I think most of the naysayers when discussing ranking for Zwift are just afraid of a worse racing experience than what they have today. Which probably amounts to them having a sweet time right now and thus having something to lose (the might podium less). But @Graham_Irvine_London is on to something important.
I’ve been a gamer all my life, a hopeless first-generation gamer who will probably be doing that in a retirement home long after I can no longer ride a bike. So I’m familiar with the ranking systems so common in many of the major competitive online games of today. Systems which work surprisingly well actually. And it seems there was never any issue creating those systems. Users never had to beg endlessly for years for them. The developers just created them and put them in. Just like that.
Now, the way it usually works is there are two ways of playing these games, both rely on matchmaking. Every player is calibrated early on. And then recalibrated in intervals. You can compare it to AutoKitten or the Zwift test model. The calibration says something about where the player stands today relative to other players. It’s a performance measure, just like the Zwift performance categories. Only they are not based on some theorycrafting some hot cycling/gaming coach came up with. They are based on a machine learning model and actual user data.
E.g. the model might take click rate into account. Typically, a good player will click the mouse much more often than a beginner. Is it better to click often? Is that how you win? No, not at all. It just turns out there is a statistical correlation between click rate and win rate, the model found out, so it uses that. And if it turned out that players with player names beggning with an ‘N’ were more prone to winning than players with names beginning with a ‘P’, then it would use that too. Completely irrelevant data in a sense but correlated with winning. A completely different approach than Zwift’s futile attempts at capturing “the essence of cycling” by relying on theories rather than facts, user data.
Then, once the player is calibrated, he can enter matches (think races) starting at a level corresponding to his calibration/start rank. He can either join ranked games or more casual, unranked games. Should he choose to play umranked, everything he does in-game still affects his calibration. He will thus always face fairly equal opposition and most players will hence have about a 50% win rate in 1-1 matchups (single or team games).
If he chooses to play ranked games, once the initial calibration is completed, the calibration measures no longer govern his mobility in the ranks past that point. Rather, it is his results. If he goes on an early winning streak, then he will climb in ranking and will soon face stiffer competition. If he starts to lose a lot instead, then the calibration obviously didn’t quite do it’s job in matching him against suitable opposition, but it’s OK because the changes in rank score will sort it out instead.
Then, on a day when the player doesn’t feel he can bring his A-game and just wants a casual game, he can still join an unranked game. The matchmaking will use his current, continuously readjusted calibration but the game result will not affect his ranking. So there is one official (“ranking”) rank score and one hidden (calibration).
You can translate calibration with performance-based category (current A-D) and rank score with a ranking we still don’t have (well, we do on ZP only it is mismanaged and distorted by the current rule set but given a new model and rule set it could work).
Tell me why this isn’t the best mix of two worlds and tell me why this wouldn’t be a vast improvement for Zwift and the community. I think it’s vastly superior to what we have and it is a proven concept already.
