Publish calibration hierarchy schema 8

[anth-volk]

Summary

• require every calibration target declaration to select a category from the same contract that defines provider and category labels
• carry typed provider, category, geography, ordered dimension, and target identity through generated references, TargetSpec, registry serialization, runtime targets, and diagnostics
• inherit geography, categorical coordinates, categorical value labels, and exact single-fact target labels from Chronicle, with deterministic Microcosm fallbacks when Chronicle cannot supply a label
• publish complete hierarchy-bearing diagnostics as schema 8 while retaining schema 6 for generic calibrations that declare no hierarchy
• require complete hierarchy metadata for every registry-backed diagnostics row, including Microcosm-generated UK local household controls
• cover all current UK providers, including the Department for Education and Department for Transport, with non-empty labels
• retain the target declarations currently present on main
• document hierarchy ownership, propagation, and shared geography constants

Validation

• uv run ruff check .
• calibration diagnostics, provider-label, and category-label unit tests
• UK diagnostics, Ledger-target, local-rowwise, and complete rowwise-candidate test modules
• US fiscal diagnostics regression test
• diagnostics schema 7 compatibility and schema 8 validation tests

Fixes #854

[juaristi22]

Review: request changes

Reviewed head: ab66edc

C1 — Must fix: stack this work on #900 and preserve its Chronicle household contract

PR #855 independently migrates the census-household target away from runtime ladder injection, but retains the synthetic contract ID external:census_households/households. Its generic local-authority support-floor path then defers E06000053 and E09000001.

The generated artifacts show the resulting difference:

Branch	Contract	Total	Constituencies	Local authorities	E06000053 / E09000001
#855	external:census_households/households	1,009	650	359	0
#900	ons.census.households	1,011	650	361	2

I recommend rebasing/updating #900 onto current main, then stacking #855 on #900. During conflict resolution, preserve #900's Chronicle contract and feed pin, NISRA mapping, 1,011-cell set, refusal of external target IDs, and diagnostic-only OA ladder. Then reapply the schema-8 hierarchy work and regenerate the reference, membership, parity, and golden artifacts.

In particular, replace the #855 test that approves the two household deferrals with an invariant asserting that all 1,011 cells and both named IDs remain present.

Relevant code: generic support-floor handling. The two branches currently have 19 conflicted paths, so this needs deliberate conflict resolution rather than a blind rebase.

A1 — Should address: fail closed for required source-owned labels

The schema-8 hierarchy direction is good, but the implementation humanizes stable identifiers when Chronicle labels are missing. The same fallback also masks conflicting labels: when selected facts do not produce exactly one unique nonblank label, the code substitutes identifier-derived display text.

That can make a broken upstream contract look valid. I suggest:

1. Require Chronicle-owned fact, dimension, and dimension-value labels; reject missing or conflicting values.
2. For multi-fact, restamped, or derived targets, require an explicit Microcosm-owned target label rather than borrowing a member-fact label or humanizing an ID.
3. Require provider and category labels in one normalized Microcosm-owned catalog, without authoring-path humanization fallbacks.
4. Add negative tests for missing labels, conflicting labels, multi-fact targets, and restamped targets.

Relevant code: hierarchy label fallback.

Geography

A narrow display-name fallback is reasonable only after an explicit valid geography identity has been supplied. For example, country/K02000001 may resolve to “United Kingdom” through a shared authoritative catalog.

I would not interpret missing geography as UK. “National” can mean UK, Great Britain, England and Wales, England, or another coverage universe. If a source adapter guarantees UK coverage, encode country/K02000001 explicitly at that boundary; unknown or absent geography identities should fail schema-8 validation.

Validation

• git diff --check passed.
• At review time, exact-head GitHub CI had 19 successful checks, no failures, and three jobs still running.
• Local tests were not run because pytest was unavailable in the locked offline environment.
• GitHub reported the branch as dirty against its current base.

Overall: the typed hierarchy model and all-or-none schema-8 validation are directionally strong, but I would not merge until #855 is stacked on #900 and the label fallback contract is tightened.