A Sequoia program is a collection of abstract tasks that do not contain any machine-specific constructs (unless they are external tasks). The abstract Sequoia program is specialized or instantiated for a particular machine (these terms are used interchangeably) at compile-time in a process referred to as mapping the program to the machine.

A Sequoia program expresses an abstract algorithm, while a Sequoia program plus a target-specific set of mapping directives represents a concrete implementation of the program. Multiple different target-specific mappings may be provided, enabling the same Sequoia program to be run on different machines.

Note that the current mapping interface that is described here and that the Sequoia compiler supports is an early prototype, and more convenient interfaces are under active development. Sequoia will always give a programmer as much control as they desire when mapping their program to a machine, and in that sense the current mapping concepts, with a few improvements, can be thought of as a "backbone" of the Sequoia system; this detailed control over mapping will always be present. In addition, however, some partial automation of the mapping process will be supported as a programmer convenience on top of the current low-level interface.