Cerebellar disease makes ordinary movements extraordinarily difficult, often resulting in endpoint errors. For example, damage to lobule VII of the vermis makes saccadic eye movements dysmetric. These symptoms have suggested that the cerebellum monitors ongoing commands and adjusts them, particularly as the movement nears the target. Yet, individual Purkinje cells (P-cells) have firing patterns that are modulated much longer than the movement. Thus, it has been difficult to decode the activities of P-cells, and their downstream nucleus neurons, with respect to computations that are necessary for control of movements. A key to this puzzle is that the inferior olive monitors the output of the cerebellum and returns to it information that appears to encode error [1–4]. This input to the cerebellum organizes P-cells and nucleus neurons into anatomical groups called micro-clusters [5–7]. Cells within a micro-cluster likely have a common feature: they respond similarly to error. Through a collaboration between Shadmehr, Soetedjo, and Kojima, we used this idea to show that in macaques, if P-cells were organized into groups based on their complex spike response to error, then their simple spikes as a population produced a rate coding that predicted parameters of the ongoing movement [8,9]. The result was a new idea: the fundamental computational unit in the cerebellum may not be an individual cell, but a population of cells that share a common preference for error. Here, we propose that P-cells that respond similarly to error are part of a network that exhibits a special property: within this network, the P-cells not only coordinate their firing rates, but also temporally align their spikes, especially during the deceleration phase of movements. That is, P-cells transmit information to the nucleus by modulating their firing rates, and synchronizing their spikes. In our hypothesis, P-cells combine disinhibition with synchronization to signal when the movement should be stopped [10]. To pursue this idea, in 2016 we built a marmoset lab, pioneered techniques to train the animals, and then used silicon probes to record from many neurons simultaneously [10,11]. We then built new tools for precise temporal analysis of cerebellar spikes [12]. Here, we propose to record from P-cells, molecular layer interneurons (MLIs), and nucleus neurons, use their error response to organize cells into populations, and then quantify both firing rates and spike timing during movements. Our proposed experiments have the potential to produce simultaneous recordings of P-cells, MLIs, and nucleus neurons, something that is unprecedented in primates. We will use this neurophysiological approach to test the anatomical basis of our hypothesis, that the inferior olive organizes the cerebellum into cell-assemblies. The data will allow us to determine whether the healthy cerebellum relies on synchronization to encode information, shedding light on conditions such as dysmetria and trem...