It seems to me it would be nice to create a benchmark that allows testing algorithms for the prediction of scalar quantities.
As an option - it would be possible to take the data sets from the NAB.
I think there are existing benchmarks or datasets for time series forecasting, mainly in an academic context. Here are some pointers: http://stats.stackexchange.com/questions/48500/where-can-i-find-time-series-data-to-assess-accuracy-of-forecast
could you please explain me how your contextual anomaly detection works?