As is said here, if you are just trying to get nodal voltages, all that matters is consistency. I get real confused with these directions. I always assume current entering a node as positive, current leaving as negative. I once had a filters class with a lot of opamps. They taught me a method that seems to work. Suppose I have a node A and a node B. First, assume there is a resistor R1 between A and B. I will always assume the current into the node (node A) from that branch is (Va-Vb)/R1. Now, assume there is a node C also attached to node A with a resistor R2 between. I will still start with Va and calculate the current into the node as (Va-Vc)/R2. On and on, for 100 nodes it is always Va minus something divided by a resistance (it also works for caps and coils impedance). There is a slight variation for a current source in the branch. If there is a current source arrow pointing in I call it I, away from the node out I call it -I. Then I just add it up, and set to zero. It always works for KCL and is usually a lot easier than some funky loop analysis. Especially with ideal opamp circuits (where you just assume V+ = V-). Hope I confused you enough.