Each of the 90 nodes is a simulated speaker placed on a 2D plane and linked into a k-nearest-neighbor social graph (4 base ties each), then partially rewired into random long-range ties — a Watts–Strogatz small-world network. Social network reach sets the rewired fraction: 0% keeps ties purely local (an innovation crawls outward from its origin); higher values let it leapfrog across the population via "weak ties".
f_i = (adopting neighbors of i) / (neighbors of i)
p_i = f_i ^ (1 − prestige/100) // adoption chance this tick
p_revert = (1 − f_i) ^ (1 + prestige/100) · (−prestige/100) // only if prestige < 0
Positive prestige makes the exponent < 1, which is superlinear ease of adoption — a small minority of neighbors is enough to tip a speaker, producing the classic S-shaped curve tracked on the right. Negative prestige makes adoption harder and lets adopters revert, so the variant can shrink back to zero even after a strong start.
- Network pane — amber nodes use the new variant, slate nodes the old one; lines are social ties along which contact happens each generation.
- S-curve pane — adoption percentage plotted against generation number; a genuine diffusion run traces the classic S shape, a suppressed one flatlines near 0%.
- Peak growth/gen — the largest single-generation jump in adoption seen so far, i.e. how steep the middle of the S-curve got.