I plead some ignorance as to the way indices are organized. Please go a little further.
One thing is very clear if you work over a slow connention: Rushmore only reads the tinest portion of the CDX file, and determines the offsets to the records that match the expression. Having those pointers, it fetches the records. The magic is that it caches smart and if it does have to read the CDX, it must be very good at estimating that right chunk to read.

>My first Foxpro Advisor article referred to the work of Tom Lewinson who theorized that Rushmode builds a bitmap where each record matching the condition is represented by a 1 and the non matching records by a zero. This bitmap would have to be in physical record sequence. Any other sort order will make it impossible to just jump from record to record.