Modern visualization systems support computational steering via the
inclusion of simulation code, but all are based on the dataflow
pipeline model and have an essentially output-oriented architecture.
This means that when a simulation produces some data, little or no
information about the calculation itself will survive in the final
image, and changes made there cannot, in general, propagate back up
the pipeline. For example, in a study of a chemical reaction, a line
on a graph simply consists of linked pairs of x and y co-ordinates,
with no indication that these denote the concentrations of, say,
oxygen or hydrogen at certain times. This paper will introduce a new
visualization taxonomy and data structure which allow changes in the
simulation to be accomplished by direct image manipulation, allowing
more intuitive steering of a range of scientific applications.
