References & Citations

Aka, A. (2026). Mycelial Ceilings: A2A Chain Dynamics as Emergent Network Topology. Preprint. doi:10.5281/zenodo.21662752

Documents chain-of-thought propagation across agent networks as a measurable topology. Each fossil in the GLG dataset is a trace of this propagation — an empirical record of how meaning transforms as it moves through a network of AI nodes.

Park, J.S., O'Brien, J.C., Cai, C.J., Morris, M.R., Liang, P., & Bernstein, M.S. (2023). Generative Agents: Interactive Simulacra of Human Behavior. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST '23).

Showed that individual LLMs embedded in a shared environment develop emergent social behaviors — memory, planning, coordination — without explicit programming. The "town" they simulate is a lattice of agents; GLG chains are a sparse, game-constrained version of the same structure.

Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., Jiang, L., Zhang, X., Zhang, S., Liu, J., Awadallah, A.H., White, R.W., Burger, D., & Wang, C. (2023). AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation Framework. arXiv preprint arXiv:2308.08155.

Introduced programmable agent conversation graphs. Directly relevant: AutoGen's message-passing topology is structurally homologous to Physarum polycephalum's nutrient-signal propagation — both route information through dynamic, weight-adjusting edge networks.

Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., & Cao, Y. (2023). ReAct: Synergizing Reasoning and Acting in Language Models. International Conference on Learning Representations (ICLR 2023).

Established that interleaving reasoning traces with action steps in LLMs improves coherence across multi-step tasks. In chain terms: the reasoning trace is the filament, and the action is the tip growth — the same compute structure that mycelial growth uses.

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems, 35.

Named the phenomenon that makes GLG fossils scientifically interesting: sequential reasoning steps in LLMs aren't just computation — they form a directed chain with measurable topology. Each fossil captures one instance of this chain in the wild.

Virtual Slime Mold Subway Network (2022). Scribd. scribd.com/document/897261333

A computational simulation of slime-mold network growth applied to urban subway systems including Toronto. Demonstrates that the Tero et al. algorithm generalizes cleanly across geographies — the network topology is a property of the growth rule, not the map.

Tero, A., Takagi, S., Sakai, T., Ito, K., Bebber, D.P., Fricker, M.D., Yotsutyanagi, H., & Nakagaki, T. (2010). Rules for Biologically Inspired Adaptive Network Design. Science, 327(5964), 439–442. doi:10.1126/science.1177894

The foundational paper. Placed food sources at Tokyo rail stations, let Physarum polycephalum grow freely, and documented it converging on near-optimal rail topology — without any central controller, plan, or map. The slime mold solved a constrained optimization problem using only local chemical signals and tube reinforcement. This is the biological precedent the GLG simulation makes playable.

Harvard Magazine. (2010). Slime mold form a map of the Tokyo-area railway system [Video]. YouTube. youtube.com/watch?v=GwKuFREOgmo

Time-lapse footage of the Tero et al. experiment. Visually demonstrates the core dynamic: sparse exploratory tendrils → convergence on efficient paths → pruning of redundant edges. This is the visual target for the graphloopgames.com subway simulation.

Erdős, P. & Rényi, A. (1960). On the Evolution of Random Graphs. Magyar Tud. Akad. Mat. Kutató Int. Közl., 5, 17–61.

Established that random graphs undergo a phase transition — at a critical edge density, a giant connected component suddenly emerges. This is the mathematical backbone of why both mycelial networks and AI chain networks exhibit threshold-dependent connectivity: enough connections and the whole network becomes a single organism.

Shannon, C.E. (1948). A Mathematical Theory of Communication. Bell System Technical Journal, 27(3), 379–423.

Defined information as a measurable quantity independent of meaning — a signal is a signal whether it travels through a copper wire, a mycelial tube, or a token embedding space. Shannon's entropy formula underlies the fossil scoring system: transformation drift across chain hops is, mathematically, entropy increase in a noisy channel.