1:many relationship

I have a situation where Saiku sometimes displays an unexpected/undesirable value, until I re-run the exact same calculation.

It is where a fact table has a 1:many relationship with a dimension table.


For example;
- Each fact is a sweetie purchase
- The sweetie-preferences dimension table can have multiple records per customer
---> Customer1 prefers Chocolate and Toffee
---> Customer2 prefers Toffee and BonBons
---> Customer3 prefers BonBons and Chocolate
-----> Each Customer has 2 records in the sweetie dimension table

If I look at sales by sweety preference; each customer (correctly) contributes their sales to two different catgories.
---> Customer's sales appear included within the "Chocolate" row AND the "Toffee" row

Code:

---

Preferred Sweeties | Sales Value
--------------------+-------------
BonBons            | 8.0  (3.0 + 5.0)
Chocolate          | 5.0  (2.0 + 3.0)
 Toffee            | 7.0  (2.0 + 5.0)

---

Then, if I remove the preference dimension from the report, the results include multi-counting (As if Mondrian expects that it can simply SUM() across the three sweeties to get the overall aggregated value).

Until I refresh the exact same query again, then it gives the 'correct' values.



Is this 1:many relationship something supported by Mondrian? Or have I just stepped over a line that I didn't even know existed?



Cheers,
Mat.