Smarter anomaly detection

We have shipped several improvements to the accuracy and consistency of our anomaly detection:

• Metaplane now uses more than one year of history to better detect and adjust for monthly seasonality.
• You can now specify if a monitor should only have an upper or lower bound. For example, you probably don't care if you get fewer product returns than normal, but you probably do care if you get a lot more.
• If Metaplane never observes a zero value, it won’t consider it in future predictions. For example, if a table drops all it's rows, and that’s not normal behavior, we’ll let you know.
• If Metaplane never sees negative values for a monitor, we’ll make sure these are not part of future predictions. If you’re monitoring warehouse costs or file sizes, for example, you should never see a "normal" range that includes negative numbers.