Its a simple idea.... kind of like a sql-view layer on top of stored data in sql-tables.
I will just type few points quickly. If you don't understand it, feel free to post more questions.
Lets say you have the following stored cubes where data has been loaded and aggregated.
CUBE1 (with stored measures meas1x, meas1y, meas1z) dimensioned by DIM1, DIM2, DIM3, DIM4, TIME
CUBE2 (with stored measures meas2x, meas2y, meas2z) dimensioned by DIM1, DIM3, DIM5, DIM6, TIME
CUBE3 (with stored measures meas3x, meas3y, meas3z) dimensioned by DIM1, DIM2, DIM3, DIM6, TIME
Now you create a reporting cube, lets call it RPT_CUBE
RPT_CUBE should be dimensioned by all the dimensions. That is important point.
It will only have calculated measures, NO stored measures.
Create one base (calculated) measure in RPT_CUBE which points to one stored measure.
So for the above 9 stored measures you will have 9 calculated measures in RPT_CUBE.
All the other measures (like time-series measures) will be created using these base measures in RPT_CUBE.
LOOP_VAR and LOOP_DENSE property settings are critical for each measure in RPT_CUBE. Often times it is correct, but I have seen cases, when it is not defined correctly, so I had to manually fix these two properties for measures in my RPT_CUBE.
When querying measures in RPT_CUBE, make sure that WHERE conditions are present for all dimensions.
Set the dimension to topancestor of a hierarchy, if it is not needed in the query.