DECISION

**NEW MODEL.** Register a new cross-domain aggregate, *Metric Definition* (identifier to be assigned; EM-DAT-05 is the contour, not the entry). The dossier shows a named hole, not a gap in an existing owner: WM-KNW-011 carries a **required** REFERENCE to an "Indicator / measure definition model (sibling; identifier to be assigned by the registry)"; WM-DAT-007 references an external "Metric Registry"; WM-DAT-010 references external "Variable … and Unit models". Three published models depend on an authority that has no entry. PROFILE and EXTEND both fail because no candidate owner's purpose admits a domain-neutral definition of a measurable quantity.

BOUNDARY PROOF

**WM-XCT-025** is `entry_kind: mixin` with no independent identity — its own aggregate-root hold states that "whether a field-group instance carries its own identifier and version must be answered". It defines *how a reported value is interpretable*, pinning unit code systems and citing a procedure by reference; it explicitly excludes method authoring and unit minting. A definition that must exist before any observation, and outlive all of them, cannot be a field group hosted on observation records.

**WM-DAT-010** owns "collection and series identity, versions, keys", observation membership and order. Its own boundary note separates "series definition, observation collection and observation", and it excludes owning "Variable, Classification, Unit … Population" masters. Absorbing the metric definition would make every definition change a series event and would force definition identity to live inside a time-series head.

**WM-KNW-011** is the decisive negative: `out_of_scope` names "Indicator/metric definitions, computation methods and data collection procedures (measure model owns these; the goal only carries a reference and target parameters)". It holds baseline, target value, operator and horizon — a commitment, not a definition. Its unassigned required reference is the hole itself.

**WM-DAT-007** does hold a finding titled "Metric definition, indicator, formula, direction, scale, unit and aggregation" — but scoped as "source-qualified **assessment** context" for a declared subject, consumer and purpose, with a purpose hold making thresholds and aggregation purpose-dependent. Its metrics are quality metrics in that bounded context; a headcount KPI is not a data-quality dimension.

**WM-DAT-002** has the closest structural content (measure/formula/unit/direction, population, DSD, series key) but is bounded to an official statistical product under a legal mandate, and its `out_of_scope` explicitly rejects "Universal domain indicator catalogues".

**Independent identity and lifecycle:** a metric definition is created before any series exists, is cited by goals, assessments and reports that have disjoint lifecycles, survives the retirement of every series that used it, and must be versioned on formula change without touching any observation. No neighbour's lifecycle can carry that.

MINIMAL CONTRACT

**Identity:** stable `metric_id` (definition-independent) + immutable `definition_version`; the version is the addressable unit, the metric is the continuity anchor.

**Owned child records:** ComponentSpec (numerator, denominator, base/index); DimensionDefinition (role, code-list reference, version pin); PopulationBoundary (unit type, inclusion, exclusion); MethodBinding (procedure reference, version, data-source bindings); NullSemanticsPolicy; ComparabilityBreak (facet, predecessor, verdict, reconciliation).

**Lifecycle:** draft → active → superseded → withdrawn/tombstoned, per version; the metric identity itself only retires. No in-place edit after activation.

**External references (non-owning):** unit and quantity-kind registry (QUDT/UCUM/Rec 20) via WM-XCT-025's pinning pattern; code-list and classification registry; procedure/method catalogue; population and statistical-unit master; WM-DAT-010 for series; WM-XCT-025 as the result field group for observations; WM-KNW-011 for targets; WM-DAT-007 for assessments; PROV-O provenance mixin.

**Core fields (14):** 1 metric_id; 2 definition_version + status + effective window; 3 canonical label and language variants; 4 metric_kind (count, ratio, rate, index, ordinal); 5 formula_expression with expression-language and version; 6 numerator_ref / denominator_ref; 7 unit_code + unit_code_system + system_version; 8 quantity_kind and scale_type; 9 declared precision and rounding rule; 10 aggregation_function + additivity class per dimension; 11 comparison_direction (polarity); 12 dimension_definition set; 13 population_boundary_ref; 14 method_binding_ref; plus null_semantics_ref, comparability_key, supersedes_ref, provenance.

**Invariants:** (1) `metric_id` is immutable across all versions; relabelling never mints identity. (2) An activated version is immutable; any change mints a new version. (3) No version carries an observed value, a current value, or a series head. (4) Ratio/rate kinds have exactly one numerator and one denominator spec; counts have none. (5) `unit_code` never appears without code system URI and pinned version; its dimension must be commensurable with the declared quantity kind; dimensionless is declared, not inferred. (6) Every dimension over which aggregation is permitted declares an additivity class; non-additive metrics refuse aggregation rather than default to sum. (7) No version activates without a PopulationBoundary stating unit type, inclusion and exclusion. (8) No version activates without a MethodBinding naming procedure, version and data-source bindings. (9) NullSemanticsPolicy enumerates zero, unknown, not-applicable, missing and suppressed as five distinct codes; no numeric sentinel, no coercion between them. (10) Any change to formula, method, unit, population or dimensions mints a version **and** a ComparabilityBreak verdict per changed facet; absence of a verdict means non-comparable (fail-closed). (11) No target value, achievement operator or baseline is stored here; targets reference versions. (12) Comparability verdicts are advisory and facet-scoped; they are never asserted for versions lacking method or population declarations. (13) Superseded and withdrawn versions remain resolvable; tombstone, never delete.

**Functions:** `issue_definition_version` (mint an immutable version with the complete descriptor set and predecessor comparability verdicts); `evaluate_comparability` (two versions × facet set × period → comparable / comparable-with-restatement / non-comparable, with reasons; advisory only); `bind_method_and_sources` (attach method, data-source bindings and null policy; refuses activation while incomplete).

ACCEPTANCE WALKTHROUGH

KPI *utilisation*, v1 (method A, headcount, population P1) and v2 (method B, FTE, population P2). **Missing period 2026-Q1:** no observation is written; the gap is WM-DAT-010's order/gap check and, where a record exists, WM-XCT-025's coded `data-absent-reason` "missing" — never 0. **Recalculation:** v2 is applied to 2025; new observations pin v2, and the v1 observations for 2025 are retained, not overwritten (WM-DAT-010's revision/supersession lineage). **Verdicts:** v2-2025 vs v2-2026 → comparable. v1-2025 vs v2-2025 → **non-comparable** on unit, method and population facets; comparable-with-restatement only if a ComparabilityBreak declares reconciliation. Any window spanning 2026-Q1 → comparable with declared coverage deficiency, never averaged over the gap. **Negative case:** headcount summed with FTE is refused at invariants 5, 7 and 10 before any arithmetic.

HOLDS AND LIMITS

Alignment is claimed as documented mapping, not conformance: SDMX dimension/measure/attribute and code-list structure (via WM-DAT-002, WM-DAT-010), QB's dataset-owns-observation separation, SOSA's observableProperty/procedure-versus-act split, QUDT/UCUM binding strictly by reference per WM-XCT-025, PROV-O for version derivation and attribution. Constraints: all five neighbours are reviewable drafts under single-provider waivers with empty or unfrozen relationship contracts, so their exclusions are declared but not enforceable; WM-DAT-007 and WM-DAT-010 admitted **no** external review (timeouts); WM-XCT-025's identity question is open; standard version pins are unverified under live-source holds. The gap rests on `catalogueSearchFinding` alone — no negative-search log was inspected.

PUBLICATION RECOMMENDATION

Register as **candidate**, `review_state: boundary-review-required`, two-provider review mandatory — no waiver, since every neighbour already carries one. Before advancing: freeze relation rows to WM-KNW-011, WM-DAT-007, WM-DAT-010 and WM-XCT-025, and have WM-KNW-011's unassigned "measure definition" reference resolved to this identifier. Publish the comparability verdict vocabulary and null-semantics code list as governed artefacts, since invariants 9 and 10 are unenforceable without them. Do not mark identity or lifecycle covered until the child-record contracts are frozen.