I think code in the SCAN loop can be shortened to

SCAN
  INSERT INTO OutTable ;
    SELECT TOP 20 PERCENT * FROM MyTable WHERE MyTable.cgroup = UCGroup.cgroup ORDER BY iscore
ENDSCAN

>The help system only states that TOP N is supported in a Non-Correlated Subquery, leaving one to assume that it is not supported in a correlated subquery.
>
>Unfortunately from what I can see you need a correlated subquery to do this in a single SQL statement, but even if that was available I think it would be very inefficient. So in my opinion you would be better off to do something like this:
>
>

>create cursor mytable (cgroup c(1), iscore i)
>SELECT * FROM  mytable INTO CURSOR OutTable READWRITE
>
>FOR X = 1 TO 20
>  cX = CHR(64+X)
>  FOR Y = 1 TO 100
>    INSERT INTO mytable (cgroup,iscore) VALUES (cX,Y)
>  ENDFOR
>ENDFOR
>
>SELECT cgroup FROM mytable GROUP BY cgroup INTO CURSOR UCGroup
>SCAN
>  SELECT TOP 20 PERCENT * FROM MyTable WHERE MyTable.cgroup = UCGroup.cgroup ORDER BY iscore;
>    INTO CURSOR Temp
>  SELECT OutTable
>  APPEND FROM DBF("Temp")
>  USE IN Temp
>  SELECT UCGroup
>ENDSCAN
>USE IN UCGroup
>

>
>In the above example the query runs 20 times, an unoptimized correlated subquery if available would run the query 2000 times. So something like this seems more efficient to me.
>
>