Remote Viewing — Reproducible Protocol

A preregistered, multisite protocol to test free-response remote viewing with strict leakage control. Viewers generate narratives and sketches; scoring is double-blind via a fixed automated text→image similarity model (e.g., CLIP) and independent human judges (5-AFC; chance = 20%). Randomization and target locks are hash-committed in advance. Optional stratification includes Local Sidereal Time windows and logging of geomagnetic Kp as a covariate. The primary KPI is rank-1 hit-rate versus 20% (binomial/permutation + Bayes factors); secondary endpoints use continuous similarity scores and inter-rater reliability. Power targets a small uplift (e.g., 26% vs 20%), requiring ~400–600 independent trials, with design-effect corrections for clustering. No optional stopping. Go/No-Go requires p<0.01 (FDR-corrected) and BF10>10. All data, code, and preregistration artifacts are released openly.