Geoff, thank you very much! I think that IBO doesn't use two cursors, in the
Developer Express help tells :

>The SmartLoadRecords mode works in the following manner:

>On activation of a dataset, the DBTreeList loads records which have a

PARENT value equal to RootValue (Locate the first record whose parent Value
equals RootValue and then fetch and load the >records when the values of the
PARENT field are equal. That is why your dataset has to be sorted by the
PARENT field). Then DBTreeList loads records if necessary - on expand
nodes and when the >DBTreeList has to retrieve the full structure of the
loaded tree node.

I don't want to fetch all records, I want to use this mode. With Paradox by
example, having two indexes the navigation with 100.000 or more records is 0
second and the dataset isn't loaded in memory. It looks that IBO when it has
a index selected uses the IBO 4 techniques to perform the locate fast but
the developer express component calls locates by ID or by PARENT but it need
the table sorted by PARENT always.

I think that you can understand better. I have tested this with TIBOTable, I
don't like this component but it setups the refinament automatically but in
the sql monitor I see like it doesn't user the ID index.

Any help more?

Thank you again.