Same thing as time paradox.
- based on your axiom - The spell of the prophecy is always accurate and has a 100% chance of being true.
Mages do not need to bother to train the guy, maybe the guy slips on a babana peel and accidentally kills the ultimate bad guy
But on the other hand, they may choose to help the guy for their own benefits, assuming it helps the CO to achieve his goal faster, and for him not to be a last guy on the planet, who kills the ultimate bad guy when no one actually cares anymore.
They (mages) may choose to help, but it does not mean their choice will matter, it still may be a banana peel accident, or 1000's different ways with or without help of the others. (That's why that CO is so good, write whatever, no need to think so much on a plan)
If the good guy is killed right at the start, it was just a wrong guy (but that a bit murky territory depending on your setting in which way they identify that CO, how do they find it, but also not a big problem in 1000's ways - they do not have full knowledge about universe, nor control - if they teleport one they can teleport few accidentally, or split bodies, or whatever)
So you can (in your fantasy world) have ultimately true prophecy, but same way as in the reality the same event may be achieved as result of unmeasurable number of consequence chains, connections, coincidences.
That stoned good guy on battle field, it even fails to create a slight disturbance in reasoning, because of not having full information about the ultimate bad guy - maybe they will find each other and smoke bong together for long an happy life, and once drunk accident happens again, damn banana.
So even if I said it similar to time paradox, but in fact it is quite far from it, and examples you had in mind(wrote down) do not do even a ripple as paradox does - it is just two points with unrestricted paths between them.
To have a paradox here it needs to have 2 ultimates/absolutes here, at least.
Prediction in general
Being able to predict something with absolute certainty for all the cases - not likely, but predicting itself as for some events it is more common that it may feels at a first glance - if I drop a stone from a plane, there is very slim chance it does not land on the surface of the planet(if I did it in the way it does not end up in a plane engine).
Difference from 100% is negligible, and similarly we do rely on predicting of consequences of events in our technologies, it basically all about getting expected results for most of the time for expected set of consequential and parallel events - start a car, fly a plane, make cpu crunch numbers, etc. Those are predictable, because they have some robustness against external events - potential barrier which has to be crossed to change it, so a movement of a random atom next/near the observable setting/system has a very slim chance to change expected outcome. (Slim means so close to 0 that you can't have seen such a number - so small it is)
So there are predictable events, and in a sense with certain power and ability to have/collect data - outcome of CO vs UBG also can be(maybe) predicted close to 100%. And how close to 100% your guys do not really know, as they are not source of that estimation, but are just users, and they not necessarily have stats to judge(a billion predictions is good stats, but without knowing how it works it gets us to just 99.9999999%, way worse than dropping stone from air plane).
So robustness of prediction depends on bifurcation points, some we do know some we do not(we are quite good, usually, identifying them in our tech stuff), but better the ability to collect information, and better prediction algorithms better the results of outcome(there are fundamental limits, it has to be some macro level events, human brains is something in between, murky territory, but it reflects in events on macro level). And when there are few or none bifurcation points - prediction can have certain probability(a bit more complex than just that, but for simplicity).
- in a sense CO is not necessarily the only guy who has capacity of killing UBG, others may have lesser chances to do so, but they are capable to do so, but they have more bifurcation points which if they cross them in a right way they may have that 100% as well, but they do have bifurcation points which makes their chances uncertain or less than 100%
In that sense CO is just one of with a set of development path's(which he has many as anyone) where each of those path's has no bifurcation points.
It basically about all that:
- If you know the enemy and know yourself, you need not fear the result of a hundred battles. If you know yourself but not the enemy, for every victory gained you will also suffer a defeat. If you know neither the enemy nor yourself, you will succumb in every battle. - @Sun Tzu, The Art of War - (probably)
In that sense a valid prediction, it less likely rely on some accidental event as path to success, or events, and it makes more sense for mages to put efforts in helping the guy.
PS - Free will
Can't avoid to add few point to that big topic, as some of the answers dive in extremes - determinism or free will.
Free will does not mean there is only unpredictable outcomes when it hits a human. Yes, a hundred years ago, or earlier it could be philosophical conundrum, enigma. But today we all can observe that behavioral patterns do exists - the whole advertisement model is build around that, focus groups, recommendation algorithms etc.
So some of them are one offer to many, some are many offers to one - and it works, sometimes better some times worse - but prediction works.
What creates complexity, is interaction in the society between agents, not only people vs people, but people vs environment, which diminishes the importance of freewill or not.
And as an example it possible to predict that most of viewers of the question will eat tomorrow, also as result of some will, as they can choose not to, same as the guy who wants to prove me wrong(write a comment btw, you are awesome, outstanding warrior of "someone is wrong on internet"). If we knew all the people, who will read it, could we determine if there is a guy who may make such a choice - choose to not eat - yes, we can try do so based on his behavioral patters. (Fbi joins the chat). Our predictions are not perfect, and ways are not mature, but still. It harder to predict who will actually read it - as it is a result of pvp and pve - but does not mean it impossible at all, or that presence of freewill stops us. There are strategies, habits people use, and in a sense we less free, aren't that fluid to render any possibility of prediction impossible.
Sure we have uncertainties, but the whole theory of ideal gas is build around random/chaotic movements of molecules(agents) and we have certain outcomes, our prediction works - not any set of chaotic events can make us helpless to predict at least something, and in some cases it can be CO vs UBG, in some cases we even do not need much to make such conclusion and prediction (wunderr waffe and genetic code to activate it, as an example)
So to have certain predictability there is no need to plunge in deterministic world, and we do have good results with chaotic not deterministic systems, predicting them, so as there is fuzzy logic approaches etc all kinds of things to deal with chaos.
So there is certain map of expectations of probabilities with certain uncertainties in them, and this map may have some unique guy whos network of probabilities ends up in slashing UBG, with or without other forces involved. Such events are probably rare, but it may exist, and may be visible for something which has sufficient strength to crunch numbers. Does not mean he is only one, but just one who can be seen.