Good ideas. Personally I'm less of a fan of hard DC's. I like it when multiple stages of success and failure (other than nat 1's and 20's) are getting implemented in games. For example rolling a 9 on check to break down a door with a DC of 10 wouldn't be a complete failure, but it would take you multiple swings and maybe even exhaust your character. Or for pickpocketing someone the higher you role the more of that person's inventory can you access. (Ok, slightly off topic because a skill check, but just to give an example how differently you could handle DC rolls)
This matches a common tabletop mechanic that failures by 5 or more are "worse failures" then failures of by only 1-4. Larian could implement something like this for these checks.
Backstab (Stealth, Passive):
...
The character will have to make a stealth check versus the opponent’s passive perception.
...
This benefits the game by making the advantage a one-time bonus and giving the player more meaningful choices to make during combat. The player would have to move the character away, provoking an attack of opportunity, and move back into the “backstab zone” to roll for advantage again.
I'm all for reducing the impact of backstabbing, but your suggestion doesn't actually increase player choice. It just changes the guaranteed backstab into a ~50% (dependent on your stealth skill) chance of occurring. In 5e (and BG3) you're already allowed to walk around an enemy without provoking an AoO, so there'd be no reason to move a character away before backstabbing. The reason why this doesn't usually work in BG3 is that BG3's pathfinding is bad.
In order for the player 'choice' to be meaningful, there has to be a downside for failure besides "you don't get advantage". I'd recommend your stealth (or acrobatics or deception) check contested by the enemy's perception (acrobatics or insight), where a failure means you attack at DISadvantage. Now there is risk vs reward.
