zero leaves, zero residual, grouping, input reordering and unrelated-region
  facts, with generated invalid cases exercising both normal selector
  resolution and direct pre-resolved compilation.
  Checking IDs:
  test_ledger_allocated_residual.py::test_exact_rational_oracle_for_partitions_recipients_groups_and_growth
  test_ledger_allocated_residual.py::test_generated_invalid_partition_rosters_refuse

• Exact typed atomic membership goes beyond cardinality: reject missing,
  duplicate, substituted, string/float-coerced and out-of-partition members;
  reject total-in-partition, wrong universes, changed source snapshots,
  periods/vintages, negative/nonfinite facts and incomplete selectors.
  Checking IDs:
  test_ledger_allocated_residual.py::test_refuses_invalid_atomic_partitions_on_both_compilation_paths
  test_ledger_allocated_residual.py::test_declared_base_roster_must_match_its_count_and_partition
  test_ledger_allocated_residual.py::test_constructor_rejects_base_roster_members_outside_partition
  test_ledger_allocated_residual.py::test_partition_dimension_filters_cannot_change_the_total_universe
  test_ledger_allocated_residual.py::test_only_exact_typed_category_equalities_are_removed_from_universe
  test_ledger_allocated_residual.py::test_total_and_atoms_require_same_typed_shared_dimension_values
  test_ledger_allocated_residual.py::test_reference_contract_refuses_unpaired_or_malformed_operands
  test_ledger_allocated_residual.py::test_future_count_period_cannot_bypass_consumer_period_policy
  test_ledger_allocated_residual.py::test_generated_invalid_partition_rosters_refuse

• Negative residuals always refuse; no clipping and no signed reconciliation.
  Zero residual with zero recipient mass is an explicit identity branch;
  positive residual with zero recipient mass refuses.
  Checking IDs:
  test_ledger_allocated_residual.py::test_negative_residual_is_refused_without_clipping
  test_ledger_allocated_residual.py::test_zero_residual_with_zero_recipient_mass_is_identity
  test_ledger_allocated_residual.py::test_positive_residual_with_zero_recipient_mass_refuses

• Growth operands come as one complete pair with matching source series and
  periods, positive denominator and a ratio in (0.5,2]. Cross-geography growth
  requires an explicit proxy while the output keeps its base geography.
  Checking IDs:
  test_ledger_allocated_residual.py::test_growth_operand_refusals_on_both_paths
  test_ledger_allocated_residual.py::test_reference_contract_refuses_unpaired_or_malformed_operands
  test_ledger_allocated_residual.py::test_explicit_proxy_preserves_output_geography

• Raw size and tenure rows cannot compile unresolved or be promoted through
  activation_status alone. Only a transformation satisfying the declared
  prerequisite activates the row; unrelated prerequisites remain parked.
  Checking IDs:
  test_ledger_allocated_residual.py::test_raw_reference_profile_requirement_cannot_be_promoted_by_status
  test_nz_target_references.py::test_raw_tenure_and_size_cannot_bypass_profile_resolution
  test_nz_target_references.py::test_raw_tenure_refuses_even_when_all_real_operands_are_present
  test_reference_profiles.py::test_transformation_must_satisfy_its_declared_profile_prerequisite
  test_reference_profiles.py::test_unsatisfied_prerequisites_remain_parked_after_transformation

• One country-neutral resolver selects explicit reference-name lists, applies
  disjoint transformations deterministically, refuses unknown/unavailable
  selections, missing names, duplicate/overlapping transformations, unsupported
  factors/interval transformations, repeated resolution and stale evidence.
  Unselected-profile edits leave selected identities unchanged. The independent
  aging_index dimension is reserved: central is explicitly deferred; low/high
  remain unavailable until G8-r2 supplies the ageing transformations.
  Checking IDs:
  test_reference_profiles.py::test_unknown_or_unavailable_profiles_refuse
  test_reference_profiles.py::test_invalid_transformations_refuse
  test_reference_profiles.py::test_resolution_is_pure_and_binds_effective_documents
  test_reference_profiles.py::test_stale_effective_documents_and_receipts_refuse
  test_reference_profiles.py::test_unrelated_profiles_do_not_change_effective_identity
  test_reference_profiles.py::test_disjoint_transformations_are_deterministic
  test_nz_target_references.py::test_v4_changes_only_the_residual_knob_and_prior_variants_keep_their_meaning

• Shared arithmetic/profile tests use synthetic consumer declarations. Country
  packages remain spec-only, with no Python entrypoints; transport retains its
  country-name/runtime and float-literal guards. Country values live in spec
  JSON, and no country-specific solver or frozen kernel is changed.
  Checking IDs and inspection:
  test_transport_target_kernels.py::test_kernels_declare_honest_country_neutral_contracts
  test_transport_graph.py::test_transport_modules_import_no_country_runtime_and_name_no_country
  test_transport_graph.py::test_transport_modules_carry_no_float_literal
  test_transport_composed_graph.py::test_transport_python_has_no_policy_literal_or_remote_path
  test_spec_only_country_packages.py::test__given_country_packages__then_no_python_files_are_present
  test_spec_only_country_packages.py::test__given_country_specs__then_no_python_entrypoints_are_declared
  Static diff inspection covers the frozen-file exclusion and shared modules.

Parked-target and scenario scope:

• Raw calibration counts are 64 active, 170 requiring profile resolution and
  539 otherwise parked; effective MAR and small-household documents have 234
  active and 539 parked references. A read-only HEAD row-payload comparison
  retains the other 603 calibration rows, all 318 precal rows and all 6 holdout
  rows. Families, IRD, receipt assignments, NZ Super and VP remain parked.
  Checking IDs:
  test_nz_target_references.py::test_residual_profiles_have_exact_recipient_rosters_and_preserve_other_targets
  test_nz_target_references.py::test_every_parked_row_refuses_direct_compilation
  test_nz_target_references.py::test_grouped_tenure_family_multiplicities_and_parked_definitions
  test_nz_target_references.py::test_receipt_snapshots_are_split_and_stratified_inputs_stay_parked
  test_calibration_columns.py::test_family_multiplicity_is_a_measure_while_nucleus_activation_stays_blocked

• V1/V2 keep low/high ageing; V3 keeps beneficiary assignment; V4 changes only
  residual allocation to small_households. Gates and reporting include V4.
  Checking IDs:
  test_nz_target_references.py::test_v4_changes_only_the_residual_knob_and_prior_variants_keep_their_meaning
  test_nz_spec_package.py::TestScenarios::test_committed_grid_satisfies_the_scenario_contract

Feed, regeneration, disclosures and golden:

• Keep the Chronicle 8f4cfd573c0d6f580d9c51cd67e8f449a42ed174 feed pin. Recompute
  the design's exact regional residual table from its Census rows. QES/PREFU
  rows stay unused. Regenerate with tools/generate_nz_target_references.py and
  verify --check against NZ_CHRONICLE_ARTIFACT.
  Checking IDs:
  test_nz_target_references.py::test_generation_reproduces_committed_resources_with_exact_feed_identity
  test_nz_target_references.py::test_inventory_asserts_exact_regional_residual_table
  test_nz_target_references.py::test_changed_feed_identity_refuses_authoring_before_reference_compilation
  test_transport_target_kernels.py::test_target_kernels_enforce_full_feed_pin

• Disclose MAR, locally computed residuals, the national/regional differences,
  proportional tenure including family trusts, V4's recipients, fractional
  targets, one ERP bridge and the separate region-99 proxy. Preserve raw
  diagnostics and per-region/profile receipts with tolerance and provenance.
  Checking IDs and inspection:
  test_nz_target_references.py::test_source_discrepancies_remain_unreconciled_and_no_totals_enter_the_fit
  test_nz_target_references.py::test_published_regional_totals_and_national_totals_remain_independent_diagnostics
  Static inspection of generator disclosure, reference_inventory.json,
  docs/nz-calibration-targets.md and docs/nz-take-up-engine-plan.md.

• Regenerate nz_country_spec.json with
  canonical_json_bytes(_loaded_spec_summary("nz")). Only fingerprint,
  resource_hashes and resources move: eight existing resource hashes change,
  reference_profiles.json adds one hash and one resource-list entry, and the
  fingerprint becomes d970b6e900b7f5d455a86158718703444ac91c15d49b2203720fc2bdf2baba7d.
  Checking ID:
  test_country_spec.py::TestGoldenCountrySpecs::test_loaded_spec_matches_the_golden_file_byte_for_byte[nz]

Build-lane evidence below is lane-reported: its logs are kept in the NZ hub's
job archive, not in this checkout. The independent review section further
down re-ran the current behaviour separately.

Reviewer mutants: all 18 killed; 25 expected failed checks and 41 passing
controls across final mutant runs. Each mutation was restored. Actual red IDs:

• wrong_recipients_23: 1 failed, 1 passed in 25.50s
  test_nz_target_references.py::test_residual_profiles_have_exact_recipient_rosters_and_preserve_other_targets
• allocation_denominator_total: 2 failed, 1 passed in 1.42s
  test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once[False]
  test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once[True]
• small_denominator_all_leaves: 1 failed, 1 passed in 1.45s
  test_ledger_allocated_residual.py::test_sensitivity_exact_recipients_and_increment_ratio
• drop_tenure_trust_3: 1 failed, 1 passed in 1.05s
  test_ledger_allocated_residual.py::test_grouping_preserves_the_third_atomic_member
• national_residual_per_region: 1 failed, 1 passed in 6.03s
  test_nz_target_references.py::test_generation_reproduces_committed_resources_with_exact_feed_identity
• total_in_partition: 1 failed, 1 passed in 0.21s
  test_ledger_allocated_residual.py::test_reference_contract_refuses_unpaired_or_malformed_operands[total_in_partition]
• duplicate_replaces_missing_atom: 1 failed, 1 passed in 0.18s
  test_ledger_allocated_residual.py::test_refuses_invalid_atomic_partitions_on_both_compilation_paths[substituted-True]
• coerce_member_type: 1 failed, 1 passed in 0.75s
  test_ledger_allocated_residual.py::test_refuses_invalid_atomic_partitions_on_both_compilation_paths[float-True]
• clip_negative_residual: 2 failed, 1 passed in 0.76s
  test_ledger_allocated_residual.py::test_negative_residual_is_refused_without_clipping[False]
  test_ledger_allocated_residual.py::test_negative_residual_is_refused_without_clipping[True]
• zero_residual_divide_by_zero: 2 failed, 1 passed in 2.61s
  test_ledger_allocated_residual.py::test_zero_residual_with_zero_recipient_mass_is_identity[False]
  test_ledger_allocated_residual.py::test_zero_residual_with_zero_recipient_mass_is_identity[True]
• erp_applied_twice: 1 failed, 2 passed in 1.34s
  test_ledger_allocated_residual.py::test_mar_closes_preserves_shares_zero_leaves_and_applies_growth_once[True]
• drop_region_99: 1 failed, 1 passed in 0.86s
  test_nz_target_references.py::test_inventory_has_exact_contract_counts_and_named_parking_reasons
• region_99_relocated_to_NZ: 1 failed, 1 passed in 2.71s
  test_nz_target_references.py::test_residual_profiles_have_exact_recipient_rosters_and_preserve_other_targets
• prune_auxiliary_selectors: 1 failed, 1 passed in 1.38s
  test_ledger_allocated_residual.py::test_prefilter_keeps_every_auxiliary_selector_and_unrelated_rows_are_inert
• raw_tenure_bypasses_profile: 1 failed, 1 passed in 5.62s
  test_nz_target_references.py::test_raw_tenure_refuses_even_when_all_real_operands_are_present
• stale_interval_or_preparation_profile: 1 failed, 1 passed in 5.13s
  test_transport_target_kernels.py::test_problem_refuses_stale_interval_or_preparation_profile[selection]
• observed_value_in_reference: 5 failed, 23 passed in 1.18s
  test_reference_profiles.py::test_observed_values_cannot_enter_effective_references[value]
  test_reference_profiles.py::test_observed_values_cannot_enter_effective_references[values]
  test_reference_profiles.py::test_observed_values_cannot_enter_effective_references[observed]
  test_reference_profiles.py::test_observed_values_cannot_enter_effective_references[observed_value]
  test_reference_profiles.py::test_unselected_profiles_cannot_carry_observed_values
• size_8_means_exactly_8: 1 failed, 1 passed in 3.10s
  test_calibration_columns.py::test_household_size_indicators_partition_all_regions_and_include_nine_people

Final execution evidence: 2,470 passed across 16 targeted files, with five
optional UK feed/authentication checks skipped (CHRONICLE_UK_FACTS unavailable).

• test_calibration_columns.py: 39 passed
• test_country_spec.py: 136 passed
• test_ledger_allocated_residual.py: 89 passed
• test_ledger_grouped_roll_forward.py: 52 passed
• test_ledger_targets.py: 436 passed
• test_nz_spec_package.py: 1192 passed
• test_nz_target_references.py: 46 passed
• test_reference_profiles.py: 28 passed
• test_spec_engine_country_bundles.py: 21 passed
• test_spec_only_country_packages.py: 18 passed
• test_transport_composed_graph.py: 145 passed
• test_transport_gate_bindings.py: 126 passed
• test_transport_graph.py: 15 passed
• test_transport_target_kernels.py: 33 passed
• test_uk_ledger_targets.py: 64 passed
• test_uk_target_references.py: 30 passed, 5 skipped

Generator --check: exit 0 against Chronicle 8f4cfd573c0d6f580d9c51cd67e8f449a42ed174.
Ruff check: all checks passed on 18 changed Python files.
Ruff format --check: 18 files already formatted.
tools/ci_test_plan.py verify: verification=ok.
git diff --check: exit 0.
Final scope audit: 31 implementation files;
no frozen-source edits; zero float literals
in the 19 transport Python modules. Golden resource hashes all match.
HEAD row audit: exactly 136 size + 34 tenure rows changed; other 603
calibration rows, all 318 precal rows and all 6 holdout rows unchanged.

The exact legacy NZ resource-list assertion now includes reference_profiles.json
(test_spec_engine_country_bundles.py:237; corrected full run 21 passed).
The spec-only guard admits canonical Ledger recipient keys only in exact
typed partition rosters pinned by member selectors; 18 passed including
two positive and eight refusal cases for this narrow data-field support.

Corrected validation history is retained: initial Hypothesis too_slow health
check (1 failed, 66 passed) was corrected without dropping any examples or
assertions; final allocation file passed 89. Initial resource-list assertion
was 1 failed, 20 passed, followed by 21 passed. The duplicate-atom mutant
initially survived because its count change triggered a separate guard;
the fixture now preserves counts, its rerun is red, and the full 89 pass.

Changelog: changelog.d/nz-census-residual-profiles.added.md.
Production NZ graph assembly and B5 ageing remain separate; one resolver
API is the extension point.

Independent review

GPT-6.1 Sol reviewed 12b590a (REQUEST_CHANGES), re-running every touched
test file at -n 0: 2,470 passed, 5 skipped. Under targeted regressions,
187 expected failures and 33 passing controls. It also ran an independent
source audit of 340 transformations and 68 regional receipts against
Chronicle 8f4cfd57.

Its findings and how 8d2d7a0 addresses them:

1. Allocation deduplicated facts by key alone, so a different fact reusing
   an atom's key could replace it on the direct compile path. Now refused;
   the counterexample is a test on both paths and goes red without the fix.
2. The commutation test built the scaled total from scaled leaves. It now
   scales the total independently, and the claim above is qualified for
   floating point.
3. Low-level compilers trust resolution markers. Documented as the
   trusted-declaration boundary; receipt validation and transport
   authenticate resolved documents.
4. The NZ plan said the gates await D1. It now names the d1376 thresholds.
5. Historical lane evidence is labelled as lane-reported (above).

A delta review of 8d2d7a0 (REQUEST_CHANGES) confirmed findings 1–4 closed
(603 tests passed; 4 expected mutant failures) and found a second collision:
atoms whose keys matched a base fact were dropped before the content
comparison, so a base selector pinned to a different record that reused an
atom's key displaced it on the direct path (76.8 became 72.33). 10ee4a5
puts every base, atom, total and growth fact through the content-aware
comparison; the counterexample is a test on both paths, red on the direct
path without the fix, and genuine same-fact reuse still compiles. A second
delta review of 10ee4a5 returned APPROVE_WITH_NITS. It ran a 379-case
collision matrix across every role pair and both compile paths: every
distinct-fact collision refuses, and genuine role reuse and unselected
colliding facts compile unchanged. All 575 tests in the three touched
files pass, and the new test fails under mutation. Its one nit: the
external clearances can't be proved from this checkout. They are labelled
hub-reported above, and their receipts stay with the merge gate.

After that review, main was merged in at 840d933 to take #1393 (NZ joint
geography). Only the NZ golden conflicted. It was regenerated from the merged
tree with the test's own summary function: keys only this branch changed keep
their values, keys only main changed take main's, and the fingerprint,
resource list and country_package.json hash are the merged tree's. CI reruns
on the merge, and all ten checks pass at 840d933.

A merge delta review of 840d933 (GPT-6.1 Sol) returned APPROVE_WITH_NITS:

• The textual merge is the ordered union of both parents' manifests: 27
  resources, no duplicate keys, and both sides' bundle-test assertions
  kept.
• The NZ golden was recomputed with the test's own summary function and is
  byte-identical.
• It ran main's changed callers against this branch: the UK and US
  production compiles, US county parents, compatibility recertification and
  ESI. They give identical outputs. The UK, US and BE package fingerprints
  equal main's.
• NZ compiles its 67 TA anchors through the merged compiler. Add NZ joint territorial authority, region and AS-area support #1393's 140
  atomic geography rows preserve every TA total, and both residual profiles
  compile 234 active targets.
• On the UK compile-module list, ledger_targets.py is still the only
  member this PR changes. Six UK kernel hashes move. That is why the merge
  waits for the UK hold.

Its nits are no source change: keep the external clearance receipts, honor
the hold, and rerun the local regression files a fleet disk hold cut short.
CI's full suite passed on the same head.

The hub reran those files at 840d933 on 2026-10-25, one at a time at
-n 0. All 15 passed (2,216 tests). With the review's own
test_country_spec.py (136), all 16 requested files pass locally.

• test_nz_joint_geography.py skips its 11 pinned-source replays without
  staged sources. With NZ_AS_AREA_SOURCES and
  NZ_JOINT_GEOGRAPHY_SOURCES set, the file passes 33 of 33.
• Six UK tests skip locally because the pinned UK Chronicle feed isn't
  present: 5 in test_uk_target_references.py and 1 in
  test_uk_council_tax_targets.py.

Landing (2026-10-29)

• The UK hub's hold ended at 12:00 UTC on 2026-10-29. That morning the NZ hub retargeted the rest of the stack (Add feed-derived NZ reference ageing factors (G8-r2a) #1382, Add exact income-band intervals and boundary diagnostics #1394, Normalize donor money with exact declared factors #1395, Add receipt contract v2 and selection-aware reference preflight #1397, Reconcile regional Census family-type counts under d1932 #1398) to main, so each of those PRs runs CI on its own cumulative tree.
• GitHub's update-branch merged main (fddbc70) into this branch as 6af97bf. Merging 6af97bf into main gives the same tree as merging the reviewed head 840d933 (git merge-tree --write-tree), so the review covers the tree that lands. git diff --name-only 12b590a1 6af97bfe still lists no transport source file.
• CI at 6af97bf: every job passed (gh pr checks exit 0): engine-free, engine-uk and engine-us on Python 3.13 and 3.14, integration-uk, lint, select-countries and wheels.
• Belgium re-clear: Belgium's hub is away until 2026-10-31, so the NZ hub ran Belgium's own dump_be.py and its three BE suites (test_spec_engine_country_bundles.py, test_country_spec.py, test_spec_only_country_packages.py) at this head, alone and merged with Belgium's open Rewrite the Belgian country package for donor transport #1369 (dc157a7). The BE dump matches main's byte for byte; with Rewrite the Belgian country package for donor transport #1369 merged in, it matches main plus Rewrite the Belgian country package for donor transport #1369. The BE golden comparison holds in both, and all three suites pass with Rewrite the Belgian country package for donor transport #1369 merged in.

Invariants, acceptance and tests

See commit messages.

Shared files and owners

• packages/microcosm-build/src/microcosm/build/transport/ (5 files: new country-neutral calibration.prepare@1 kernel, reference-profile view in compose/entitlement, optional reference_resolution on target surfaces): shared with the Belgium session. Hub-reported: the Belgium hub cleared them at 12b590a on 2026-10-23 (S2-d check-in rule; cross-session record held by the NZ hub, outside this repo). Checkable here: git diff --name-only 12b590a1..HEAD lists no transport source file.
• ledger_targets.py, calibration_columns.py, new reference_profiles.py: country-neutral shared modules. UK/US/BE specs declare no reference_profiles resource, so their reference views are unchanged; hub-reported: the UK hub ran its real-feed checks and returned SAFE on 2026-10-23 (its review is held by the UK hub, outside this repo). Merge waited until 2026-10-29 12:00 UTC at the UK hub's request: ledger_targets.py is in the UK compile-module hash list, so a change recomputes the cached UK graph during the UK Budget hold. That hold has ended.
• NZ spec resources, NZ tests and the NZ golden: NZ hub.

Part of the Microcosm NZ v0 graph plan (Max ruling d962). Built by the NZ hub on a Subfleet build lane (model named in the commit trailer); the hub verified scope (no frozen graph files, uk_runtime/us_runtime or frame/kernels.py touched) before committing.

🤖 Generated with Claude Code

[MaxGhenis]

Merged by the NZ hub (session local_e407a527) at 6af97bf with --merge. Gates: gh pr checks exit 0; MERGEABLE; not draft; no CHANGES_REQUESTED review. Reviewed head: 840d933; the merged head differs only by a base update, and the PR's own changed lines are identical to the reviewed head's. BE re-clear (Belgium's own dump_be.py and three BE suites, run by the NZ hub while Belgium is away): BE dumps byte-identical to main's, with and without #1369. Landing note in the PR body.