The same 8-chamber tunnel graph as the 3D scene is laid out flat: each ant is a token performing a random walk across the graph's 11 edges. Arriving at a chamber, it rolls the no-backtrack bias — with that probability it excludes the tunnel it just came from and picks uniformly among the rest; otherwise every connected tunnel (including reversing) is equally likely. This is a first-order Markov chain on a graph, not a shortest-path or pheromone-optimization search — there is no food source and no reinforcement, so the process converges purely from graph topology and the backtrack bias.
Every tunnel is colour-coded by how much traffic it has carried, and the panel tracks the normalized Shannon entropy of the traffic distribution:
p_i = uses(edge_i) / total_transitions
H = -Σ p_i·log2(p_i) / log2(11)
H → 1 means traffic has spread almost evenly across all 11 tunnels (a well-mixed colony); H well below 1 means a few tunnels dominate — raise the no-backtrack bias and watch H climb as ants stop doubling back on themselves and spread further before looping home.
- Chambers — the 8 cavern hubs, sized by how many tunnels meet there.
- Tunnels — colour brightens with cumulative traffic since the last reset.
- Busiest tunnel — the single edge carrying the largest share of all transitions so far.