The idea to record where the low values of the function f were to be found was initially implemented as follows (Stanford 1968):

Points were sampled in A. The locations of the best p % were recorded by dividing the coordinates of A into equal sized intervals and noting how many points each interval contained. This together with the history of the best point gave hints about where the local minima were to be found. Below a hypothetical result for the Branin problem.

                                    X
        XX          X               X
       XXXX       XXXXX          XXXXX
      XXXXXXXXXXXXXXXXXX   XXXXXXXXXXX                
x1 |---1------------3---------------2--|

        XX                      X
       XXXXX                   XXX
      XXXXXXX                  XXXX
     XXXXXXXXX  XX XXX XX XXXXXXXXXX
x2 |----2-3--------------------1-------|

This was not very practical and lead to the idea to try to directly identify the clusters. I did not then know about clustering techniques and is one of probably many that have rediscovered the technique.