The scene in 30 seconds

Watch the 30-second Pikon World demo

The demo shows boats, a storm, territory conquest feedback and a stone wall being built. It uses a local development fixture.

One polygon model, several uses

Each island has 24 clipped Voronoi territories. The Dart model creates a seeded set of sites using farthest-point sampling, then clips the island's convex coastline against the half-planes between each pair of sites.

For sites A and B, the boundary is their perpendicular bisector:

normal = B - A
limit = (B dot B - A dot A) / 2
keep P where P dot normal <= limit

The clipping helper computes the signed distance to that boundary. This is the edge-intersection part from the current Dart implementation:

final a = polygon[i], b = polygon[(i + 1) % polygon.length];
final da = side(a), db = side(b);
if (da <= 1e-10) result.add(a);
if ((da < -1e-10 && db > 1e-10) ||
    (da > 1e-10 && db < -1e-10)) {
  final t = da / (da - db);
  result.add(LandPoint(
    a.x + (b.x - a.x) * t,
    a.y + (b.y - a.y) * t,
  ));
}

Here side(P) is P.x * normal.x + P.y * normal.y - limit. The small tolerance handles points close to the clipping plane. This implementation assumes a convex island footprint.

The scene reuses these territory polygons for the land surfaces, picking and the fortification perimeter. A fortification change can therefore animate construction along the same boundary that defines the territory. Keeping these uses on one model makes it easier to keep visual boundaries and interaction aligned.

Working on the world independently

The world lives in its own package with an example app, so scene behavior can be developed independently of account and multiplayer state. Mobile rendering also has its own quality policy: an initial memory-based preset, followed by quality reductions when frame pacing drops.

I'd be interested in how other Dart developers share geometry between rendering, picking and construction animation.

Try Pikon

• iOS — App Store
• Android — Google Play

Disclosure: this development note was drafted by an AI assistant from the project's current implementation.