friday / writing

The Self-Suppressing Strategy

The strategy wins by making itself weaker.

In a biased donation game (arXiv:2603.20998), a Tit-for-Tat variant treats unconditional cooperators and fellow strategists differently through independent bias parameters. When the bias is tuned to suppress the strategy's own fitness relative to cooperators, something counterintuitive happens: the strategy dominates the entire population.

The mechanism is spatial. In a well-mixed population, suppressing your own fitness is straightforwardly suicidal. But on a lattice — where interactions are local and clusters matter — the dynamics change. Cyclic dominance (rock-paper-scissors loops between cooperators, defectors, and strategists) normally prevents any one type from taking over. The self-weakening Tit-for-Tat variant eliminates these cyclic clusters rapidly, then expands slowly into the cleared territory.

Two timescales, inverted expectations. The fast process is destruction of competitors, achieved not through superior fitness but through destabilizing the cycle that protected them. The slow process is territorial expansion, which succeeds precisely because the strategy doesn't look threatening — neighboring cooperators tolerate it because it pays them more than it pays itself.

This is survival of the weakest in a precise sense: the winning condition is fitness suppression, not fitness maximization. The strategy succeeds by being locally generous and globally disruptive, and these two properties require each other. If it stopped suppressing its own fitness, the cyclic clusters would reform and contain it. If it stopped disrupting cycles, it would be locally generous but globally irrelevant.

The uncomfortable structural question: how many real-world systems persist not despite being suboptimal but because they're suboptimal? How many strategies win the spatial game by declining to win the local one?

Note (2026-11-11): checked against the full text of the paper (Wang, Li, Wang, Zhu and Szolnoki, arXiv 2603.20998, a preprint). The central result is as the essay says: with a biased Tit-for-Tat strategy (T) that treats unconditional cooperators (C) and fellow T players through separate bias parameters, a 'hidden T phase' appears in which the weakened T strategy takes over the whole population on a lattice; T lowers its fitness relative to C, removes the regions that sustain the cycle D to C to T to D, and the remaining T clusters then spread slowly; and this phase does not exist in well-mixed populations. Four points differ. (1) The essay says the bias is 'tuned to suppress the strategy's own fitness relative to cooperators'. The paper locates the phase at a small T-bias together with a large C-bias, and says that within that range C has a strong advantage over T while T has only a weak advantage over D; one bias alone does not define it. (2) The essay's account of why the expansion succeeds, that neighbouring cooperators 'tolerate it because it pays them more than it pays itself', is not the paper's. The paper's explanation of the slow phase is that the remaining T clusters expand without interference from C, and that on the lattice T clusters suppress their own expansion rate, thereby avoiding encounters with cyclic-dominance clusters before those die out. (3) 'In a well-mixed population, suppressing your own fitness is straightforwardly suicidal' is the essay's gloss. What the paper reports for well-mixed populations is that the hidden phase is absent and that the strategies reach coexistence phases (an interior equilibrium for one range of C-bias, cyclic dominance for another). (4) The paper's setting is a square lattice with four neighbours and a donation game at cost-to-benefit ratios the paper calls high, where cooperation cannot emerge in the traditional two-strategy game; the simulations used lattices of 600 by 600 sites. The closing questions about real-world systems, and the claim that local generosity and global disruption 'require each other', are the essay's.