The document proposes an adaptive code offloading framework called EMCO that uses evidence-based learning. It aims to optimize the code offloading decision process by treating offloading as a global learning process based on analysis of offloading traces from many mobile devices, rather than a local decision. The framework uses fuzzy sets and linguistic variables to represent parameters like bandwidth, and rules to determine whether to offload code based on these variables and cloud/mobile conditions. Preliminary results show the percentage of times code is offloaded or not for different combinations of bandwidth, data size, instance load, and other factors. The conclusions discuss how periodic cloud analysis can empower mobile devices with knowledge gained from traces to allow self-adaptive off