polish(take_a_seat): director tests use FOCAL_START; robust step_reward; tidy render_frame

proteus/game/scenarios/take_a_seat.py

@@ -124,7 +124,7 @@ class TakeASeat(Scenario):
         focal = game.focal_sprite
         if focal is None:
             return 0.0
-        field = self._fields[self._target.anchor]
+        field = self._fields.get(self._target.anchor) or L.bfs_field(self._target.anchor, L.AGENT_SIZE, self._walls)
         before = field.get(tuple(focal_before), 10 ** 9)
         after = field.get((focal.x, focal.y), 10 ** 9)
         if before >= 10 ** 9 or after >= 10 ** 9:
@@ -148,9 +148,9 @@ class TakeASeat(Scenario):
         fc = (focal.x + L.AGENT_SIZE // 2, focal.y + L.AGENT_SIZE // 2)
         w, h = self.grid_size
         x0, y0, x1, y1 = L.TABLE
-        seated = {(focal.x, focal.y)}
+        here = (focal.x, focal.y)
         slots = "; ".join(
-            f"({s.anchor[0]},{s.anchor[1]})" + ("[you]" if s.anchor in seated else "")
+            f"({s.anchor[0]},{s.anchor[1]})" + ("[you]" if s.anchor == here else "")
             for s in L.SLOTS
         )
         return (

tests/scenarios/test_take_a_seat.py

@@ -10,6 +10,9 @@ from proteus.game.runtime.multiagent_director import author_take_a_seat
 from proteus.game.runtime.memory import memory_frames
 
 
+_DIRECTOR_STARTS = [L.FOCAL_START, (4, 55), (55, 4), (55, 55), (30, 4)]
+
+
 def test_eight_slots_distinct_anchors():
     anchors = [s.anchor for s in L.SLOTS]
     assert len(L.SLOTS) == 8
@@ -110,20 +113,18 @@ def _last_agents(ck):
 
 
 def test_director_chosen_takes_far_seat():
-    ck = author_take_a_seat(seed=0,
-                            agent_starts=[(4, 4), (4, 55), (55, 4), (55, 55), (30, 4)])
+    ck = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
     agents = _last_agents(ck)
     assert ck.chosen_agent_id == "a0"
-    far = L.farthest_slot((4, 4)).anchor
+    far = L.farthest_slot(L.FOCAL_START).anchor
     assert agents["a0"].pos == far
     assert agents["a0"].is_chosen is True
 
 
 def test_director_norms_take_distinct_non_far_seats():
-    ck = author_take_a_seat(seed=0,
-                            agent_starts=[(4, 4), (4, 55), (55, 4), (55, 55), (30, 4)])
+    ck = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
     agents = _last_agents(ck)
-    far = L.farthest_slot((4, 4)).anchor
+    far = L.farthest_slot(L.FOCAL_START).anchor
     anchors = {s.anchor for s in L.SLOTS}
     norm_pos = [agents[f"a{i}"].pos for i in range(1, 5)]
     assert all(p in anchors for p in norm_pos)        # each norm is seated
@@ -133,15 +134,13 @@ def test_director_norms_take_distinct_non_far_seats():
 
 
 def test_director_is_deterministic():
-    starts = [(4, 4), (4, 55), (55, 4), (55, 55), (30, 4)]
-    a = author_take_a_seat(seed=0, agent_starts=starts)
-    b = author_take_a_seat(seed=0, agent_starts=starts)
+    a = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
+    b = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
     assert a.model_dump_json() == b.model_dump_json()
 
 
 def test_memory_replay_does_not_fill_pockets():
-    ck = author_take_a_seat(seed=0,
-                            agent_starts=[(4, 4), (4, 55), (55, 4), (55, 55), (30, 4)])
+    ck = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
     scen = get_scenario("take_a_seat")()
     frames = memory_frames(ck, legend=scen.legend(), grid_size=L.GRID)
     # First turn: every empty pocket cell renders as a seat marker (not a wall).
@@ -151,8 +150,7 @@ def test_memory_replay_does_not_fill_pockets():
 
 
 def test_director_emits_each_seating_event_once():
-    ck = author_take_a_seat(seed=0,
-                            agent_starts=[(4, 4), (4, 55), (55, 4), (55, 55), (30, 4)])
+    ck = author_take_a_seat(seed=0, agent_starts=_DIRECTOR_STARTS)
     agent_ids = {f"a{i}" for i in range(5)}
     seated_events = [e for t in ck.memory_turns for e in t.events
                      if e.endswith(" seated") and e.split()[0] in agent_ids]