Given we could do it over again, what would we do differently?

Truthfully, the only difference would be to generate a graphical output tool, which allows the intermediate work performed by HORE to be viewed, early in the design phase. HORE currently produces real-time movies of wavefront expansion or ripup and reroute passes. This graphical output was crucial to debugging and optimizing HORE; however, it was considered a low priority and not fully implemented until after we "thought" the router complete. Having graphical output early in the design phase would have allowed for more rapid debugging and facilitated the task of optimization.