Strange LSB flip bug when passing data out of component

-- SPI Slave 
-- Tailored to work with netduino

library ieee;
use ieee.std_logic_1164.all;
use ieee.std_logic_arith.all;
use ieee.std_logic_unsigned.all;

entity netduino_bridge is
	generic( data_width: integer := 8
				);
	port( -- Hard ports
			clk: in std_logic;
			SCLK, MOSI, CS: in std_logic;
			MISO: out std_logic := '0';
			-- Soft ports
			target_out: out unsigned(3 downto 0);	
			target_in: in unsigned(3 downto 0);
			status: in unsigned(3 downto 0);
			cmd: out unsigned(7 downto 0)
			);

end netduino_bridge;


architecture behavioural of netduino_bridge is
	
	type state is (IDLE, Rx, Tx);
	signal SPI_state: state := IDLE;
	signal Rx_data, Tx_data: unsigned(data_width-1 downto 0);
	signal Rx_count, Tx_count: integer := 0;
	signal latch_data: std_logic := '0';
	signal target_reg: unsigned(3 downto 0) := "0001";
begin

	target_out <= target_reg;

-- Command decoding process
	process(latch_data)
	
	variable lock, search: std_logic := '0';
	
	begin
	
		if rising_edge(latch_data) then
		
			case Rx_data(7 downto 4) is		
			
				-- Set new target
				when x"1" =>
					Tx_data <= Rx_data;
					target_reg <= Rx_data(3 downto 0);

				-- Query current target/status
				when x"2" =>
					Tx_data <= status & target_in;
					
				-- Find 'real' zero
				-- A command to set the inclination to real zero level
				-- rather than zero relative to last position (based on step count)
									
				-- Echo
				when x"4" =>
					Tx_data <= x"4" & Rx_data(3 downto 0);		
			
				when others =>
					Tx_data <= x"E0";
			end case;
			
		end if;
		
	end process;


-- Communication process
	process(CS, SCLK, Rx_count, Tx_count) is
	begin
		
		if(CS = '0') then
		
			if rising_edge(SCLK) then	
				-- Receive			
				if(Rx_count < data_width) then
					Rx_count <= Rx_count + 1;
					Rx_data(data_width-1-Rx_count) <= MOSI; -- Shift in data on MOSI
				end if;
				-- Flag new data (Latch asynchronously)
				if(Rx_count = data_width-1) then
					latch_data <= '1';
				end if;	
				-- Hopefully time in here to decode the command and provide a reply.			
			end if;
			
			if falling_edge(SCLK) AND Rx_count = data_width then
			
				-- Transmit
				if(Tx_count < data_width) then
					Tx_count <= Tx_count + 1;
					MISO <= Tx_data(data_width-1-Tx_count);
				end if;
					
				
			end if;
		
		else
			-- Stop latching data_out to Rx_data
			Latch_data <= '0';
			-- Reset counters
			Tx_count <= 0;
			Rx_count <= 0;
			
		end if;
		
	end process;
	
end behavioural;

-- Signals and constants

	-- Positional
	signal azimuth, elevation: integer := 0;
	signal target_new: unsigned(3 downto 0) := x"1";
	signal target_set: unsigned(3 downto 0) := x"0";
	signal cmd: unsigned(7 downto 0) := x"00";
	
	-- Status flags
	signal Az_idle, El_idle: std_logic;
	signal status: unsigned(3 downto 0); -- ? zeroed, search, lock

-- Netduino interface
	netduino: entity work.netduino_bridge port map(mclk, net_SCLK, net_MOSI, net_CS, net_MISO, target_new, target_set, status, cmd);
	
------ PROCESSES ------	


	-- Deal with netduino commands
	
	process(target_fix) is
	begin
	
		-- if both axes are idling, accept new position.
		if(Az_idle = '1' and El_idle = '1') then
			
			case target_new is
				-- zero position
				when "0000" =>
					azimuth 	 <= 0;
					elevation <= 0;
				
				-- Sat 1	
				when "0001" =>
					azimuth 	 <= 50;
					elevation <= 50;
				
				-- Sat 2
				when "0010" =>
					azimuth 	 <= 100;
					elevation <= 300;
				
				-- Sat 3
				when "0011" =>
					azimuth 	 <= 200;
					elevation <= 180;
				
				-- Sat 4
				when "0100" =>
					azimuth 	 <= 300;
					elevation <= 100;
			
				-- Unrecognised; do nothing
				when others =>
					null;
			end case;
		
		end if;
		
		-- wait until pointing is finished and report lack of lock
		if (Az_idle = '0' AND El_idle = '0') then
			status(0) <= '0';
		else
		-- once pointing finished, update status
		target_set <= target_new;
		status(0) <= '1';
		end if;
		
	end process;