Applying the discovery–understanding–validation distinction: a policy warning about understanding cannot be counted as a measured failure of validation, and faster discovery alone measures neither. Google DeepMind's November 2024 essay, corrected in May 2025, identifies reliability and scientific understanding as separate concerns alongside limited policy evidence. It documents an institutional policy argument, not field-wide prevalence. To test that argument, investigators would need separate measures of what was produced, what reviewers could explain, and what claims they could check. This measurement proposal is synthesis, not a finding of the essay. Its historical capability discussion is not a November 2026 inventory.

Evidence status: attributed_policy_analysis_with_measurement_synthesis.