In the Yee/B.Olson diagrams Condorcet methods give quite ideal results. I proposed ages ago that one might study also voter distributions that give cyclic preferences. That would show also some differences between different Condocet methods. I'll try to draft some simulation scenarios.
In a typical simulation there is one "heap" of voters that can move around. Another approach would be to have a smaller heap of voters at the position of each candidate. This can be said to be natural. Every candidate is seen to represent some supporter group or party, and there is a concentration of voters close to each candidate. In order to move the balance of voters in the diagram area one can add one larger and flatter heap of voters to this. Then one can move this larger heap so that the average voter (or median voter) moves around in the diagram (average location should be easier to count but median could be more interesting as a diagram). This way one can draw all the pixels of the diagram. Now to cyclic preferences. We can rotate all the small heaps around the centre of the diagram (few degrees) so that each small heap is no more exactly where the candidates are but next to them. This way one can get also circular preferences. This is still "natural" in the sense that the candidates may well not represent the median opinion of their "own party" but they (or their supporters) may be biased in one direction. (The larger heap can again be moved around to move the average/median voter spot.) One can generalize this style so that in addition to having a set of candidates that can be drawn anywhere on the diagram one could have also limited number of small heaps of voters that can be put anywhere on the diagram (and shown in the diagram using some special marks). This is just one approach to richer modelling of the voter space. The small heaps could typically be of same size to keep the diagram easy to grasp. The larger heap is used just to move the overall balance of the voters. There could be also other ways to move the balance but this one is at least quite simple and understandable. The target is to be able to model also some slightly more complex voter distributions than the basic normal distribution of voters in a two dimensional space. And to be able to make more detailed comparisons, e.g. between different Condorcet methods. Juho ---- Election-Methods mailing list - see http://electorama.com/em for list info