Been working on it. There's an information gap where it's much clearer who the top, say, 90 17-18-year-old NHL prospects are right now, vs. the top 90 14-16-year-olds. There are always some clear standouts in the younger group, but the consensus thins out rapidly, e.g. this what Eliteprospects' 2025 NHL draft preview page looks like now:
https://www.eliteprospects.com/draft-center/2025 - Misa, a few more notables, and only 20-odd other guys. But I've been a little more aggressive with giving good ratings to the top youngest guys in the last 2-3 years (and toned down my over-rating of the current draft year, which was part of the problem), so the situation should be improved.
After those first few years, the real guys disappear and the game's player generation takes over, and at that point the player pool is pretty steady. Long-term tests 20+ years out are showing a player pool in the rosters and drafts that's very comparable to what's there are the beginning.
There'll always be a difference. Classic mode result generation is frozen now and won't be adjusted further unless we find it doing something really odd. With 2D mode, whenever we make gameplay changes I then do a bunch of testing to see what that's done to the statistical results, then do a bunch of tweaking to get the stats back at the proper levels. So while they'll never work exactly the same as classic mode, both should be producing realistic results.
More scouting depth is still something I want to do, it was on my early planning list for 9 but I wasn't able to come up with a particularly fun/satisfying way of handling it, so we left it alone for now.
Mid-draft trade offers may be a little tough, the game engine isn't great at interrupting an ongoing event like that; we'd have to do it in a way that doesn't make the draft clunky and slow to advance. You'll notice in the new college recruiting "drafts", where teams compete against each other to sign players (requiring a lot of ongoing AI recalculations), that it tends to take much longer to finish the draft (even if you just auto-complete it) than a regular draft. So mid-draft trading would need to find a way around that problem.
We've done a few things, noted above, to get the AI trading more actively, including offers to human teams.
The big problem there is the absolute boatload of text that has to be written and maintained to make that kind of system enjoyable and non-repetitive. SI has no problem doing that because they have a core team of 100+ people, where ours can be counted on one hand. So we need to be a bit more abstract, for now.