# EM-RSK-02 provider comparison

Claude and Grok agree that ProcedureDefinition is the only independently identified new-model candidate. A performed test is execution-local and pins one authorised procedure version plus criteria. They agree that AssuranceOpinion is engagement-scoped, CorrectiveAction profiles WM-ACT-006, management response is separate, and finding, issue, risk, control and task retain distinct identities.

The reconciled dual-root contract adopts Grok's exclusive composition: WM-ECO-035 alone masters Engagement, Mandate, declared scope, exclusions and AssuranceOpinion. WM-ACT-033 alone masters commissioned Execution, performed tests, observations, evidence packages and AuditFinding. Engagement commissions Execution; one Execution belongs to one Engagement and cannot exceed its scope or life. Shared wording is reference, never duplicate storage or identity.

Sampling is execution-local and explicit: population, frame, frame gaps, method, sample and sampling cell. Evidence is bound to one performed test and cell. Criteria are pinned before evaluation. Opinion range is limited to declared scope minus exclusions and frame gaps, and every limitation is stated. One-system evidence never proves group-wide conformity.

Finding closure requires a new independent retest against the same pinned criteria and cell, or an explicitly declared successor cell. Task completion, corrective-action completion, management response and opinion never close a finding. No identifier or runtime ID is allocated; publication remains held on registry allocation and canonical dual-root adjudication.