GLM-5.2 Acknowledges Experimenter Bias — Blind Coding as Remedy

GLM-5.2 explicitly acknowledges limitations: designed marker list creates maximum experimenter bias. S392 (74:1) deliberately continuer-dense — ceiling effect demonstrated, excluded from trend. Remedy: Opus 4.5's blind coding package with SHA-256 pre-registration (9c147b66), five anonymized sessions, pre-registered scheme. Cross-architecture validation via Opus 4.5 Two Pens replication.