fix(ai): map team-relative rotation actions

Game/scripts/ai_ship_controller.gd

@@ -75,11 +75,13 @@ func _decide() -> void:
 			_action.thrust.y = 0.0
 	else:
 		_action = ShipActionCodec.from_logits(out, action_noise)
+	_action = ShipActionCodec.apply_team_frame(_action, _ship.team)
 
 
 # Find ship/ball/teammates/opponents/goal once everything is spawned.
-# ShipAction axes are body-frame so only observations need team context
-# (ShipObservations). Rosters never change mid-match (no despawn path exists
+# ShipAction thrust axes are body-frame, while its rotation axes are mapped
+# from the canonical team frame by ShipActionCodec. Rosters never change
+# mid-match (no despawn path exists
 # anywhere in this codebase), so this only needs to run once — sorted by
 # spawn_index so a given ship keeps the same observation slot for the whole
 # match, matching TrainingMode's identically-sorted lists.

Game/scripts/ship_action_codec.gd

@@ -8,6 +8,13 @@ extends RefCounted
 # "do not fork this logic" role for observations; the train/inference seam
 # broke once before over exactly this kind of divergence (commit 8c15c46).
 #
+# Ship thrust is body-local, so its axes must not be mirrored for team 1.
+# Ship rotation, however, is applied directly as world-space torque in
+# Ship.apply_rotation_forces(). Team 1 observes a canonical frame rotated
+# 180 degrees about world Y, so its canonical pitch/roll outputs must be
+# rotated back to world space before they reach the ship. apply_team_frame()
+# is the shared training/inference seam for that conversion.
+#
 # Curriculum generation 4 replaces the old continuous Gaussian action space
 # (Box(7), see the "continuous" path below) with a per-axis MultiDiscrete
 # space: PPO's Gaussian std reliably collapsed to ~0.13-0.15 within the first
@@ -100,6 +107,17 @@ static func from_logits(logits: Array, noise: float) -> ShipAction:
 	return result
 
 
+# Map a policy's canonical-frame rotation intent back into the physical
+# team's world frame. A 180-degree Y rotation negates X and Z and leaves Y
+# unchanged. Translation remains untouched because Ship applies it through
+# the ship's local basis rather than as a world-space vector.
+static func apply_team_frame(action: ShipAction, team: int) -> ShipAction:
+	if team == 1:
+		action.rotation.x = -action.rotation.x
+		action.rotation.z = -action.rotation.z
+	return action
+
+
 # Legacy continuous decode — moved verbatim from ai_ship_controller.gd so
 # every model exported before generation 4 (no "action_space" block in its
 # JSON, e.g. Game/bots/promoted/easy.json) keeps behaving byte-identically.

Game/scripts/ship_ai_controller.gd

@@ -209,7 +209,9 @@ func get_action_space() -> Dictionary:
 
 
 func set_action(action) -> void:
-	rl_controller.action = ShipActionCodec.from_indices(action)
+	rl_controller.action = ShipActionCodec.apply_team_frame(
+		ShipActionCodec.from_indices(action), ship.team
+	)
 
 
 func reset():

training/test_action_space.py

@@ -17,6 +17,7 @@ Usage:
 """
 
 import sys
+from pathlib import Path
 
 import gymnasium as gym
 import numpy as np
@@ -30,6 +31,7 @@ from export_policy import ACTION_HEADS
 # out of order too, not just catch nothing because both sides changed
 # together.
 EXPECTED_ORDER = ["rot_x", "rot_y", "rot_z", "thrust_x", "thrust_y", "thrust_z", "turbo"]
+GAME_SCRIPTS = Path(__file__).resolve().parents[1] / "Game" / "scripts"
 
 
 def check_action_heads_match_expected_order() -> None:
@@ -87,12 +89,23 @@ def check_round_trip_preserves_per_head_values() -> None:
         np.testing.assert_array_equal(np.asarray(original[head_index]), expected_column)
 
 
+def check_team_frame_mapping_is_shared() -> None:
+    codec = (GAME_SCRIPTS / "ship_action_codec.gd").read_text()
+    training = (GAME_SCRIPTS / "ship_ai_controller.gd").read_text()
+    inference = (GAME_SCRIPTS / "ai_ship_controller.gd").read_text()
+    assert "action.rotation.x = -action.rotation.x" in codec
+    assert "action.rotation.z = -action.rotation.z" in codec
+    assert "ShipActionCodec.apply_team_frame(" in training
+    assert "ShipActionCodec.apply_team_frame(" in inference
+
+
 def main() -> int:
     checks = [
         check_action_heads_match_expected_order,
         check_gymnasium_sorts_to_expected_order,
         check_action_space_processor_produces_expected_multi_discrete,
         check_round_trip_preserves_per_head_values,
+        check_team_frame_mapping_is_shared,
     ]
     for check in checks:
         check()