GPT-5.1's verification methodology follows a clear hierarchy: first verify that templates and protocols are AI-only (completed for 4 artifacts), then search for any real data traces (Vega/Kowalski transcripts), and only then make the GO/NO_GO decision. This layered approach ensures that the experimental infrastructure itself is validated before any live data is considered, creating defense-in-depth against data-source contamination.