Biochemical oscillations are ubiquitous in nature and allow organisms to properly time their biological functions. In this paper, we consider minimal Markov state models of non-equilibrium biochemical networks that support oscillations. We obtain analytical expressions for the coherence and period of oscillations in these networks. These quantities are expected to depend on all details of the transition rates in the Markov state model. However, our analytical calculations reveal that many of these details -specifically, the location and arrangement of the transition ratesbecome irrelevant to the coherence and period of oscillations in the limit where a high chemical affinity drives the system out of equilibrium. This theoretical prediction is confirmed by excellent agreement with numerical results. As a consequence, the coherence and period of oscillations can be robustly maintained in the presence of fluctuations in the irrelevant variables. While recent work has established that increasing energy consumption improves the coherence of oscillations, our findings suggest that it plays the additional role of making the coherence and the average period of oscillations robust to fluctuations in rates that can result from the noisy environment of the cell.In principle, R depends on all the details of the rates k ± i in the network. However, in line with a large body of work that generically connects energy dissipation to accuracy in biophysical processes [20][21][22][23][24][25], it has been suggested that irrespective of these details the affinity bounds the coherence of biochemical oscillations [8,17,18,26]. In particular, Barato and Seifert recently conjectured an upper bound on R as a function of the number of states N and the affinity A of the biochemical network [8]. The bound is saturated when the network is uniform; that is, when all of the counterclockwise (CCW) rates in the network are equal and all of the clockwise (CW) rates are equal. In this uniform limit, the bound tells us that only two variables (N and A) determine R. However, the bound is a weak constraint for non-uniform networks with arbitrary rates [8,17] and hence it is unclear which variables control the time scales of the oscillator in the presence of rate fluctuations. If the time scales depend sensitively on all of the rates in the network, then they might vary dramatically with any fluctuations. Conversely, if they depend on only a small subset of the variables, then they will be robust to any fluctuations arXiv:1808.04914v2 [cond-mat.stat-mech]