This is the next installment in the series looking at Zwift racing development following on from my articles “Zwift – What’s Next“, “Matchmaking: How a Simple System Could Revolutionize Zwift Racing“, and “How can Zwift Develop a Platform for Fair Racing“.
These articles garnered a significant community response, and from these comments it felt like there was consensus amongst the community for the direction we would like to see things move. I contacted Zwift to see if I could share these ideas more directly, and in parallel a dialogue was opened up on the Zwift Forums with input from Zwift’s Head of Content and Programming, Mark Cote.
It’s great news that Zwift are now engaging with the community directly on this topic – speaking to Mark it is clear that this approach will help Zwift be more accurate as they develop a solution, as well as help us all understand what is feasible and some of the challenges the development team will face as they work through this.
So what are Zwift planning to do? What are the big challenges they will face, and what can we do as a community of racers to help?
Let’s start by looking at what has been shared on the forum thread to this point. Discussion has generally revolved around two development opportunities – a results-based ranking system with matchmaking is generally considered to be the ‘utopian’ ideal, but the more pressing issue seems to be category enforcement at race entry and improvements to better split the categories to disincentivize sandbagging.
We want fair competition, just like all of you.
Ultimately, we believe in an ELO style ranking system, like has been raised by many of you. This is not a simple add given our current infrastructure, but ultimately this is where we look to go.
If you were to make a power-based system for category recommendations or auto categorization:Mark Cote
– What data would you use to establish a rider’s performance?
– How would you categorize riders?
This was a great question to get the community thinking. First of all it is clear that a rankings/matchmaking system is a significant development – an entire new backend management system and significant changes to the user experience would be needed, but this long term vision is shared by both Zwift and the community. The question then becomes one for the short-medium term: how can we improve the category system as it exists today?
Mark then went on to share some of the current thinking:
“Thanks all for the thoughtful feedback. Since you asked where our current work and thinking is. With ZwiftPower starting here:
1. Data to Establish Rider’s Performance
We’ve been migrating the Power Duration Curves server side over the past many months and building a microservice that allows us to call power data based on a few different attributes. The power curve has a ton of data from 1s to 2hrs and allows us to pull really any data across this array. ZwiftPower takes this off of your last three races. We’re currently considering a maximal aggregation of power curve array values over the past 30 days. This would include more data that is currently considered for your category.
2. How Would You Categorize Riders?
With regards to categories, we’ve discussed the balance of race density relative to power bands. On one end, we want races to have a good field size and we want that field to be fairly matched. i.e. It would be awesome if autocategorization was enabled for ALL events and there were infinite bands, tied to pack density, but this is super difficult to do in the short term.
So right now, we have 4 categories (potentially +1 for A+). The feedback we’ve generally heard is that D’s could almost be split into three while C/B/A could each respectively be split. This might be too much, so your feedback on the 4 categories and the W/kg splits is what we’re asking for feedback on.“
This is some really interesting insight into the progress the Zwift team are making. Storing a rider’s power curve data allows for a much broader understanding of their strengths and weaknesses, which could allow for a more even splits across categories. Probably the biggest benefit of this approach is the impact on sandbagging – if the arbitrary line to separate categories becomes more dynamic, it becomes very difficult to manage your performance to deliberately stay in a lower category.
The key thing is that this category is enforced up front and from a UI perspective, a rider’s category is made clear in-game. This has also been confirmed as the preferred approach by Zwift:
“By auto categorization, we fix it on the front end, not penalize on the backend – this is generally preferred.
Following more interesting discussion from the community, Mark updated with further progress:
“We’re working on next steps right now. There are a few mega initiatives going on the software team right now that are in front of Competition Fairness so I can’t confirm timing. As this is looking to be in the near term queue, Flint and I have been engaging along with a few other PMs on calls with some of you on this forum to gather feedback and thoughts. There is active work on Power Curves and outlining the systems and rules for upcoming competition series, some by Zwift and some in partnership with WTRL. There is upcoming work on recommended and auto categorization, so this feedback will fuel those directions.
I hope this gives some clarity – I know we all want details and a launch date, but we do not have that yet.”
I caught up with Mark to discuss some further ideas and how the current work is progressing, and he was able to share the below. This is really exciting progress and I am sure you will agree with me that this dialogue with the community is a welcome change. I will hopefully be able to share more articles in the future with some insight from the Zwift development team.
“The team at Zwift is working hard to bring some of the most requested features to life, such as the recent release of route badge achievements tracking. Our forum conversations on fairness in competition shows how aligned we are with the vision of the community, but there are other priorities ahead of this work. Fortunately, with a great partner in WTRL, we are hoping to test some of these early ideas within upcoming events.“
WTRL released the Zwift Classics race schedule this week, and it appears that these races will use a 6-class system instead of the 4-category system used in the past. Is this WTRL testing some early auto-categorization ideas? We’ll get our answer soon, as the series begins July 13th.