Not my intention to take anything away from the winner (won by
@corwin42 way back in Oct 2011) but it seems as if the attached project solves the same puzzles 3 to 4 times faster than that of the winning project. Project is based on the Dancing Links algorithm of Donald Knuth. It uses inline Java code - code is very short and can very easily be converted to B4A. Have not tried to add any optimization to the project - just adapted it slightly to fit B4A project. Competition puzzles have been hard coded in the project (so Sub Activity_Resume does not look the same) but points of time measurement (start/finish) are however the same.
CORRECTION EDIT: change this line
Log(Round((DateTime.Now - starttm) / 9 ) & "ms (per puzzle)")
to
Log(Round((DateTime.Now - starttm) / 7 ) & "ms (per puzzle)")
Still at least 2 to 3 times as fast when measured on the same device...