More entertaining than I expected, although my first impression is that Draft is heavily dominated by two metas: "NG with a dozen locks" and "play a card, summon 5+ copies from your deck". Often both in the same deck.
I'd suggest giving an option to choose the keg or 100 ore for the reward. One of the things I enjoyed about the old Arena was that if you won 4 or 5 games, you could win back most of your entry fee and could usually play again right away. From my perspective, I won a keg type that I didn't want and I'm left without enough ore to enter again, so I guess that's it for me playing this mode for a day or two...
My other suggestion would be that post-game pack selections should include at least one choice from your faction. Otherwise don't bother giving us Devotion card choices based on our Leader. I got forced to handicap myself after one match because I had no MO faction choices while playing a Blood Scent deck that was heavily dependent on multiple copies of Unseen Elder, a devotion card.
If I decide to keep playing Draft, I'll probably just stop buying kegs entirely since the powder payout appears to be much better than the 10 for 100 ore that I usually get (and powder is all I'm interested in these days). It seems like it'll provide the option to get "guaranteed" powder as opposed to gambling for epic / legendary premiums.
My first outing resulted in a 3-4 record, and I got a keg + 45 ore + 45 powder. 45 is such an odd number, that it's only significant attribute is that it's evenly divided by the number of wins I had. Low sample pool, but I'm fairly confident guessing we get 15 ore + 15 powder per win.
I'm also assuming the cards will be premium. Otherwise, I'd similarly be inclined to win no more than 5 games.
EDIT - So I miscounted... I was 4-3 with the keg/45 ore/45 powder reward. I managed to get some RP for ore, so I went again and finished 3-3-1, although
a draw counts as a loss, or "not a win" at any rate. 3 wins was good for a keg/30 scraps/40 powder reward. So rewards definitely scale by victories, but we need some more data points to determine how many resources we get at each point.