flutter3d_game_strategy 0.8.0

A fourth genre: ground made of samples, units that move as a crowd rather than one at a time, and orders given to a selection instead of to a body.

flutter3d_game_strategy

A fourth genre: the ground is made of samples, units move as a crowd instead of one at a time, and orders go to a selection instead of to a body.

final sim = StrategySimulation(ground: heightfield);
for (var i = 0; i < 500; i++) {
  sim.add(Unit(position: Vector3(i % 25 * 1.5, 0.0, i ~/ 25 * 1.5)));
}

for (final unit in sim.units) {
  unit.order = UnitOrder.moveTo(Vector3(60.0, 0.0, 60.0));
}
sim.step(1.0 / 60.0);

Why a strategy is the fourth genre

The three genres before it each move one body under a camera attached to it. When they agree about what an engine owes a game, that agreement says less than it seems to, because all three have the same shape. A strategy game has the other shape: a camera over a map, orders given to a selection, a crowd in place of a hero. It is also the first game to use three things the engine already had and nothing had exercised yet: instanced meshes, the picking pass, and the flow fields in the navigation.

What is measured rather than assumed

Question	Answer	Where it came from
What does a crowd cost to step?	10 000 agents in under 1 ms	descend 256 µs, separation 717 µs, transforms 19 µs
What does a crowd cost to draw?	50 000 hold a 120 Hz display	the ceiling above that is the CPU encoding the frame, ~0.07 µs an instance
What does an order cost?	0.52 ms on a two-metre grid	8.4 ms on a half-metre one, for the same walk

The last row is why this package bakes a coarser grid than a shooter does. An order pays for the field, and the units walking it cannot tell the difference.

What is here

A Unit is a record of numbers, not a body. A UnitOrder points somewhere or nowhere. Each step walks the crowd over a Heightfield: descend a shared field, push overlapping neighbours apart, settle back onto the ground. The NavGrid under all of that is baked at two metres and re-baked whenever a Building takes cells out of it; a building is a footprint, not a collider. Placing a building also moves any unit standing under it, because a unit left in a cell that no field can reach stops walking for the rest of the match without any error.

On top of the walking there is an economy. A ResourceNode is what the map holds. A HarvestJob is the loop a worker runs when nobody is giving it orders: out, dig, home, drop. A Stockpile belongs to a side, not to the map, and a Producer turns a pile back into units.

FogOfWar keeps two states on a lattice of its own. A cell is explored once anybody has been near it, and stays explored for good; it is visible only while somebody is near it now. The fog is part of the rules as well as the picture. The policy that decides where to send workers queries it, so a side that has not found a seam cannot dig it and has to send somebody to look, and the rendering draws the same lattice the rules read. Bot plays a side through the same handles a player has, and Match runs two bots to an economic win, or to a draw when neither side got further than the other.

Around the simulation, MapCamera watches a place instead of a body. It still drives CameraRig, so it gets the rig's smoothing and first-frame cut without a copy of its own. Selection finds units with a ray the application unprojects. Formation arranges a squad at its destination instead of giving every member its own goal, which is why twenty orders do not cost twenty fields. StrategyVisuals is the one file in the package that draws.

What is not here yet

There is no combat yet. A package that added it before it could carry a crowd across a hill would have been guessing about the part that was actually uncertain. Every unit is also the same kind: there are no types with different reach or armour. Nothing here can be snapshotted, recorded or given a seeded random source, so a run can be watched but not yet replayed or rolled back.

Sight is a radius, not a line through the terrain, so a ridge hides nobody. Casting a ray per cell per source is the one cost in this genre that grows with both the crowd and the map, and nothing has needed it yet. There are exactly two sides, each with one stockpile and one delivered total. That is enough for one policy playing itself, and not enough for a game.