Each ant is an agent on a 6 m x 6 m forest floor. It senses the pheromone field P at three points 0.12 m ahead (left, centre, right, 0.6 rad apart) and turns toward one of them with the Deneubourg choice function
p_i = (k + P_i)^n / sum_j (k + P_j)^n k = 1, n = 2
dP/dt = D laplacian(P) - (ln2 / T_half) P + deposits
deposit: 0.35 /s outbound, 1.4 /s when carrying prey
Outbound ants speed up on a trail (0.12 m/s vs 0.05 m/s off-trail), pick up prey within 7 cm and carry it home at 0.09 m/s with a homing turn plus trail following. An outbound ant that stays off-trail for 45 s gives up and walks home empty. Positive feedback: more prey found, more pheromone, more ants follow, until the patches run out and the trails evaporate.
Simplifications, stated plainly: real Eciton raids are driven by contact interactions and a wide swarm front, not by a single scalar field; this model reproduces the trail-recruitment loop, not the swarm front's fan geometry. Ant bodies are drawn about 10 times their real length so they are visible; speeds, deposit rates and the 45 s give-up time are plausible round numbers, not species measurements; prey needs one ant, there is no cooperative carrying, no day-night cycle and no bivouac. The forest props are static scenery outside the 6 m arena.