One of the fundamental problems of using optimization models that represent complex systems – e.g. power systems on their path towards achieving net-zero emissions – is the trade-off between model accuracy and computational tractability. Many applied optimization models that use different time series as data input have become increasingly challenging to solve due to the large time horizons they span and the high complexity of technical constraints with short- and long-term time dynamics. To overcome computational intractability of these optimization models, the dimension of input data and model size is commonly reduced through time series aggregation (TSA) methods. However, applying TSA for optimization models that are governed by varying time dynamics simultaneously is quite challenging. TSA methods mostly focus on short-term dynamics, and rarely include long-term dynamics due to the inherent limitations of TSA. As a result, longer-term dynamics are not captured well by aggregated models, which is imperative for reliably modelling many complex systems. Moreover, traditional TSA methods are based on the common belief that the clusters that best approximate the input data also lead to the aggregated model that best approximates the full model, while the metric that really matters –the resulting output error in optimization results – is not well addressed. This belief is mainly based on the lack of theoretical underpinning relating inputs and output error, rendering existing methods trial-and-error heuristics at best. We plan to challenge this belief by discovering the currently unknown relation between input and output error, and to overcome existing TSA shortcomings by developing the novel theoretical TSA framework for optimization models with varying time dynamics, thereby tapping into unprecedented potential of computational efficiency and accuracy. If this project is successful, it would have untangled the Gordian knot of data aggregation in optimization.