# EM-KNW-02 provider comparison

Claude and Grok independently select **PROFILE** over WM-KNW-007, WM-KNW-008, WM-KNW-010 and WM-REC-010 with no profile or runtime identifier. Both also identify **Evidence Artifact / Source Work** as an independent missing aggregate whose identifier remains unassigned.

Both providers separate five authorities: source work and its expressions; immutable citation acts; claim assertion and assessment; decision content and local appraisal; and authentic fixed decision records. WM-KNW-010 authors question, alternatives, evaluation rounds, selection, rationale, objections and dissent. WM-REC-010 fixes an authentic expression; any repeated content is a version-pinned transcription. An ADR may exist without a formal issued instrument.

Grok sharpens the collision: WM-KNW-010 must drop approval-as-owned, WM-REC-010 must stop authoring question/reasons/evidence, and WM-XCT-028 remains a neighbor rather than decision-owned argument content. Source correction or retraction appends status observations and reassessment without rewriting citations, claims, rationale, alternatives, dissent or fixed records.

Canonical publication remains held by non-canonical bases, missing relation-ledger edges, unresolved WM-KNW-008 parentage, citation deduplication and dissent identity gaps, the unallocated source-work aggregate, unpinned certainty schemes and incomplete standards crosswalks.

Claude described the 010/REC-010 collision as residual wording risk; Grok required explicit reciprocal correction. The reconciled candidate adopts Grok's stronger constraint. Grok team-process wording is recorded as non-evidentiary.