Indeed k/d is an overall good proxy when taken with a step back, eg for a whole game. When looking at a single round, it ends up being noisy. Which is not to say it’s useless! In the end this is a shooter game, and to shoot is in the name (and not to die but that’s another matter) :D Now to the matter to evaluating odds of winning a combat.
To develop a bit on smokes: with some domain-specific knowledge you could approximate its usefulness. Smoke in catwalk on D2 allowing T to go through mid while a CT stands on shirt unable to take info nor shoot? Now that’s a good smoke. But this would require very tedious work an be non exhaustive. You could make a rating of "how many people walked in front of an opponent thanks to a smoke". Or how many bullets missed a CT defusing in the smoke. Possibilities are endless but not perfect.
For example scope.gg rates flashbangs by how many people were blinded. And I’d not be surprised if they gave a negative ranking to flashbangs that were heard but did not blind anyone: those give info to the opponent without helping you.
Now if you made a follow up that’d be great but you’ll end up meta-gaming instead of gaming. But that’s fun too!