Supplying Grad/Jacobian causes solver to think a solver is infeasible?

I'm easing my way into optimization and so I've created a scheduling problem (MINLP) that slowly getting more complex. It solves fine (with the Bonmin solver - I'm using the Opti toolbox because I wasn't too happy with the Matlab's GA's solutions) when I don't supply the jacobian or gradients, but if I do it fails and claims the problem is infeasible?

I must be doing something wrong...

I'm generating the derivatives with the symbolic math toolbox like so:

% Gradient
symGrad = jacobian(totCost);
grad = matlabFunction(symGrad,'vars',{X}); 
% Jacobian
symJac = jacobian(nlCon);
jac = matlabFunction(symJac, 'vars',{X});

Is there a better way to do it?

The whole file is here on dropbox

It's a very simple problem where I'm trying to schedule 3 generators over 10 time steps while minimising cost. The constraints are that they can be throttled between 70-100%, or be switched off completely.

Also just a shout-out to Jonathan Currie. You're a champ! Opti is a great tool