Well this is where it gets confusing because the theory says to assume all being summed or subtracted and let the signs take care of themselves. a +K would probably end up as +-k or would the + still throw it ?
The theory says that you must add all the currents converging into the node.
In order to do that, you must make sure that the currents are incoming into that node, consistent with the global orientation you chose for the circuit.
For example, when writing the equation for the left node, you did:
left node: (120-x)/2 + (0-x)/(-i5) + k + (-14.14-i14.14)/(-i5) = 0
Since you are adding all these currents, you are using the fact that they are converging into node x: a <---(k)--- b, and also: a <---(-14.14-i14.14)/(-i5)--- b
When it's time to write the equation for node 'y', you must add currents converging into that node, so:
a <---(k)--- b turns into: a ---(-k)---> b
a <---(-14.14-i14.14)/(-i5)--- b turns into: a ---(14.14+14.14i)---> b
Note that a <---(k)--- b and a ---(-k)---> b are saying exactly the same, but in order to add currents converging into the node, you must choose the latter. So the equation is:
right node: (i120-y)/4 + (0-y)/(i4) - k + (14.14+i14.14)/(-i5) = 0
So, yes, in the second eqation you used +(-k) = -k.
Anyway, for me it works better to choose a clear orientation for each wire and separately add incoming and outgoing currents in each node.