OLAP time dimension - how are the weeks allocated

Question

Could someone please help me understand what logic Oracle uses when one makes use of Oracles OLAP time dimension based on weeks.
I am use Oracle OLAP 11.2
I have a Time dimension that has the following hierarchy:-
YEAR
QUARTER
MONTH
WEEK
For calculating the weeks ID I make use ISO week and year i.e. IYYYIW
For calculating the end date I use the following:- NEXT_DAY( l_date, 'MON' ) -1
i.e. the weeks end date is the Sunday.
According to me this is the required result.
Problem is that for some months there are 3 weeks allocated which makes no sense to me.
I cannot understand the logic used in allocating the weeks.
The following is an example:-
the following weeks were allocated to the month February
201306 (end date= 10-2-2013)
201307 (end date= 17-2-2013)
201308 (end date= 24-2-2013)
but the following week was allocated to January which makes no sence to me,
I would have expected it to be found in February
201305 (end date= 3-2-2013)
Week 201309 (end date= 3-3-2013) was allocated to March which according to me is correct..
Another example is week *201030 (end date= 1-8-2010)* that is allocated to July while I would have expected that it should be August.
I would have thought that it uses the end date that is placed in the mapping to determine the month to place it in.
Could some one please explain what the reason for this could be.

Answer 1

Oracle OLAP model/design (in this case, at least) cannot compensate to coverup existing flaws in the relational model/design.
I dont think this is a valid hierarchy even considering relational model/design.
Weeks do not fit in below Months (calendar months).
You can force fit them by making Weeks the source of truth and making all Months logical Months like "Business Month", "Business Quarter", "Business Year" etc. which will be composed of whole weeks but differ significantly from the calendar definition of Month, Quarter, Year etc.
You are better off modeling time as composed of two hierarchies:
H1: DAY -> MONTH -> QUARTER -> YEAR and
H2: DAY -> WEEK
Alternately if you dont want to introduce DAY level (lower granularity), you can split up the Time dimension into 2 different dimensions. And also modify your star schema to add an additional time dimension key to fact table - one key pointing to WEEK and another pointing to MONTH (independently).
TIME_MO: MONTH -> QUARTER -> YEAR
TIME_WK: WEEK
The fact data will need to be split up to correctly identify the coirrect MONTH-WEEK combination for the relational record.
E.g:
Fact table should have 2 dimension Fks for WK_KEY as well as MO_KEY so that a fact figure of 1000 evaluated over 7 days is split up into 300 from 29-31st of previous month and attached to MO_KEY of previous month and another record of 700 covering days from 1-4 and recorded/tied to same WK_KEY but next month (next month's MO_KEY).
HTH
Shankar