Ok, so thought (I have no idea if this would actually work, and I don't think any factions have the necessary casters at this point)
Luck/Think/Predict (or possibly math). Effect: win almost any battle (read below to see how)
Ok, so the basis for this one is solid, but there may be a flaw in there somewhere (possibly in predictmancer choice)
So in book 0, two things have been told.
1- luckcasters can de-buff and buff. I'm going to assume they can de-buff enemies. It has also been determined that it is hard to buff/debuff more than a unit.
2- (just a theory) luck is stolen from somewhere else.
So the predictmancer, if they can, predicts which of the enemies will roll high, and which of the allies will roll low. Each round, the luckmaster then selectively buffs and de-buffs units (which has been established to be easier than whole armies) so that your side has good (not perfect but good) luck, and your enemy has bad. theoretically, if point 2 is correct, this should actually be easier (with practice) because you are not "pulling a 4 out of the air", you are just switching the luck of two dice. What's more, you are not wasting a debuff of enemies who would have rolled poorly anyway, and you don't have to maintain the link after that roll. Essentially, this allows a precision use of luckmancy, using less energy and allowing you to possibly even do it multiple times in a turn.
The upshot is this: your units consistently roll average or better, and their units consistently roll average or worse. short of epic warlord bonuses, this should be enough to swing the fight. If I'm right and not missing something. This, of course, helps you in no way against ambushes.
The real problem is that I'm not sure if this use of predictmancy or mathmancy, and if you need a lookmancer in the link to look that closely at your enemies.