Each new prototype we developed engendered new interface ideas. Moreover, each prototype brought change to the way that the neuroscientists carry out research. The new interfaces are tools that let neuroscientists learn more about brains, and are not burdens. Indeed, now that browsing is easy we no longer think that the collection of images is large, but rather are looking to fill in the gaps in the ``globally sparse, locally dense database.''
Depending on how you count, we produced between a few dozen and a hundred prototype versions. Had we expended any effort optimizing the code for any version we would have been wasting our time - the code was destined for an honored burial in the scrap heap. Actually such effort would have been worse than a waste, since we quite naturally would have been loath to discard what we slaved over, and would have resisted change. We did address an evident bottleneck with wavelets, but even there we just modified existing code.
Already, the fifth iteration is completing and we are heading towards a sixth, a seventh, and many many more. With each iteration, we stretch to find the tools we need to support our users. Just as the first four iterations brought us to Tcl/Tk, TclProp, distributed constraints, and wavelets; the next few iterations will lead us to discover or invent abstractions and technologies. And, with each iteration we lift our users just a bit higher, expanding their horizons, and ensuring another iteration of improvement. The story doesn't end here, it is still being written. We wouldn't have it any other way.