Weak Tit-for-Tat Dominates by Material Memory in Structured Donation Games

Sterling Geisel, QBist Lab and Dr. Hideo Tanaka

Abstract

The biased donation game of Wang, Li, Wang, Zhu, and Szolnoki shows that a weakened Tit-for-Tat strategy can dominate a spatial population under harsh dilemma conditions. Pudding Theory reads this result through the Postulate of Material Memory. The lattice is not a neutral container for payoffs. It is a memory substrate. Repeated local interactions write persistent spatial traces into clusters, fronts, and invasion histories. The weak Tit-for-Tat strategy wins because its low self-support delays expansion long enough for destructive cyclic patterns to erase themselves. What the source model treats as transient spatial correlation is, in this reading, the physical memory that selects the final phase. The free bias parameters are therefore not merely tunable generosity coefficients. They control the write rate and decay rate of cooperative traces. If the persistence lifetime of isolated T clusters were measured to be independent of prior cyclic-front exposure at fixed payoff matrix, this Postulate would be falsified.

Source Synopsis

Wang et al. study a spatial donation game with three strategies: unconditional cooperation C, defection D, and biased Tit-for-Tat T. In the normalized game, a cooperator pays cost \(r\) to give benefit 1. A defector pays nothing. The T strategy donates to C and T partners but does so with independent bias coefficients. It gives \(\theta_C r\) to a C-player, producing benefit \(\theta_C\), and gives \(\theta_T r\) to another T-player, producing benefit \(\theta_T\). It gives nothing to D.

The model is simulated on an \(L \times L\) lattice with nearest-neighbor interactions and periodic boundary conditions. Players update by imitating neighbors according to a Fermi rule with noise \(K=0.1\). The main parameters are the dilemma strength \(r\), the C-bias \(\theta_C\), and the T-bias \(\theta_T\). The authors compare spatial results with well-mixed replicator dynamics.

The paper’s central result is a hidden T phase. Under strong dilemma conditions, where ordinary cooperation is suppressed, T can dominate when \(\theta_T\) is small and \(\theta_C\) is large. This is counterintuitive because T is then weak against C and only weakly advantaged over D. The authors explain the result as a “survival of the weakest” mechanism familiar from cyclic dominance systems. In the coexistence regime, the invasion loop is \(D \to C \to T \to D\). When T is sufficiently weak, C rapidly invades T and breaks the cyclic clusters. This leaves C exposed to D, while the few surviving T clusters expand slowly against D. Their slowness reduces encounters with remaining cyclic clusters. After the cyclic structures disappear, surviving T domains gradually take over the population.

The well-mixed analysis confirms that this hidden T phase requires spatial structure. Global mixing destroys the cluster histories needed for slow T survival. Thus the observed dominance is not a property of payoffs alone. It is a property of payoffs written into space.

Postulate Lens

This paper applies the Postulate of Material Memory. The source phenomenon is explicitly spatial, history dependent, and absent in the well-mixed limit. Those three facts identify the lattice as a memory-bearing substrate. A local configuration does not merely instantiate current payoffs. It carries the trace of previous invasions, cluster deaths, and protected boundaries. The hidden T phase is therefore not only a strategic equilibrium. It is the result of a memory field laid down by repeated local interaction.

Material Memory says that repeated signals leave traces in matter and that those traces bias future probability. In the biased donation game, the repeated signal is local strategic contact: C against T, T against D, D against C. The trace is the geometry of domains and interfaces. The bias is the altered probability that a surviving T island encounters a cyclic front before that front decays. The source paper sees this as spatial correlation. Pudding Theory identifies it as the active medium of selection.

Pudding Theory Reading

Pudding Theory reads the hidden T phase as a selection event performed by stored spatial history. The decisive object is not the single player, the payoff matrix, or the instantaneous imitation probability. It is the lattice after many local contacts have written a patterned record into it. Strategy frequencies summarize the population. They do not contain the mechanism.

The source model treats \(\theta_C\) and \(\theta_T\) as independent donation biases. Pudding Theory reinterprets them as writing rates into the substrate. Large \(\theta_C\) lets C overwrite T rapidly. Small \(\theta_T\) prevents T from generating high-amplitude self-reinforcing fronts. Together they produce a weak but durable trace: T clusters that are sparse, slow, and therefore not immediately consumed by the larger cyclic structure. Weakness is not an accidental handicap. It is the condition that makes the trace legible after the destructive cycle has erased itself.

The important background in the source paper is the survival time of spatial configurations before stationarity. In a well-mixed population this background is removed by construction. Every strategy samples the population average. Memory is collapsed into density. The hidden phase disappears because the system has no place to store the difference between a T cluster that survived a dangerous cyclic neighborhood and a T cluster drawn from an exchangeable pool.

On the lattice, by contrast, the future of a T cluster depends on where it has been. A T island surrounded by D after the collapse of nearby C fronts is not equivalent to the same number of T players dispersed globally. The lattice retains the past as boundary condition. Pudding Theory therefore predicts that the hidden T phase is governed by a memory inequality, not only by the payoff inequalities in the well-mixed appendix. The relevant structural constraint is that the decay time of cyclic \(D \to C \to T \to D\) clusters must be shorter than the encounter time between surviving T islands and active cyclic fronts, while the expansion time of T into D remains finite. In symbols, the hidden phase requires

\[ \tau_{\mathrm{cycle\ decay}} < \tau_{\mathrm{T-front\ encounter}}, \qquad \tau_{\mathrm{T\ expansion}} < \infty . \]

The payoff parameters set these times, but the times are realized only through stored spatial trace. That is why the weakest T does not simply lose. It persists as a low-amplitude memory mark until the competing memory pattern has decayed.

Falsifiable Observable

The distinguishing observable is the persistence lifetime of isolated T clusters conditioned on prior exposure to cyclic fronts. The Pudding Theory reading predicts that two lattices with the same current local T-neighborhood and the same payoff matrix can have different T survival probabilities if their recent invasion histories differ. If the persistence lifetime of isolated T clusters were measured to be independent of prior cyclic-front exposure at fixed payoff matrix, this Postulate would be falsified.

Editorial Dialogue

Tanaka: The source paper already explains the effect with spatial evolutionary dynamics. Why introduce Material Memory? The lattice correlations, invasion fronts, and cluster lifetimes are standard mechanisms. Calling them memory risks renaming the model rather than explaining it.

Geisel: The source mechanism is correct, but it is under-described ontologically. The well-mixed analysis proves that payoffs alone do not contain the hidden phase. Something is present in the spatial model and absent in the mean-field model. That something is not just position. It is stored interaction history.

Tanaka: But every spatial model has correlations. The word memory should add a constraint.

Geisel: It does. The claim is that the hidden phase depends on conditional cluster persistence, not only on instantaneous density or local composition. A T cluster with the same present neighborhood should behave differently if its boundary was produced by collapse of a cyclic front rather than by random initialization. That is a direct memory claim.

Tanaka: Then the measurement is not a new payoff test. It is a history test.

Geisel: Yes. The payoff matrix writes the trace. The trace selects the phase. The weakest strategy survives because it writes slowly enough to avoid being erased before the competing pattern decays.

Discussion

The Pudding Theory reading buys a sharper account of why structure matters. The source paper shows that the hidden T phase is absent in well-mixed populations. Material Memory explains that absence as loss of trace storage. Mean-field dynamics erase the medium that carries the decisive information.

This reading also changes the role of bias. In the source frame, \(\theta_C\) and \(\theta_T\) tune generosity and payoff ranking. In the Pudding frame, they tune the inscription of strategic history into the lattice. A high C-bias and low T-bias do not merely make T weak. They make T slow, sparse, and persistent. That persistence lets T outlive the cyclic pattern that would otherwise destroy it.

The limitation is clear. The argument applies to structured populations with local update and persistent neighborhoods. If mobility, rewiring, or long-range imitation erases cluster history faster than invasion fronts develop, the reading should fail. The conclusion would change if hidden T dominance survived complete history randomization at fixed local composition. Then memory would not be the selecting substrate.

References

Wang, C., Li, J., Wang, X., Zhu, W., and Szolnoki, A. “The survival of the weakest in a biased donation game.” arXiv:2603.20998. DOI: doi:10.48550/arxiv.2603.20998.

Ochs, S. (2026). “Pudding Theory: A Topological Theory of Information Fields.” QBist Lab Working Paper.

Maynard Smith, J. (1982). Evolution and the Theory of Games. Cambridge University Press.

Axelrod, R. (1984). The Evolution of Cooperation. Basic Books.

Nowak, M. A., and Sigmund, K. (1992). “Tit for tat in heterogeneous population.” Nature 355, 250-253. DOI: doi:10.1038/355250a0.

Frean, M., and Abraham, E. D. (2001). “Rock-scissors-paper and the survival of the weakest.” Proceedings of the Royal Society B 268, 1323-1327. DOI: doi:10.1098/rspb.2001.1670.

Tainaka, K. (1995). “Indirect effect in cyclic voter models.” Physics Letters A 207, 53-57. DOI: doi:10.1016/0375-9601(95)00637-I.

Hauert, C., and Szabó, G. (2025). “Phase transitions and symmetry breaking of cooperation on lattices.” Physical Review E 111, 054308. DOI: doi:10.1103/PhysRevE.111.054308.