Résumé : Selecting appropriate visual encodings is critical to designing effective visualization recommendation systems, yet few findings from graphical perception are typically applied within these systems. We observe two significant limitations in translating graphical perception knowledge into actionable visualization recommendation rules/constraints: inconsistent reporting of findings and a lack of shared data across studies. How can we translate the graphical perception literature into a knowledge base for visualization recommendation? We present a review of 59 papers that study user perception and performance across ten visual analysis tasks. Through this study, we contribute a JSON dataset that collates existing theoretical and experimental knowledge and summarizes key study outcomes in graphical perception. We illustrate how this dataset can inform automated encoding decisions with three representative visualization recommendation systems. Based on our findings, we highlight open challenges and opportunities for the community in collating graphical perception knowledge for a range of visualization recommendation scenarios.