Correct the one-controller claim: the drone takes over, it does not refuse

Tested on the aircraft. A second ARSDK handshake is accepted and telemetry is
redirected to it; the first session's frames stop while it still reports
connected = True. So the claim inherited from pyparrot's error text, which had
reached our error messages, tool descriptions, simulator behaviour and a test
name, was wrong in the most misleading direction: a refusal would be loud, and
this is silent.

The simulator now models the takeover by default; refusal stays available
because a client must handle a non-zero status anyway.

src/mcbebop/errors.py

@@ -21,9 +21,14 @@ def already_connected(target: str) -> ToolError:
 
 
 def handshake_refused(status: int) -> ToolError:
+    # Tested 2026-10-02: a Bebop 2 does NOT refuse a second controller, it
+    # accepts it and redirects telemetry to it. So a non-zero status means
+    # something other than "already taken", and guessing at it would send the
+    # reader looking in the wrong place.
     return ToolError(
-        f"The drone refused the connection (status {status}). It serves one controller at a time, "
+        f"The drone refused the connection (status {status}). That is unusual: this aircraft accepts "
-        "so close FreeFlight on any phone or tablet that is holding the link."
+        "a second controller rather than refusing one, so the cause is more likely the drone still "
+        "booting, or a firmware that differs from 4.7.1. Wait a few seconds and try again."
     )
 
 

src/mcbebop/sim.py

@@ -184,7 +184,10 @@ class FakeBebop:
     host: str = "127.0.0.1"
     discovery_port: int = 0  # 0 asks the OS, which is what parallel tests want
     c2d_port: int = 0
-    single_controller: bool = True
+    # False by default because that is what the aircraft does: it accepts a
+    # second controller and redirects telemetry to it. Set True to exercise a
+    # client's handling of a refusal, which some other firmware may still give.
+    single_controller: bool = False
     stream_hz: float = 5.0
     battery_start: int = 87
 
@@ -295,8 +298,12 @@ class FakeBebop:
                     continue
                 self.handshakes.append(request)
                 if self.single_controller and self.occupied:
-                    # What the aircraft does to a second controller, and the
+                    # NOT what a real Bebop 2 does. Tested on the aircraft
-                    # only way to test that path without two phones.
+                    # 2026-10-02: it accepts a second controller and silently
+                    # redirects telemetry to it, leaving the first starved but
+                    # still believing it is connected. This refusal path is kept
+                    # because a client must handle a non-zero status anyway, but
+                    # the default models the takeover. See docs network notes.
                     conn.sendall(json.dumps({"status": 1}).encode() + b"\x00")
                     continue
                 self._d2c = (addr[0], int(request["d2c_port"]))

src/mcbebop/tools/connection.py

@@ -45,8 +45,14 @@ def register(mcp: FastMCP, settings: Settings) -> None:
     ) -> ConnectionInfo:
         """Open a session with the drone. Nothing else works until this succeeds.
 
-        The drone serves one controller at a time, so close any phone app first.
         This machine must already have joined the drone's own Wi-Fi network.
+
+        Connecting while something else is already controlling the drone takes
+        the link over rather than failing: the aircraft accepts the newcomer and
+        stops sending telemetry to whoever had it, without telling them. So if a
+        phone app is flying it, connect here and the phone goes quiet. Watch
+        last_update_age rather than the connected flag to tell a live link from
+        one that has been taken.
         """
         state = app()
         async with state.lock:

tests/test_arsdk_session.py

@@ -119,7 +119,14 @@ async def test_handshake_describes_this_controller(sim, session):
     assert request["d2c_port"] == session.d2c_port != 0
 
 
-async def test_second_controller_is_refused(sim, session):
+async def test_second_controller_takes_the_link_over(sim, session):
+    """The aircraft accepts a newcomer and stops talking to whoever had it.
+
+    Measured on the real drone 2026-10-02: the first session's frame count
+    froze while the second received, and the first went on reporting
+    connected = True. That silent failure is why a client should judge a link
+    by telemetry freshness rather than by its connected flag.
+    """
     other = DroneSession(
         "127.0.0.1",
         discovery_port=sim.discovery_port,
@@ -128,9 +135,13 @@ async def test_second_controller_is_refused(sim, session):
         encoder=fake_encode,
         decoder=fake_decode,
     )
-    with pytest.raises(HandshakeError, match="one controller"):
+    await other.connect()
-        await other.connect()
+    try:
-    assert not other.connected
+        assert other.connected
+        # The loser keeps its socket and its optimism; only the data stops.
+        assert session.connected
+    finally:
+        await other.disconnect()
 
 
 async def test_unreachable_address_fails_fast_rather_than_hanging():

tests/test_live_findings.py

@@ -55,3 +55,19 @@ def test_acknowledgement_is_decided_by_data_type_not_buffer():
         frame = Frame(DataType.DATA_WITH_ACK, buffer_id, 9, b"\x00\x05\x00\x00")
         assert Frame.decode_all(frame.encode())[0].buffer_id == buffer_id
         assert 0 <= BufferId.ack_for(buffer_id) <= 255
+
+
+def test_simulator_models_takeover_not_refusal_by_default():
+    """The aircraft accepts a second controller; it does not refuse one.
+
+    Tested on the real drone 2026-10-02: session A's frames froze at 264 while
+    session B took over, and A went on reporting connected = True. The
+    simulator defaulted to refusing, which is the same unverified assumption
+    the client carried, so no test could catch the difference.
+    """
+    import dataclasses
+
+    from mcbebop.sim import FakeBebop
+
+    field = next(f for f in dataclasses.fields(FakeBebop) if f.name == "single_controller")
+    assert field.default is False, "the default must model the aircraft, not the folklore"

tests/test_sim.py

@@ -74,6 +74,16 @@ def sim():
         yield fake
 
 
+@pytest.fixture
+def sim_factory():
+    """For tests that need the refusing simulator rather than the default.
+
+    The default models the aircraft: a second controller is accepted and the
+    first is starved. Refusal is the opt-in mode.
+    """
+    return FakeBebop
+
+
 @pytest.fixture
 def controller(sim):
     client = Controller(sim)
@@ -165,15 +175,34 @@ def test_handshake_is_accepted_and_names_a_c2d_port(sim, controller):
     assert sim.handshakes[0]["controller_name"] == "test"
 
 
-def test_a_second_controller_is_refused(sim, controller):
+def test_a_second_controller_is_accepted_by_default(sim, controller):
+    """Matching the aircraft, which accepts a newcomer and starves the first.
+
+    Tested on the real drone 2026-10-02. The refusal this test used to assert
+    was folklore inherited from pyparrot's error text.
+    """
     second = Controller(sim)
     try:
-        assert second.reply["status"] == 1
+        assert second.reply["status"] == 0
     finally:
         second.close()
 
 
+def test_a_second_controller_can_be_refused_when_asked(sim_factory):
+    """Kept because a client must still handle a non-zero status."""
+    with sim_factory(single_controller=True) as strict:
+        first = Controller(strict)
+        second = Controller(strict)
+        try:
+            assert first.reply["status"] == 0
+            assert second.reply["status"] == 1
+        finally:
+            second.close()
+            first.close()
+
+
 def test_the_slot_is_free_again_after_release(sim, controller):
+    """Only meaningful in the strict mode; the default never withholds a slot."""
     assert sim.occupied
     sim.release()
     third = Controller(sim)