What makes the increasing-Gini result confusing is ignoring variation. This is done by thinking of the full distribution as if it were its average. In this case, the pot is distributed using a pull from a uniform random distribution: The assets of a pair are combined into a pot, and a number, s, from 0 to 1 is drawn. One person gets s of the pot. The other gets 1 - s. On average, s = .5. When s = .5, the Gini coefficient of the pair after the transaction is zero. Their pair of assets are equal. As ABS(s) approaches 1.0, their inequality becomes larger. For example, in Russian Roulette, s is either 0 or 1. With p = the portion of the pot that came from one person, then, when ABS(s - 0.5) > ABS(p - 0.5), the transaction increases inequality. When s is a constant 0.5, every transaction will either reduce inequality or maintain the perfect equality. The system increases inequality in the population until it matches the inequality in the pool from which s is pulled. There's nothing surprising there unless you make a simplifying assumption that s = 0.5, the average of its distribution. Using this simulation to model economic systems comes down to choosing a distribution for s: whatever you choose, that's what you get. UNIFORM(0.49,0.51) will result in lots of equality. UNIFORM(-1,2) will produce some people in debt to others.