One of the reasons I stopped playing DOTA 2 was because I could not figure out how to slice it into small enough pieces to deliberately practice. And over the course of the 800 hours I put in, the impact I had on each game stayed relatively inconsequential. Which ultimately felt really unrewarding. My win rate was basically chance.
I think there were two main things at play - one was a long practice loop, and the other was difficulty in self-assessment.
Each DOTA game lasts about 45 minutes to an hour and a half (on the long end). I basically never had time to even play two in a row. Now I could break the loop into something smaller, like a specific action in the game. For example, hitting creeps. But it is the late game where things really open up, and if I only got the practice late game maneuvers once it just was never going to be enough. Other small game actions are actively bad to do in certain situations, so repeatedly practicing them is not so much the challenge - the challenge is knowing when to do them.
The second issue was self-assessment. For very small foundation actions I could assess. Such as the gold example given above. But for pretty much everything else I could not tell if I made the right choice. I could not tell if I bought the right items, I could not tell if I positioned myself correctly. I could not assess if I made a good trade. One of the beauties of DOTA 2 is its complexity. There is almost nothing that is always bad. I died - surly that is bad. Yeah, but did you damage the other player so that he too will have to run back to base? Then maybe you dying as a support was actually good.
A key piece of practice in multiplayer games is to be able to determine how you specifically played, separate from your team. This means that the overall outcome of the game (win or lose) is not a good indicator of your game. But I couldn’t find concrete other metrics the judge my game by. I could not tell if I played a good game or not. Some games I felt I played well, but my gold and XP was abysmally low or vice versa.