I wonder whether you would be able to expand a little more on this (very old) post, Pierre. I found it quite useful and I would probably enjoy if you went "on and on" for a few more sentences at least...
Well, that is quite open-ended... SN is build on top of a relational database. Currently JET, but SQL Server is coming very soon (I've got a paying customer who's willing to contribute to change to that back-end).
Each field in SN is actually a query against a table. That way, fields can be added/removed easily. Plus this removes the 255 fields limit that many databases have. Such an approach also greatly reduces collisions, hence making multi-user access less problematic. Finally, with 100's of fields, a regular table would be like a sparse matrix, hence not efficient.
Grids are based on queries (source, filter, datefilter, alphafilter, sort). This allows for filtering/sorting on fields which are not even displayed in the grid. Being based on queries, also allows other apps access to the grid. So you can mail-merge in Word with SN data, Excel can view SN grid data, Access can link to and generate a report on SN data. No need to export, no syncing, this is live database reads.
The UI builds on this data framework. Rich-text, outlines, forms, field equations, inheritance, etc
Hope this helps, I've got to run to pick up my 4 year-old from daycare