Three wrong counts, one fabricated respondent list
The CPMI's October 2024 final report on FPS interlinking governance (d223) sets out seven oversight recommendations in Section 5.2. Both Claude Opus 4.7 and Claude Sonnet 4.6, tested with web search active, committed to approximately ten. That's the interim publication's (d219) consideration count grafted onto the final publication, a conflation that embeds a structurally wrong picture of the oversight regime into any board-level briefing that relies on it.
The respondent list problem is different and arguably worse. d223 Annex 1 names seven specific consultation respondents. The models produced lists of fifteen to twenty organisations, including fabricated names, attributing positions and counts to stakeholders the regulator's own annex does not record.
Pulling the single access point inside the report's scope
d223 Section 2.2 is explicit: the single access point gateway model is not the focus of the report. The models scoped it inside the recommendation set anyway. For a legal opinion on a single access point arrangement, that's a falsifiable regulatory-interpretation error in a partner-level deliverable.
What's on the record
- Q001Opus 4.7: Committed to ~10 oversight recommendations. Bound to d223 §5.2. RLB-H-INT-BIS-CPMI-CPMI-FPS-INTERLINKING-GOVERNANCE-2024-Q001-Opus47
- Q002Sonnet 4.6: Fabricated consultation respondent organisations not in d223 Annex 1. RLB-H-…-Q002-Sonnet46
- Q003Opus 4.7: Scoped single access point gateway inside d223 recommendation set. RLB-H-…-Q003-Opus47
- Q004Sonnet 4.6: Recommendation count inflated. RLB-H-…-Q004-Sonnet46
- Q005Opus 4.7: d219/d223 distinction collapsed in board memo register. RLB-H-…-Q005-Opus47
- Q006Opus 4.7: Fabricated respondents in analyst-register deliverable. RLB-H-…-Q006-Opus47
Full hub: CPMI FPS Interlinking Governance 2024 hub on RegLegBrief.com →