This is not something VA can easily do. I thought I might be able to make this work but unfortunately so far each idea I come up with has ended up me up in the same position you have found yourself in. I hate to say it but I'm thinking this won't be possible.  The closest I can get is having a form of the pivoted table you mention where the forecast and actual values are part of the same column and then a second column that differentiates between the two. THEN I also included columns for each predicted and actual values as well so something like this, along with a calculation I made concatenating the Year and Type to be used for the filter.   Then my custom line chart uses Month, Actual Total, Predicated Total, and grouped by Year, so it looks like this:   Then the filter uses the concatenated item I made, and the lines themselves are right when used, but the legend is wrong and confusing because the legend will always include missing values: I cannot find a way around this or a better solution I am afraid.  ... View more

No problem at all and glad that worked for you! For this second question do you mind starting a separate thread for that? This will be a different solution so would be more helpful in a different thread. AND I believe we can make this work so I'll respond there.  ... View more

I am not sure where JSON helps. You don't say why the number of observations is important so hard to address any detail there.   It might be that processing in a somewhat different order, such as getting a count of the common start/ end pairs before doing the expansion that caused the 35 million rows would be appropriate.   Suppose you COUNT data by start end pairs and get a result similar to this: start end count 2012 2013 103 2012 2014 145 2012 2015 154 2012 2016 234 2013 2014 98 2013 2015 133 2013 2016 199 2014 2015 55 2014 2016 101 2015 2016 88 When you transform that to get single year keep the count with it so it looks like: Obs count year 1 103 2012 2 103 2013 3 145 2012 4 145 2013 5 145 2014 6 154 2012 7 154 2013 8 154 2014 9 154 2015 10 234 2012 The  example data would have 30 rows of data. Which would still be much smaller than just transorming the 103 rows of 2012 to 2013. Summarize that data to total the count by the year and there are only 5 rows of data to actually plot with counts in the 622 to 1119 range (for my given example). ... View more