# EM-DAT-05 provider comparison

Claude and Grok independently select **NEW MODEL** for a cross-domain Metric Definition aggregate, while both forbid guessing a numeric identifier. They agree that no current Vercy owner has compatible identity, purpose and lifecycle: result fields are a mixin, observation collections own series and vintages, goals own target commitments, quality assessments own contextual findings, and official statistics own products.

The reconciled candidate keeps two durable records: stable Metric identity and immutable DefinitionVersion. It owns formula/components, quantity-kind and unit/scale references, dimension and additivity declarations, population rules, method/source bindings, absence semantics and explicit comparability declarations. It references observations, series, goals, targets, procedures, unit and population masters, quality assessments and reports.

Grok sharpened four rules adopted in candidate.2: quantity-kind or phenomenon change creates a new Metric identity; facet comparability may narrow but never silently widen the aggregate verdict; restatement creates new observation vintages without overwriting history; and equal-authority definition conflicts are not silently merged. Fixtures now include headcount-plus-FTE refusal, undeclared-additivity refusal, preserved restatement vintages and target-version pinning.

Both reviews support a bounded identifier-unassigned research candidate only. Registry allocation, neighbouring draft holds, frozen relation rows and canonical publication remain outside this checkpoint.