fix(*): apply the locomotion mask during in-game/eval inference, not just training

AIShipController (eval + real gameplay) ran the raw policy output unmasked
regardless of allow_vertical/allow_pitch_roll, while ShipAIController
(training) correctly discarded those axes for grounded curriculum stages.
A grounded-trained model's untrained vertical/pitch-roll output reached the
ship as noise during eval, understating it against models that were never
handicapped this way.

Game/scripts/ai_ship_controller.gd

@@ -17,6 +17,16 @@ extends ShipController
 # Uniform noise magnitude added to each action axis (0 = play at full skill).
 @export_range(0.0, 1.0) var action_noise: float = 0.0
 
+# Must mirror whatever the model was actually trained with (see
+# ShipAIController's identical exports on the training side, curriculum
+# stages 1-2 in TRAINING.md). A model trained grounded (mask on) never got a
+# reward gradient on these axes, so its raw output there is untrained noise —
+# leaving this true for such a model doesn't make it fly well, it just lets
+# that noise reach the ship instead of being discarded like it was in
+# training. Set false to match a grounded-trained model's actual behaviour.
+@export var allow_vertical := true
+@export var allow_pitch_roll := true
+
 var _policy: PolicyNetwork
 var _action := ShipAction.new()
 var _ticks_until_decision := 0
@@ -53,13 +63,13 @@ func _decide() -> void:
 	# 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[0]) if allow_pitch_roll else 0.0,
 		_axis(out[1]),
-		_axis(out[2])
+		_axis(out[2]) if allow_pitch_roll else 0.0
 	)
 	_action.thrust = Vector3(
 		_axis(out[3]),
-		_axis(out[4]),
+		_axis(out[4]) if allow_vertical else 0.0,
 		_axis(out[5])
 	)
 	_action.turbo = out[6] > 0.0

Game/scripts/training_mode.gd

@@ -91,6 +91,12 @@ var _agents: Array[ShipAIController] = []
 # Eval mode state (see header comment)
 var _eval := false
 var _eval_models: Array[String] = ["", ""]
+# Per-model locomotion mask — must match how each model was actually trained
+# (see AIShipController's identical exports), so a stage 1/2 (grounded)
+# candidate isn't unfairly penalized by untrained aerial noise during eval
+# that its training environment never had.
+var _eval_allow_vertical: Array[bool] = [true, true]
+var _eval_allow_pitch_roll: Array[bool] = [true, true]
 var _eval_episodes := 20
 var _eval_goals := {0: 0, 1: 0}
 var _eval_draws := 0
@@ -123,6 +129,8 @@ func _start() -> void:
 		for team in [0, 1]:
 			var bot := AIShipController.new()
 			bot.model_path = _eval_models[team]
+			bot.allow_vertical = _eval_allow_vertical[team]
+			bot.allow_pitch_roll = _eval_allow_pitch_roll[team]
 			spawn_ship(team, 0, bot)
 		return
 
@@ -161,6 +169,10 @@ func _parse_eval_args() -> void:
 		_eval_models[0] = args["eval_model_a"]
 		_eval_models[1] = args["eval_model_b"]
 		_eval_episodes = int(args.get("eval_episodes", str(_eval_episodes)))
+		_eval_allow_vertical[0] = _typed_like(args.get("eval_allow_vertical_a", "true"), true)
+		_eval_allow_vertical[1] = _typed_like(args.get("eval_allow_vertical_b", "true"), true)
+		_eval_allow_pitch_roll[0] = _typed_like(args.get("eval_allow_pitch_roll_a", "true"), true)
+		_eval_allow_pitch_roll[1] = _typed_like(args.get("eval_allow_pitch_roll_b", "true"), true)
 
 
 # TrainingMode @export names a curriculum run may override from the cmdline.

TRAINING.md

@@ -122,6 +122,13 @@ appends to `training/eval_history.json` — the long-term progress record.
 Evaluate each new candidate against the previous promoted bot and a fixed
 early reference to see absolute progress over time.
 
+If a model was trained with the locomotion mask on (curriculum stages 1-2 —
+see below), pass `--grounded-a`/`--grounded-b` for whichever side it's on.
+The eval otherwise runs `AIShipController` fully unmasked regardless of how a
+model was trained, so a grounded model's untrained vertical/pitch-roll output
+reaches the ship as noise it never had to contend with during training —
+this understates it, not a neutral comparison.
+
 ## Difficulty tiers
 
 A bot is `(model, reaction_ticks, action_noise)` — configured on the Match