how is this more "random" than the old training methods
The "old" methods were 100% deterministic. If you train skill x, it gives a certain increase to a skill and a smaller increase to other skills, based on age and trainer level. So nothing random at all, at least that I could see.
The "new" methods, we don't know yet, at least for the non-primary training. But if you look at the quote that I pulled from the news above, it implies that you can't predict what happens to the other 10%. Which means it is not deterministic. Which is the definition of random, statistically speaking.
As for the 10% not being a big deal, in general I agree. However, as with any random process, there is always the possibility of getting a bad sample. Which means that you could get 10 training sessions in a row where all the training goes to one or a few skills. And those will add up to something significant. It is unlikely but it is possible.
Plus... Over a career, it adds up. Certainly it adds up for the primary skill. If you trained a player for 5 years, that's 70 weeks. And 10% of that is 7 weeks. You're losing a half of a season of training. And that half of a season goes to other skills - which may or may not add to anything significant (depending how lucky/unlucky you are).
Anyhow, I think we're just debating semantics, so really not much to see here.
Run of the Mill Canadian Manager