Your goal is to place your political platform so as to win the election. You also get to choose what type of election you want: Min or MaxVoting and variations of those.
You will not know how many strategic voters, non-voters, or voters who will choose not to give two candidates the same score there will be. Neither will you know how many candidates there will be. But you can choose rational or random competing candidates.
When you understand how the voting category, MinVoting and MaxVoting, affects the location of the winning platforms in relation to the majority opinion of the voters (MO) you will understand how we can end our political division.
What relevance does political party have if all the viable candidates must move toward the Majority Opinion of the voters?
Each AI candidate solves the same question: “Given the voting rules, the voter distribution, the other candidates’ likely positions, the estimated number of strategic voters (0-30%), and the potential for a new candidate to enter the race, where should I stand to win?”
This is how the AI candidates reason in this simulation:
Each candidate has an ideology in the range -1 (far left) to +1 (far right). Candidate A always leans left (-1), Candidate B always leans right (+1), and any additional candidates have ideologies randomized for each election. This gives every candidate a brand identity that voters can recognize.
Each candidate assumes there are more candidates than they can see. Even if only three appear on the ballot, every AI reasons about a hypothetical additional entrant who could appear if territory is left uncovered. This keeps them from collapsing toward the same column, because doing so would invite a challenger to capture the abandoned space.
Each candidate treats the player as a fixed peer. They know a player is in the race, they just don’t know which column the player will pick. They assume the player will play to maximize the player’s own chance of winning.
Each candidate estimates the strategic voter rate at 0% to 30%. When the rate is hidden, they pick a random value in that range (in 5% increments) and play their best response to that estimate. Real candidates don’t know exactly how many voters will bullet-vote — they reason about a plausible range.
Each candidate prefers winning over getting a higher score. Candidates evaluate every possible column on a three-tier scale:
A position that scores 1000 but loses to a stronger opponent is rejected in favor of a position that scores 800 but wins. Candidates prefer Tier 2, then Tier 1, then Tier 0 — and within a tier they prefer the higher score.
Each candidate iteratively responds to the others until no one wants to move. All candidates (real AIs, the assumed player, the hypothetical entrant) repeatedly recompute their best move based on where the others have just landed. This continues until the field reaches a stable equilibrium (no candidate wants to change position) or a cycle is detected.
What to take away: The same set of rational candidates, with the same uncertainty about strategic voters, the same threat of entrants, and the same voter distribution, play very different strategies based on the voting rule. MinVoting forces them apart to defend territory. MaxVoting pulls them together because there’s no penalty for proximity. That difference — emerging from rational behavior, not imposed by design — is what these simulations let you see.