IMO when people complain of unrealistic drafting in Zwift, it’s not so much the general drafting algorithm that they’re responding to, but some ways in which it seems Zwift may be breaking/cheating the full physics model to make packs look more realistic.
Specifically, the lack of collision detection (i.e., riders are allowed to occupy the same space) in Zwift would mean an accurate physics model would create a very unrealistic-looking pack - basically, everyone would be squished together on the 2nd row of the pack, briefly rotating to the front before getting kicked back to the 2nd row.
In order to stretch the pack out, I believe Zwift is fudging the physics to keep riders moving up/back.
If e-racing is to become a real thing, I feel Zwift needs to be transparent about their model - they need to publicly document all of the ways their model works, including these fudging cases (sticky draft being one of the very obvious physics “fudges.”)
My hypothesis on Zwift pack dynamics is that there is a bias being applied to riders to keep them moving through the pack.
I’m not certain about this but I feel that once you hit the front, you get penalized until you hit the back of your group. I’ve observed that I’ll often drop far back in the group even while maintaining the same power level that brought me to the front.
In order to save energy, I’ve taken to letting myself hit the back of a group and then maintain just enough power to very slowly drift toward the front. It seems much easier to move up than to maintain position at the front.
For people who’re used to racing in real life, this is completely counter-intuitive. Your instinct is to try to keep yourself up front - you never want to give up positions if you can avoid it. Zwift’s algorithms, I believe, are actively punishing you for this - because you don’t move toward the back of the pack, Zwift’s bias stays in effect, forcing you to maintain a high power level to keep your position in the pack.
IMO, this is a case where Zwift may have taken the wrong path: It makes for a realistic-looking visual image of the pack, but is unrealistic from a perspective of race tactics - diminishing its value greatly for people who want to race online and offline.
In any case - I may be wrong about my theory, but there’s definitely some “black magic” in Zwift’s algorithms they haven’t revealed. I feel they owe it to the racing community to be open about this (and to give the community an opportunity to give feedback).