The term "parallel terraced scan", coined by Douglas Hofstadter, refers to a special type of heuristic breadth-first search where all elements of level n are first subjected to some sort of test to see which are the most promising, and then the elements of level n-1 under these most promising ones are visited.
The parallel terraced scan is thought by some (notably the Fluid Analogies Research Group) to mimic how the human mind solves certain types of problems, e.g. brain teasers or Bongard problems.