You are absolutely right about the fact that it is madness to launch a query which would load 7000 items. I was just trying the limits of the system. Furthermore, I am coming from a Java world where we use a relational db every day; there I'm used to filter out of millions of entries some objects you need, and this relational database is actually very fast at that. The point you make about linking the TODO's to a specific user is off course the way you would do it in an object database using a rich domain. The idea about moving TODO's from one collection to another collection (the history) is also sound. Instead of changing the status field on a TODO (and filter on that status), just move it to the history. Like this, you never end up with a collection of many thousands of items. Just need to think more OO when modeling the domain. And don't care how it gets persisted :-).

The 'toy' projects I've created with Magma don't have a rich domain. But on my job we're investigating Smalltalk/Seaside & we've decided that we will use Magma as a database. We are building a kind-off bigger POC. (http://www.squeaksource.com/SunnysidePlanning2).

If I understand you correctly on this, it means that even if the sorting for instance says: 200/2999, that if I read the first 20 items, these might still change afterwards, so I should warn the user that these are preliminary results. Or offcourse use #sortedBy:.