Skip to content

Let a parentless NZ reference person under 18 head a unit (B8 amendment) - #1628

Merged
MaxGhenis merged 3 commits into
mainfrom
nzhub/b8-reference-teen-heads
Nov 16, 2026
Merged

MaxGhenis merged 3 commits into
mainfrom
nzhub/b8-reference-teen-heads

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

Why

NZ's first real-data transport run (G8-x step 4) refused the whole frame in unit construction with "The unit rule refuses a parentless child whose reference person is a dependent child." NZ's rule set refuse_dependent_reference_person: true, and _heads checks it with a frame-wide .any(), so one household aborts CREATE.

The NZ hub counted the households with the build's own code (CREATE's reader, string-dtype fix and declaration) on the pinned donor:

  • 9 of 57,240 households trip it, 0.0024% of household weight.
  • Case (ii), 6: the reference person is 15 or 16 with a parent present, and another child has no co-resident parent. This is the refusal B8 states.
  • Case (i), 3: the reference person is 16 or 17 with no co-resident parent. The constructor kept them dependent and treated them as their own parentless child.
  • Every one has a member aged 18 or over; none is minors-only.
  • Without the 9, the strict rule builds the frame. With the flag false, the whole frame builds.

What changes

The NZ hub amends B8 (2026-11-15). Literal B8 would still refuse case (ii), and the declaration has no setting that keeps case (ii) while dropping case (i).

  • Rule. refuse_dependent_reference_person is false in NZ's benefit_unit_rule.json. A reference person under 18 with no co-resident parent heads a unit, and the dependent children who name them as parent 1, and every other dependent child with no co-resident parent, join it. In case (ii) the reference person stays dependent, and the parentless child follows the caregiver chain into the unit that holds the reference person, their parent's. Every frame that built before builds the same units: under the old flag, any parentless reference candidate made _heads raise.

  • Disclosure. The rule file and docs/nz-benefit-unit-rule.md record the amendment: its counts, the new non-statutory approximation (in unobserved_statutory_limbs), and the Accommodation Supplement consequence (a minor heading a unit with a dependent reads as a sole parent; on this donor the 3 new heads are one-person units). They also record the refusal that remains: the caregiver loop, "Dependent caregiver pointers do not reach an adult."

  • Receipt (shared code). benefit_unit_diagnostics adds two keys:

    • parentless_reference_person_heads, which shows the new branch (none under own-unit placement, which makes every parentless candidate a head);
    • younger_independent_teen_parents, parents under 16 who aren't dependent. With it, the three teen-parent counts partition parents under 18. Before, a partnered parent under 16 was in neither count, and the docstring's partition claim was false.

    The five existing expressions are unchanged. Only NZ declares transport.create.

  • Tests.

    • The two refusal fixtures now expect the built partition and roles. New fixtures pin the donor's shapes and a minors-only household.
    • The property tests draw reference persons of any age. An independent oracle predicts the caregiver-loop refusal, and the adult, partition and reference-head invariants are checked on what builds.
    • test_strict_reference_flag_is_read_from_spec keeps the shared strict branch and the flag's round trip covered.
    • test_reference_heads_count_only_the_reference_branch runs each placement with the strict flag on and off.
  • Golden and changelog. The NZ fingerprint and the rule-file hash are regenerated; only those two lines move. There is a changed fragment.

Review

  • Design. An in-session design workflow (four designers, two adversarial critics) produced the edits. The hub hand-checked the two designs whose critics didn't run and every expected count.
  • Independent review. Round 1 (Subfleet job 20261115-144611-mc1628-r1, at 08d4460) raised two P1s and one P2. The changelog said every listed reference person heads a unit, though case (ii)'s stays dependent. The rule text said a reference head's own children join their unit, though a child follows parent 1. The reference-head count also counted own-unit heads. cfdd3c2 fixes all three. It also folds in the Belgian port's wording notes and regenerates the golden; 1fec5b2 clarifies one docstring sentence. Round 2 (job 20261115-180127-mc1628-r2, at 1fec5b2) approved: all three closed, all seven mutants caught, and the NZ, Armenian and Belgian golden hashes recomputed to match. Its one P3, on wording inherited from main ("otherwise dependent" in the under-16 parent passages), goes to a later wording PR.
  • Belgian port. It OK'd the shared hunks at 08d4460 and re-confirmed them on read at cfdd3c2 and 1fec5b2 (its optional docstring note), on the condition that CI pass the new test. Both engine-free legs ran test_transport_dependent_children.py: 29,673 passed, none failed.
  • Not Max's call. This stays the NZ hub's call on Settle and apply the NZ dependent-child benefit-unit rule (G8-u) #1226's grounds: the NZ package is inactive and publishes no number, and the spec is reversible data. Round 1 marked it "Needs Max" because the docs and changelog carry the donor counts. Round 2 flagged the same. Repo docs and changelog fragments are not publications: the repo has no docs site, GitHub Pages or homepage. Merging cuts no release either, since the repo's only workflow is test.yml.

🤖 Generated with Claude Code

MaxGhenis and others added 3 commits November 15, 2026 14:44
NZ's first real-data transport run refused the whole frame in unit
construction: refuse_dependent_reference_person was true, so a household
whose reference person was a parentless minor, or a parented minor living
with another parentless child, aborted CREATE. On the pinned donor 9 of
57,240 households did this (0.0024% of household weight), each with an
adult present.

The NZ hub amends B8: the flag is false, so a reference person under 18
with no co-resident parent heads a unit, and a parentless child in a
household whose reference person is a parented dependent child joins the
unit that holds the reference person. Every frame that built before builds
the same units.

- benefit_unit_rule.json and docs/nz-benefit-unit-rule.md record the
  amendment, its counts, the non-statutory approximation, the
  Accommodation Supplement consequence and the refusal that remains.
- benefit_unit_diagnostics gains parentless_reference_person_heads and
  younger_independent_teen_parents, so the receipt shows the new branch and
  the teen-parent counts partition parents under 18.
- Tests: the refusal fixtures now expect the built units; the property
  tests draw minor reference persons, with an oracle for the caregiver-loop
  refusal; a strict-flag test keeps the shared strict branch covered.
- Golden fingerprint and rule hash regenerated; changelog fragment.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review round 1 (Subfleet job 20261115-144611-mc1628-r1, head 08d4460):
- P1: the changelog said every listed reference person heads a unit. A
  reference person with a co-resident parent stays dependent. The fragment
  now describes the two cases separately and notes that a household whose
  caregivers never reach an adult is still refused.
- P1: the rule text said a reference head's own co-resident children join
  their unit. A child follows parent 1, who can be another adult. The
  passages now name the dependent children who name them as parent 1.
- P2: parentless_reference_person_heads counted heads made by own-unit
  placement. It now counts none under that placement; a new test covers
  each placement with the strict flag on and off.

Belgium's notes 1, 2 and 5 on 08d4460: the diagnostics docstring, the
teen-parent and reference-placement wording in the rule and its docs, and
the reported-count bullet. The docs now say "dependent child" where they
said "child". The rule bytes changed, so NZ's golden hash and composition
fingerprint are regenerated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ring

Under own-unit placement, strict frames build and count no reference head,
so "refuses those frames" applies only to the other placements (Belgian
port's optional note on cfdd3c2).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 16, 2026 03:53
@MaxGhenis
MaxGhenis merged commit 67ab33e into main Nov 16, 2026
12 checks passed
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

Merge audit (NZ hub). Squash-merged at reviewed head 1fec5b2 as 67ab33e.

  • Checks. All 12 passed on 1fec5b2, including the seven the ruleset requires. Both engine-free legs ran test_transport_dependent_children.py (29,673 passed, none failed).
  • Main since CI. Main gained only Bind Brussels and Flemish post-18 education flags to enrolment #1611 and Make Belgium's gate rows pass the transport gate registry #1626. Neither touches this PR's files, tests or NZ resources.
  • Review round 1 (Subfleet 20261115-144611-mc1628-r1, at 08d4460) raised two P1s (changelog and rule wording) and one P2 (the reference-head count under own-unit placement). cfdd3c2 fixed them.
  • Review round 2 (Subfleet 20261115-180127-mc1628-r2, at 1fec5b2) approved. It confirmed all three closed and caught 7 of 7 mutants. The NZ, Armenian and Belgian golden hashes matched on recompute. Its one P3, on wording inherited from main, goes to a later wording PR.
  • Belgian port. It OK'd the shared unit_construction.py bytes at 1fec5b2 (sha256 faf131c8…), on the condition that CI pass the new test, which it did.
  • "Needs Max." Both rounds flagged the donor counts in the docs and changelog. Repo docs and changelog fragments are not publications: the repo has no docs site, Pages or homepage, and merging cuts no release. This stays the NZ hub's call, on Settle and apply the NZ dependent-child benefit-unit rule (G8-u) #1226's grounds.

MaxGhenis added a commit that referenced this pull request Nov 16, 2026
Main brought nine first-parent merges since c4ad232: #1626 (Belgium gate
rows), #1628 (NZ B8 reference teen heads), #1621 (County Files household
measures sidecar and schema manifest tools), #1540 (L0 release practice
docs), #1609 (gate battery shared scopes), #1619 (ACS partnership_income
kept at zero where SEMP holds it, #1387), #1629 (untimed composed stage-scan
test), #1594 (withdraw a release build) and #1541 (Microcosm name sweep).
Main changed nothing under packages/microcosm-fit or packages/microcosm-graph.

Four pin files conflicted; each block took main's text plus this branch's
own comment lines, placed after main's #1387 sentence:
- docs/evidence/spec-engine/us-f0-coverage.json: five digest blocks, main's
  side (the branch has no comment there).
- packages/microcosm-build/src/microcosm/build/spec_engine/inventory_coverage.py:
  main's #1387 comment, then the branch's #1543 sentence, then the digests.
- packages/microcosm-build/tests/engine_free/shared/test_spec_engine_loader.py:
  main's #1387 comment, then the branch's fit-input-checks sentence, then the
  golden.
- packages/microcosm-build/tests/engine_workflow/us/test_us_multispine_pool_tool.py:
  main's #1387 comment, then the branch's #1543 line, then spec_sha256.
No code or test file conflicted, and no other file was changed by both sides.

The tools then wrote every value on the merged tree:
- tools/repin_us_spec_engine_contracts.py: EXPECTED_HASHES seed_map
  d5898b93 and seed_protocol 206be396, the loader golden b7c843a1 and the
  multispine pool spec_sha256 4062e22e (its first pass moved the coverage
  tool test's covered_item_count to 40; its second pass restored 41, so that
  file equals main's).
- tools/spec_engine_coverage.py: us-f0-coverage.json's eight digest lines
  (42395/42395 configuration fields; 41/41 inventory checks).
- fit.qrf/pins.json keeps the branch's values: the merged fit sources equal
  the branch's and the kernel's implementation hash reproduces, so
  tools/graph_parity_repin.py was not run.
The repinner's quote style was then reformatted with ruff format.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant