diff --git a/src/game/engine/src/modules/ai/ai_turn_bridge.gd b/src/game/engine/src/modules/ai/ai_turn_bridge.gd
index b0926377..35ad0489 100644
--- a/src/game/engine/src/modules/ai/ai_turn_bridge.gd
+++ b/src/game/engine/src/modules/ai/ai_turn_bridge.gd
@@ -271,6 +271,11 @@ static func _replay_learned_actions(action_log: Array, player: RefCounted) -> in
 	var applied: int = 0
 	var deferred_founds: int = 0
 	var dropped_variants: Array[String] = []
+	# Gate (b) instrumentation — record (intended_axial, resulting_gdscript_pos)
+	# for the first few moves so the Inc-2 gate can prove the GDScript world
+	# actually moved to the policy's chosen target. Capped to keep the log tight.
+	var parity_samples: Array[String] = []
+	const PARITY_SAMPLE_CAP: int = 6
 	for entry: Dictionary in action_log:
 		var kind: String = String(entry.get("kind", ""))
 		match kind:
@@ -279,14 +284,24 @@ static func _replay_learned_actions(action_log: Array, player: RefCounted) -> in
 				# the move-as-attack case at the target hex).
 				var off: Vector2i = Vector2i(int(entry.get("to_col", 0)), int(entry.get("to_row", 0)))
 				var ax: Vector2i = HexUtilsScript.offset_to_axial(off)
+				var uid: int = int(entry.get("unit_id", -1))
 				var move_json: String = JSON.stringify({
 					"MoveUnit": {
-						"unit_id": int(entry.get("unit_id", -1)),
+						"unit_id": uid,
 						"to_hex": [ax.x, ax.y],
 					}
 				})
 				if DispatchScript.dispatch_action(move_json, player, index_maps, city_name):
 					applied += 1
+					# Position-parity proof: read the unit's resulting GDScript
+					# position and compare to the intended axial target.
+					if parity_samples.size() < PARITY_SAMPLE_CAP:
+						var u: RefCounted = (index_maps.get("units", {}) as Dictionary).get(uid)
+						if u != null:
+							var match_str: String = "OK" if u.position == ax else "MISMATCH"
+							parity_samples.append(
+								"u%d→intended%s got%s [%s]" % [uid, str(ax), str(u.position), match_str]
+							)
 			"queue":
 				var queue_json: String = JSON.stringify({
 					"SetProduction": {
@@ -302,10 +317,18 @@ static func _replay_learned_actions(action_log: Array, player: RefCounted) -> in
 				dropped_variants.append(String(entry.get("variant", "?")))
 			_:
 				dropped_variants.append(kind)
+	print("[p1-29k] replay applied=%d (move/attack/queue dispatched into GDScript)" % applied)
+	if not parity_samples.is_empty():
+		print("[p1-29k] position_parity: %s" % str(parity_samples))
 	if deferred_founds > 0:
 		print("[p1-29k] DEFERRED %d FoundCity action(s) to Inc-3 (p1-29j boundary)" % deferred_founds)
 	if not dropped_variants.is_empty():
-		print("[p1-29k] dropped %d uncovered variant(s): %s" % [dropped_variants.size(), str(dropped_variants)])
+		# Summarize dropped-variant counts rather than dumping the full toggle
+		# storm (the policy oscillates Fortify↔Unfortify — see Inc-2 report).
+		var dropped_counts: Dictionary = {}
+		for v: String in dropped_variants:
+			dropped_counts[v] = int(dropped_counts.get(v, 0)) + 1
+		print("[p1-29k] dropped %d uncovered action(s): %s" % [dropped_variants.size(), str(dropped_counts)])
 	return applied