Game theory usually assumes that followers are rational -- that they choose strategies maximizing their payoffs given what the leader has committed to. Sam Ganzfried's new framework, the evolutionarily stable Stackelberg equilibrium, replaces this assumption with something older and stranger: followers adopt strategies that survive invasion by mutants. The leader -- modeled as a physician choosing a treatment protocol, or a regulator setting policy -- optimizes against a population whose behavior is shaped not by calculation but by evolutionary stability. The resulting equilibrium can differ sharply from the standard Stackelberg solution, because an evolutionarily stable strategy in the follower population need not be a best response in the classical sense.
The structural insight is that the leader's power actually depends on the followers being governed by dynamics the leader cannot directly command. In a standard game, the leader announces a strategy and followers rationally comply; the leader's advantage comes from commitment. Here, the leader's advantage comes from prediction of a process that runs on its own logic. The evolutionary dynamics are not a simplification of rationality -- they are a different kind of regularity, one that makes the follower population's behavior more constrained in some directions and less controllable in others. The leader must optimize against a population that is stable in a specific, invasion-proof sense, and this stability narrows the set of reachable outcomes while making those outcomes more robust. In the cancer treatment application, the physician cannot instruct cell phenotypes to adopt a strategy; she can only shape the selective landscape and rely on the evolutionary process to settle. The leash works precisely because it is not a chain.
Note (2026-11-10): I checked the abstract (current and first version) and searched the full text of arXiv 2603.18385 (Ganzfried). The setting and the two selection rules match, though the essay does not name them: the current paper considers leader-optimal (OSESS) and leader-pessimal (PSESS) selection among ESSs, where the first version had follower-optimal. The essay says the resulting equilibrium “can differ sharply” from the standard Stackelberg solution. In the paper's cancer-treatment experiment (its Table 4) it reports that the optimal physician strategies vary slightly between the two solutions and the optimal quality of life is slightly higher for OSESS than for the standard Stackelberg solution. The essay's regulator setting policy is not in the paper, which names only the physician and cancer cell phenotypes, and the line about the stability making outcomes more robust is the essay's gloss.
(arXiv:2603.18385)