Signal detection techniques in longitudinal databases have been extensively piloted in the U.S. Food and Drug Administration's Sentinel Initiative. The Sentinel Innovation Center's leveraging of electronic health record (EHR) data presents new challenges when accounting for data missingness and the capture of key features in unstructured clinical text. How can we prioritize the signal among the noise, or the right hay in the haystack?