feat(*): Fix exported-policy action order to gymnasium's sorted-key layout, add wall-contact penalty and stronger ball-touch reward, and wire Spectate to run01 vs run02

Game/scripts/ai_ship_controller.gd

@@ -49,14 +49,15 @@ func get_action() -> ShipAction:
 func _decide() -> void:
 	var obs := ShipObservations.build(_ship, _opponent, _ball, _attack_goal_position)
 	var out := _policy.forward(obs)
-	# Output layout matches the flattened training action space (Box(7)):
-	# thrust xyz, rotation xyz, turbo (> 0 means on).
-	_action.thrust = Vector3(
+	# Output layout is the trainer's flattened action space (Box(7)), which
+	# gymnasium orders by SORTED key name — rotation xyz, thrust xyz, turbo
+	# (> 0 means on) — NOT ShipAction's thrust-first declaration order.
+	_action.rotation = Vector3(
 		_axis(out[0]),
 		_axis(out[1]),
 		_axis(out[2])
 	)
-	_action.rotation = Vector3(
+	_action.thrust = Vector3(
 		_axis(out[3]),
 		_axis(out[4]),
 		_axis(out[5])

Game/scripts/ship_ai_controller.gd

@@ -15,9 +15,14 @@ extends AIController3D
 # Reward shaping weights. Dense terms accrue per physics tick (60 sim-ticks
 # per sim-second); event terms fire once. Exported so tuning needs no code
 # edits. Goal rewards are added by TrainingMode, which owns goal events.
-@export var ball_touch_reward := 0.1
+@export var ball_touch_reward := 0.25
 @export var velocity_to_ball_weight := 0.001
 @export var ball_velocity_to_goal_weight := 0.004
+# Per-tick penalty while touching the arena enclosure (walls/floor/ceiling).
+# At 60 ticks/sim-second this is -0.3/s: a ship parked on a wall for a full
+# 30 s episode loses ~9 — comparable to conceding — while a brief graze
+# costs almost nothing.
+@export var wall_contact_penalty := 0.005
 
 var ship: Ship
 var rl_controller: RLShipController
@@ -84,6 +89,15 @@ func _physics_process(delta):
 		var ball_progress := ball.linear_velocity.dot(ball_to_goal.normalized())
 		reward += ball_velocity_to_goal_weight * ball_progress / ShipObservations.BALL_SPEED_SCALE
 
+	# Dense penalty: every tick spent in contact with the arena enclosure
+	# (contact monitoring is already on for the ball-touch reward). Ships
+	# bumping each other or the ball is fine — only the boundary counts.
+	if wall_contact_penalty > 0.0:
+		for body in ship.get_colliding_bodies():
+			if body is ArenaBoundary:
+				reward -= wall_contact_penalty
+				break
+
 
 func _on_ship_body_entered(body: Node) -> void:
 	if body.is_in_group("ball"):