We present a possible strategy for filling the missing data of the CATS benchmark time series prediction competition. Our approach builds upon an appropriate embedding of this time series and the use of bagging of multilayer perceptrons (MLPs). We exploit time-reversal symmetry for prediction within the first four gaps, linking the missing state to symmetrically-located information both in the past and future. One-shot forecasting is then performed for each missing value from distant-enough delays. The suitability of the proposed embedding is assessed empirically by t-testing the goodness-of-fit of models built in symmetric versus asymmetric input spaces. Since this approach cannot be pursued for forecasting the continuation of this time series, in the right end we perform standard, non-iterated forward predictions. Expected error levels are provided according to the performance on test data.