We will monthly receive a CSV file with about 3 million records of finacial transactions generated by a third party using a bookkeeping program we can not control.

This is how the records in the file will look like:
Explanation:
B: means the amount was booked
A: means the amount was annulated, cancelled, all the data of the original booking is copied, only B becomes A

- the amount due of invoice nr 12 is $1000.
- invoice nr 13 is cancelled totally, it is useless to import it into the database.
- the amount due of invoice nr 14 is $3000, once it was $2000, but that was an error and it was cancelled, then it was changed to $2500, but that was cancelled too, the last thing we know is that it is now $3000.

This should be imported into our database:
The records are stored ordered by the invoice_nr and the time the transaction was performed by the bookkeeping program.

The logic is: import only the last record of a group of records (grouped by invoice_nr), if its booking/annulation flag is set to 'B'.

Can we put some logic into the LOAD command to make it perform the filtering for us ? From the test data we received, a relatively large portion consists of booking/annulation/booking and booking/annulation.

AFAIK, you cannot ...

An option may be to load the data into a 'temp' table and then use a CURSOR(which is defined with the filter criterion) to load data into the main table

Sathyaram