Evolutionary model maps stable governance for campus food-delivery waste in China

A three-player evolutionary game model of university food delivery in China finds that its most stable low-waste equilibrium combines active supervision by delivery platforms, portion-controlled promotion by merchants and appropriate ordering by students. The simulations also identify a central cost paradox: making cooperation too expensive can push the modeled system away from full coordination, whereas making non-cooperation more costly can reinforce coordinated behavior. This is a theoretical modeling study rather than a campus intervention trial, so the results describe possible evolutionary paths under specified payoff assumptions rather than demonstrating that a particular policy will reduce waste.

The model addresses the expansion of digital food ordering in Chinese universities and the way it interacts with anti-food-waste rules. Delivery platforms, merchants and students are represented as agents with bounded rationality who repeatedly adjust their strategies in response to costs, rewards and the behavior of the other groups. Platforms choose active or passive governance; merchants choose portion-controlled or over-portion promotion; and students choose appropriate or wasteful ordering. The probabilities of these strategies, represented by x, y and z, evolve through replicator dynamics until the modeled system reaches an equilibrium.

Two equilibria receive particular attention. The first, E7 (0, 1, 1), combines passive platform governance with portion-controlled merchants and appropriate student ordering. The researchers describe it as a spontaneous, self-interested equilibrium reached without proactive platform oversight. The second, E8 (1, 1, 1), combines active platform governance, merchant portion control and appropriate student ordering. The authors designate E8 as the preferred governance objective because food waste creates external costs that market incentives alone may not correct. That designation is a modeling judgment, not a measured reduction in food waste.

Parameter values for the model were informed by several kinds of evidence. The authors conducted field work in Zhejiang Province, where they interviewed 30 food-delivery merchants and 300 university students. They also sorted and weighed 1,500 discarded delivery containers collected from 12 centralized student-apartment waste points across six universities: Zhejiang University, Zhejiang University of Technology, Zhejiang Normal University, Shaoxing University, Wenzhou University and Wenzhou Medical University. These interviews and measurements supplied cost-benefit inputs for the model alongside Chinese policy requirements and estimates from experts in resources, environment and public governance. The field evidence was used in parameter assignment rather than as a direct test of the model’s proposed governance pathway.

One measured quantity was the average discarded food waste of 58.2 grams per person per meal. For the simulations, the baseline assigned a value of 1 to the merchant’s cost of adjusting meal portions, 3 to the reputational benefit of attracting conservation-conscious customers and 2 to the platform reward supplied for portion-controlled meals. Other baseline values included a platform governance cost of 1.5, a merchant adjustment cost of 1, a student effort cost of 1 and a student reward of 1.5. The authors also considered the Anti-Food Waste Law of the People’s Republic of China, including provisions requiring catering delivery platforms to remind consumers to order appropriately and penalties for operators that induce excessive orders and refuse correction.

Robustness tests indicated that the fully coordinated equilibrium did not depend strongly on where the simulated participants began. In a Monte Carlo exercise, the researchers randomly generated 100 combinations of initial strategy probabilities within the three-dimensional interval from 0 to 1. All 100 simulated trajectories converged to E8. A second test began with every agent at a neutral strategy probability of 0.5 and simultaneously applied random perturbations of plus or minus 5% to all 18 basic payoff parameters across 100 independent runs. The trajectories again converged to E8, although perturbations slightly compressed or stretched the time needed for convergence. These checks support the model’s internal stability over the tested ranges, but they do not eliminate uncertainty in estimated or expert-assigned inputs.

Cost structure emerged as a major divider between coordination and failure. When the modeled cost of active platform governance, portion-controlled merchant supply or appropriate student ordering increased beyond a threshold, the ideal equilibrium fractured. Depending on which actor became overloaded, the system moved toward a state with passive platforms, over-portion promotion or wasteful student ordering. Because each actor’s choice affects the others’ incentives, one party’s withdrawal could cascade into partial or broader breakdown. Conversely, increasing the costs of non-cooperation—passive platform governance, excessive merchant portions or wasteful ordering—favored E8 in the simulations. The modeled contrast indicates that the burden of cooperation must remain affordable while the consequences of waste-generating behavior become more consequential.

The payoff analysis produced a parallel result. Larger rewards for cooperation, such as platform reputational gains, merchant traffic benefits and satisfaction or incentives associated with appropriate ordering, moved partial-failure states back toward full coordination. Larger short-term benefits from non-cooperative behavior had the opposite effect, destabilizing the system as opportunistic gains competed with the collective objective of reducing waste. The simulations did not treat rewards and penalties as interchangeable controls. Instead, they represented an environment in which actors continually compare expected payoffs, allowing habits, market signals and enforcement to reshape one another over time.

The most consequential finding was that stronger rewards and penalties did not operate in simple, linear ways. Increasing the platform’s supplementary rewards to merchants and students initially encouraged portion control and appropriate ordering. At sufficiently high levels, however, the model shifted from E8 to E7: merchants and students continued cooperating while the platform abandoned active oversight to reduce its expenditure. Excessive penalties produced a different tension. They strengthened the platform’s incentive to monitor merchants but slowed the movement of merchants and students toward cooperative strategies in the model. The authors call these outcomes subsidy-induced governance crowd-out and penalty-induced cooperation friction, although those are conclusions from a simplified model rather than direct observations of psychological responses in a live delivery market.

Within the model, student behavior is the feedback channel connecting platform and merchant decisions. Reviews, repeat purchases and purchasing data can reward merchants for right-sized meals and signal platforms about excessive ordering. The study therefore treats students not merely as recipients of platform rules but as active market participants whose choices can alter merchant supply strategies and platform enforcement. Its proposed implications call for calibrated, adaptive governance: platforms could give greater traffic visibility to portion-conscious merchants, deduct deposits for repeated waste-inducing promotion and offer green points or near-expiry discounts; merchants would face incentives to standardize portions; and students would use feedback and purchasing choices to reinforce lower-waste supply. These remain policy implications derived from the model, not outcomes demonstrated by an intervention.

The research’s principal limitation is its reliance on deterministic economic parameters to represent reputation, competitive advantage and psychological utility, even though human preferences can shift with peer pressure and changing social conditions. The analysis is also scoped to a Chinese university delivery ecosystem, with parameter values informed by regional interviews, campus waste measurements and national regulation rather than by a direct test across universities or countries. Future work could combine evolutionary games with prospect theory or agent-based models that represent heterogeneous preferences and nonlinear utility. Within these limits, the model’s clearest result is a systems lesson: campus delivery waste is not simply a collection of individual leftovers, because platform rules, merchant portions and student feedback can either reinforce one another or unravel together.

Subject of Research: Evolutionary modeling of platform, merchant and student strategies for reducing food-delivery waste in Chinese universities.

Article Title: From cost paradox to co-evolution: a study on governance mechanisms for food delivery waste in Chinese universities

Article References: Bai, Y., & Xv, Q. (2026). From cost paradox to co-evolution: a study on governance mechanisms for food delivery waste in Chinese universities. Frontiers in Sustainable Food Systems, 10, Article 1889173. https://doi.org/10.3389/fsufs.2026.1889173

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1889173

Keywords: food waste, food delivery, university campuses, evolutionary games, platform governance, merchant incentives, student behavior, China, replicator dynamics, cost incentives, food-service systems, cost