It's an interesting problem. During an upgrade the agg_schedule table, which controls the summary tables, gets reloaded, so there are two copies of the schedule for building the tables. That means that the summary tables get loaded twice going forward - with the same data. Since the daily table is built from the 15 minute table, it will have not only duplicates caused by the scheduler, but it's counts will be 2x what they should be. Likewise the weekly and monthly tables, which are built from the daily table, will have duplicates and counts which are 4x what they should be. TAC should be able to help you with this. The solution is:
0) turn off the summary job in the windows scheduler while you do this, turn it back on when done.
a) ensure that the agg_schedule table in the ciscoadmin database only has 1 row for each summary (10 rows total)
b) rebuild the 15 minute tables, (select distinct * from call_15 into temp t1 with no log; truncate call_15; insert into call_15 select * from t1; drop table t1). That might not work with the application_summary_15 table as it can be very large, you might have to unload that to a file (unload to "myfile" select distinct * from application_summary_15;...... load from "myfile" insert into application_summary_15;
c) You can do the same thing for the daily, weekly and monthly tables. If you can figure out when the duplication started, you can update their counters to be 1/2 or 1/4 of what they are.
Not something for the feint of heart. Call TAC.