Here's a little puzzle that's cooking my brains. What can be done to solve?

Question

Guys
Insightful wisdoms herein, the following small problem has landed with me.
Suppose I run a bank and there is a big table of financial transactions. Every credit card therein gains points for every dollar they spend. Keep it simple, 1 point for $1.
A third party company sells gifts in exchange for the points, and we send a file every night of the number of points that must be added to our member's account over at their store.. Joe Schmoe spent $200 today, add 200 points to his account. This part is easy
Problem comes when something is refunded with us (the "bank"), the points balance at the third party CANNOT be decremented (dumb, I know), so that $1000 purchase of a TV netted Joe 1000 points even though he took the TV back a week later and got his money back. Refunds can occur at any time, and you cannot assume that an item will be refunded properly between two runs of the report that prepares the points file (i.e. before sending over the points, you can't add up all the spends, add up all the refunds and send the difference as the points, because some refunds won't have happened yet).
The proposal is that as a result, Joe Schmoe's card should go into "points debt" of 1000 points, and we only start sending new points to the third party once Joe has spent more than $1000 again.. So points updates should be paused and if Joe spends over the next 4 days:
500
300
150
90
then the last transaction for $90 on day 4 clears his outstanding points debt of -50 points and we should send 40 points over to the third party company
If such a problem could be entirely solved within a SELECT, then that's cool but I have development ability on the reports system that runs the SQLs so I can add options to run other query types after, or add support for returning a cursor and implementing this in PL/SQL with lots more advanced logic
Regards
cj

Answer 1

I am unsure of the efficiency of this and it is hard to tell if this is what you need without example data and output, but I think this is what you are looking for.
Suppose we have sample data like this (where negative numbers are normal transactions - debits, and positive numbers are refunds or credits back to the account.
with t as (
select 1 id, sysdate - 5 dte, -1000 amount from dual UNION ALL
select 1, sysdate - 4, -500 from dual UNION ALL
select 1, sysdate - 3, 500 from dual UNION ALL
select 1, sysdate - 2, -250 from dual UNION ALL
select 1, sysdate - 1, -300 from dual)
select * from t;
ID                     DTE                       AMOUNT                
1                      28-FEB-09                 -1000                 
1                      01-MAR-09                 -500                  
1                      02-MAR-09                 500                   
1                      03-MAR-09                 -250                  
1                      04-MAR-09                 -300                  
5 rows selectedFirst step is to get the actual number of points the customer should have at each date.
with t as (
select 1 id, sysdate - 5 dte, -1000 amount from dual UNION ALL
select 1, sysdate - 4, -500 from dual UNION ALL
select 1, sysdate - 3, 500 from dual UNION ALL
select 1, sysdate - 2, -250 from dual UNION ALL
select 1, sysdate - 1, -300 from dual)
select t.*,
       sum(-amount) over (partition by id order by dte rows unbounded preceding) actual
from t;
ID                     DTE                       AMOUNT                 ACTUAL                
1                      28-FEB-09                 -1000                  1000                  
1                      01-MAR-09                 -500                   1500                  
1                      02-MAR-09                 500                    1000                  
1                      03-MAR-09                 -250                   1250                  
1                      04-MAR-09                 -300                   1550                  
5 rows selectedThe next step is slightly more tricky - I am going to use the max() analytic function combined with a sliding window so we can eliminate the reduction in the actual. See below:
with t as (
select 1 id, sysdate - 5 dte, -1000 amount from dual UNION ALL
select 1, sysdate - 4, -500 from dual UNION ALL
select 1, sysdate - 3, 500 from dual UNION ALL
select 1, sysdate - 2, -250 from dual UNION ALL
select 1, sysdate - 1, -300 from dual)
select b.*,
       max(actual) over (partition by id order by dte rows unbounded preceding) curr
from (select t.*,
             sum(-amount) over (partition by id order by dte rows unbounded preceding) actual
      from t) b;
ID                     DTE                       AMOUNT                 ACTUAL                 CURR                  
1                      28-FEB-09                 -1000                  1000                   1000                  
1                      01-MAR-09                 -500                   1500                   1500                  
1                      02-MAR-09                 500                    1000                   1500                  
1                      03-MAR-09                 -250                   1250                   1500                  
1                      04-MAR-09                 -300                   1550                   1550                  
5 rows selectedSo as you can see, this gives the total number of points at each date that should have been sent to the customer.
And so finally, we use lag on that result and find the difference (using NVL to handle the first case):
with t as (
select 1 id, sysdate - 5 dte, -1000 amount from dual UNION ALL
select 1, sysdate - 4, -500 from dual UNION ALL
select 1, sysdate - 3, 500 from dual UNION ALL
select 1, sysdate - 2, -250 from dual UNION ALL
select 1, sysdate - 1, -300 from dual)
select id,
       dte,
       amount,
       curr - nvl(lag(curr) over (partition by id order by dte),0) diff
from (select b.*,
             max(actual) over (partition by id order by dte rows unbounded preceding) curr
      from (select t.*,
                   sum(-amount) over (partition by id order by dte rows unbounded preceding) actual
            from t) b) d;
ID                     DTE                       AMOUNT                 DIFF                  
1                      28-FEB-09                 -1000                  1000                  
1                      01-MAR-09                 -500                   500                   
1                      02-MAR-09                 500                    0                     
1                      03-MAR-09                 -250                   0                     
1                      04-MAR-09                 -300                   50                    
5 rows selectedAnd that should give the points to send at each date.