Hot on the heels of the release of seed score V2, we are excited to announce a fast follow iteration that directly hits some of the feedback provided that riders were being over or under seeded because of their weight. Thank you to everyone who provided feedback and shared your experiences in the V2 rollout thread.
At 20:00 UTC today, December 6 (convert to your local time zone here) we will be starting to reprocess historical seed values, floor values, and race results for all racers and events since the last reset on Sep 25, 2024. We appreciate your patience during this time as scores will fluctuate a bit until the reprocessing is complete.
We expect reprocessing to take between 7 and 8 hours total. Once it’s complete, we will verify the data and notify this thread that processing is done. At that time, you will be able to see your updated score. You will also be able to look back at historical race results in Zwift.com and Zwift Companion app to see how your score changed in each race.
What does the new formula look like?
The new formula is based on 90 day best 5 minute power (both raw watts and w/kg). This formula is a modified version of compound score, developed by the ZwiftRacing.app creators and community to work well with Zwift and reduce weight bias. We will now be seeding racers by mapping this modified compound score into our 1-1000 range. Because of the validations that were built for V2, we were able to iterate quickly to create and deploy V3.
Keep the feedback coming, and thanks again for the patience as we work to make racing fun and fair!
I am sorry but this is pure comedy.
P.s. I am seeing a lot of people with 999. In a good scoring system there would be a couple people with the top score, not 100s.
It loses all the appeal of competing up.
Feels like a mad house making these decisions.
There are not 100s of people at 999, not even in v2. I didn’t do an exhaustive search but even amongst pro eSports riders, I was aware of 1 person at 999 and another at 998. Also, the scores will be reprocessing for the next bit of time, so even if you see a score change, I’d wait until Zwift says the processing has finished - likely some number of hours from now - before you take a score you see as someone’s new v3 seed.
Common sense would have been for them to sandbox this first and then analyse the output. They would have seen really quickly that a lot of riders were at 999 and they could have looked at their profiles. But no, Zwift test their solutions on us in a live environment and let us, the paying customer, tell them if it works or not.
This topic stirred a lot of discussions and complaints and a good part of the community arguments were right (although the tone not always). I personally would have done a couple of things quite differently, but I really appreciate the effort some of you at Zwift put in and I want to pay some respect to you for staying with it and still trying to improve. For me and many others (a dedicated minority), racing is the heart of Zwift. Thanks!
Edit: Please dare some more transparency. If you don’t release the formula on of us will re-engineer it anyway…
There is at least 10 999s from the names I am familiar with.
Not that it matters, speaking from a gaming perspective.
V3 will fix it, v4 will bring in 7m power and we will all get 678 for some reason. V5 will of course silence the critics because Zwift product managers are so attentive and responsive.
I am pretty sure all this is an intern with excel :)))
P.s. if you were not aware that this is banter and jokes now you are. I am breaking balls, come on.
Maybe a bit of wiggle room is still needed so people can see their score drop from seed after a bad performance. Hard to say exactly because our experience of seed being floor was clouded by the inaccuracy of seed V1. 15% does seem unnecessary now. 5%? Or some fixed number.
Variable pen boundaries should be the solution to people being stuck at the bottom of a pen, not allowing scores to be dropped so far that someone can’t help but win.
if we need floor it has to be a fixed number if Pen Ranges are fixed. We can’t have higher ZRS riders droping down 2 pens and lower ZRS riders just stuck in the same place as 15% at 350 is only 50pts but at 700 is 100pts.
I think I know the answer to this, but are the power metrics that are mapped to derive ZRS, equally spaced across the 1000 scores possible. In other words, if there were 100,000 zwifters in the universe, then each point in the ZRS scale would have about 100 members with that score? Or is that, pretending an alternate universe where all zwifters were tour pros, the system could have 100% of riders with a score over 900 and nothing below that would used?
I’d like something below seed to account for injuries and the like, so the floor should be able to fall below your seeding. Unless the seed decays and is recalculated every so often.
This is especially true if there isn’t a weight modifier attached to the 5 min W x 5 min W/Kg formula. I did a race series in summer of 2023 that was organized based on CP and it was awful. As someone who weighs 105-ish kg, my raw W had me racing in the A pen even though my 5 min power was 3.5 W/kg because it was also 375 W (in old language, I was a C with a current zFTP of 310W/2.9 W/kg, my ZRS last week was 468, this week it’s 363).
Going up against the fast guys meant that every race I was off the back on the first little bump if not before. Couldn’t even stay hooked on in Watopia’s desert when they started surging less than 5 minutes into a race (and forget about anything rolling). I definitely don’t want to go back to that if there is a floor at my CP that I can’t fall below.
Yep. I was surprised by that. As a mountanbiker irl I was almost wacked during my first e-season. I was used to constant flow of jokes and insults and self destruction. And expecting a lot more banter here as we can ride holding our phones all the time.
I’m pretty sure they will use the same distribution as before. There are far less people at the top of the scale than at the bottom, it will not be evenly distributed. It is a bell curve.
Ok that makes sense, but does the bell curve determine the mapping, or the mapping determines the shape of the curve? Put another way, if a distribution curve is applied to the mapping process, scores of inactive riders could naturally move up and down as the general populations’ scores move up or down from racing. Am I making sense? lol
EDIT: Or, theoretically, are all positive score gains in a given race, added together, made to equal all score drops added together from the same race?
The curve is not forced. From the sample distributions of the Zwift population I’ve seen, it is bottom (left side) heavy. A very small percentage (single-digit percentage) of riders sit above 700 (30% of score band) in ZRS score.
Thanks. I remember seeing some of those curves, but it was hazy as to whether they were curves of zwifters who had raced (and recently), or all zwifters.
