Currently you can raise an availability alert if a threshold if breached for a number of consecutive tests. However this is applied at quite a high level. for some thresholds such as CPU usage it would be very useful to be able to set the threshold upon the rolling average value or the 95% level rather than the current snapshot; this will enable short term spikes to be excluded.
Originally I though setting this on the threshold was the best approach, but it would be more flexible to create computed attributes. The computed attributes would be a logical and mathmatical function of the attributes available to that monitor. examples include: average CPU over last n samples