Binned Kepler photometry of KOI populations above/below the period–radius valley (top/bottom) with/without (red/blue dots) injected refraction signature (solid red line). Prior to binning, a random sample of 300 light curves were injected with a simulated refraction effect—limited to a maximum flux increase of ∼10 ppm—in order to assess its significance across the out-of-transit photometry. To protect against the potential for in-transit data to contaminate the out-of-transit refraction signal near ingress/egress by means of Kepler’s 30 minutes cadence, we exclude data within ±0.1 T14 from ingress/egress accordingly.