The gold terrain is an idea-novelty landscape: height h(x,z) is a sum of Gaussian "idea peaks" scattered around a central topic. Peaks near the middle are low (obvious ideas everyone reaches first); the tallest, most original peaks sit farther out, so reaching them takes real exploration:
h(x,z) = Σ A_i · exp(-((x-x_i)² + (z-z_i)²) / 2σ_i²)
Each glowing dot is a simulated student wandering this space. Every step blends a random heading with a gradient-ascent heading toward the nearest peak, weighted by constraint (b): a purely random walk (b=0) covers ground but rarely lands on tall peaks; a fully guided walk (b=1) beelines for the closest peak — often a low, obvious one — starving exploration. Real creativity research (Stokes; Rietzschel & Nijstad) finds a similar inverted-U: moderate constraints tend to raise average originality more than none or too much.
direction = normalize((1-b)·random + b·gradient)
speed = base · (1 + timePressure)
p(idea) = 1 − 0.7 · evaluationFear
Periodically each student "drops an idea" at their current position (sampled as a point on the landscape) unless suppressed by evaluation fear — modelling self-censorship and production blocking observed in real group brainstorming (Diehl & Stroebe, 1987). Three live readouts track Guilford's classic divergent-thinking dimensions:
- Fluency — total ideas generated and the current rate per minute.
- Originality — the average landscape height (novelty) of every idea dropped, 0–100.
- Flexibility — how many of the 24 angular/radial zones around the topic have been touched by at least one idea, i.e. category-switching rather than one narrow theme.
Try it: push constraint to 0% (chaos, high fluency, low originality), then to 100% (tunnel vision, one obvious idea repeated), then find the middle ground that maximises originality without killing fluency — then add evaluation fear and watch all three collapse, exactly as it does in an intimidated classroom.