I worked through this issue once. The project specified a tree - but there was a bunch of data. I think my solution was use the "expand" event on "cluster" nodes to read in data that was not initially loaded. The cluster nodes the user never opened were empty throughout the session.

All the data should be accessible - and I am sure the interface and files are laid out so that the necessity and expense of searching or browsing through open or closed "orders" is minimal.

Sounds like an "industrial" app. I have had 2000 orders/day projects (pick lists - packing lists - invoices - returns and accounts receivable). Since I moved into mostly energy and finance projects - the total transactions are smaller - but the dollar amounts per transaction are much much bigger.

It feels good to know that users are tickling keys to drive a program we wrote that manages some of the more sensitive aspects of their operations.

>My apps work primarily on local VFP tables. So when the user opens a form, he's opening a buffered table for the main table (sales orders header, for instance) and a buffered view that displays the sales order line items from the currently selected sales order. Click on the list tab and he can find any sales order in the table. It may be a table with 1000 sales orders or 10,000 or 100,000. They are all accessible. And why not?
>
>>Who in the world would scope 50 THousand records to a user interface?:-) That rediculous - irregardles of the containing control.