diff --git a/components/actor/hendiCtrl.py b/components/actor/hendiCtrl.py
index bb4c8a0..afec45a 100755
--- a/components/actor/hendiCtrl.py
+++ b/components/actor/hendiCtrl.py
@@ -213,9 +213,23 @@ class HendiCtrl:
 		if self.ser.is_open:
 			try:
 				self.remoteEnable(False)
-			except HendiException:
+			except Exception:
+				# Not just HendiException (a protocol-level NAK) - a genuine
+				# unplug fails deep inside pyserial itself (termios.error/
+				# OSError from flushInput()/readline() on a dead fd), and
+				# disconnect() must never let that escape: every caller here
+				# (HeaterHendi.process()/activate()'s own except handlers,
+				# AHeater.open()'s context-manager exit) calls disconnect()
+				# specifically to recover from a comm failure, so if this
+				# raises too it propagates all the way out of HeaterTask's
+				# on_process() and permanently kills that task's loop - the
+				# device then never gets process()'d again even after a
+				# later successful reconnect.
+				pass
+			try:
+				self.ser.close()
+			except Exception:
 				pass
-			self.ser.close()
 		self.sw_id = None
 		self.sw_ver = None
 
diff --git a/tasks/heater.py b/tasks/heater.py
index ed67cef..24b7c92 100755
--- a/tasks/heater.py
+++ b/tasks/heater.py
@@ -122,35 +122,57 @@ class HeaterTask(ATask):
 		# connected.
 		await self.send({'Connected': self.device.connected})
 
-		with self.device.open():
-			while True:
-				# Closed-loop: TC has full control. Open-loop: direct manual power only.
-				power_soll = self.power_actor if self.closed_loop else self.power_soll
-				self.power_set_changed(power_soll)
+		# Outer retry loop - belt-and-suspenders around the tick loop's own
+		# per-call guards below. A comm failure a device-level guard didn't
+		# anticipate (e.g. a raw pyserial/termios exception escaping from
+		# somewhere other than device.process(), the one call already
+		# wrapped below) must never be allowed to fall out of this
+		# coroutine entirely: nothing restarts a dead ATask, so a single
+		# uncaught exception here would permanently stop this device from
+		# ever being process()'d again - even after a later, otherwise-
+		# successful reconnect. See the incident this was added for: a
+		# mid-brew unplug raised termios.error from deep inside
+		# HendiCtrl.disconnect() itself (fixed there too), which escaped
+		# all the way out of this loop and killed it for the rest of the
+		# process's life.
+		while True:
+			try:
+				with self.device.open():
+					while True:
+						# Closed-loop: TC has full control. Open-loop: direct manual power only.
+						power_soll = self.power_actor if self.closed_loop else self.power_soll
+						self.power_set_changed(power_soll)
 
-				# Calculate duty cycle
-				power_low = self.get_next_smaller_power(power_soll)
-				power_high = self.get_next_greater_power(power_soll)
-				power_step = power_high - power_low
-				power_diff = power_soll - power_low
-				duty = power_diff / power_step
+						# Calculate duty cycle
+						power_low = self.get_next_smaller_power(power_soll)
+						power_high = self.get_next_greater_power(power_soll)
+						power_step = power_high - power_low
+						power_diff = power_soll - power_low
+						duty = power_diff / power_step
 
-				on_count = pulse_period_count*duty
+						on_count = pulse_period_count*duty
 
-				self.pulse_counter += 1
-				if self.pulse_counter >= pulse_period_count:
-					self.pulse_counter = 0
+						self.pulse_counter += 1
+						if self.pulse_counter >= pulse_period_count:
+							self.pulse_counter = 0
 
-				if power_soll == 0 or self.pulse_counter >= on_count:
-					self.device.set_power(power_low)
-				elif self.pulse_counter < on_count:
-					self.device.set_power(power_high)
+						if power_soll == 0 or self.pulse_counter >= on_count:
+							self.device.set_power(power_low)
+						elif self.pulse_counter < on_count:
+							self.device.set_power(power_high)
 
+						try:
+							self.device.process()
+						except Exception as e:
+							print(f"HeaterTask: comm error, marking disconnected: {e}")
+							self.device.disconnect()
+						await asyncio.sleep(self.interval)
+			except Exception as e:
+				print(f"HeaterTask: unexpected error, recovering: {e}")
 				try:
-					self.device.process()
-				except Exception as e:
-					print(f"HeaterTask: comm error, marking disconnected: {e}")
 					self.device.disconnect()
+				except Exception:
+					pass
 				await asyncio.sleep(self.interval)
 
 
diff --git a/tasks/stirrer.py b/tasks/stirrer.py
index ed847d9..b52218b 100644
--- a/tasks/stirrer.py
+++ b/tasks/stirrer.py
@@ -66,11 +66,24 @@ class StirrerTask(ATask):
 		# observer.
 		await self.send({'Connected': self.device.connected})
 
-		with self.device.open():
-			while True:
+		# Outer retry loop - see HeaterTask.on_process()'s own comment for
+		# why: nothing restarts a dead ATask, so an exception escaping this
+		# coroutine entirely would permanently stop this device from ever
+		# being process()'d again, even after a later successful reconnect.
+		while True:
+			try:
+				with self.device.open():
+					while True:
+						try:
+							self.device.process()
+						except Exception as e:
+							print(f"StirrerTask: comm error, marking disconnected: {e}")
+							self.device.disconnect()
+						await asyncio.sleep(self.interval)
+			except Exception as e:
+				print(f"StirrerTask: unexpected error, recovering: {e}")
 				try:
-					self.device.process()
-				except Exception as e:
-					print(f"StirrerTask: comm error, marking disconnected: {e}")
 					self.device.disconnect()
+				except Exception:
+					pass
 				await asyncio.sleep(self.interval)