Automated wildlife detection from satellite imagery could transform ecological monitoring, but no study has quantified the domain gap between drone-trained detectors and real satellite data at matched resolution. We present results from a same-day paired acquisition at Año Nuevo Island, California (March 19, 2026): a drone orthomosaic at 2 cm ground sample distance processed through production pinniped and seabird detectors, paired with a Maxar WorldView Legion scene at 36 cm acquired the same day. Drone-trained detectors applied to synthetically degraded imagery retain partial detection through 36 cm, but the same detectors on real satellite imagery produce zero detections. This sim-to-real gap confirms that sensor-aware domain adaptation is required for satellite-scale wildlife census.