I cannot get closed loop network to run.

%first, I prepare the open network in the usual way 
net=narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);

% Prepare Data
X = tonndata(inputs,false,false);
T = tonndata(target,false,false);
[x,xi,ai,t] = preparets(net,X,{},T);

% Train on first half of data 
% r is the length of the data
% 2 represents the delays
hw=floor(r/2);
x_firsthalf=x(:,1:hw-2);
t_firsthalf=t(1:hw-2);
net = train(net,x_firsthalf,t_firsthalf,xi,ai);

% Get the (in-sample) open loop results on the first half of the data
[yopen,xf,af] = net(x_firsthalf,xi,ai);
% These results are good, as expected.

% Convert to closed loop network
[netc,xi,ai]=closeloop(net,xf,af);

% Get the closed loop results on the second half of the data
x_secondhalf=x(:,hw-2+1:end);
[yclosed,xf,af] = netc(x_secondhalf,xi,ai);
% Returns error message: "Error using network/sim (line 270)
% Number of inputs does not match net.numInputs."

% Yet x_firsthalf and x_secondhalf have the same numbers of input columns.
% What am I overlooking?

[Walter Roberson]

But what is size(xf) and size(xi) ?

Hypothethically, the closeloop() might have changed the number of inputs required.

[Kevin Johnson]

In [netc,xi,ai]=closeloop(net,xf,af), size of xf is 2x2, and size of xi is 1 x 2. The target column was dropped by closeloop. When I remove it before using netc, netc executes (though the closed loop results are very poor unfortunately!). Thanks for your help.