That may be true if everyone in the race has a similar longer duration power (ie zFTP). You start with a group of 50 people and anywhere from 20 to 50 end up contesting the finish (depending on terrain) because they all have similar zFTP. Not surprising then that the person with the best sprint is going to be the winner.
The problem is, because ZRS is using 30 second power you are getting outliers put into either a category too high (for sprinters) or a category too low (complete non sprinters). 30 seconds power means squat when you are putting people with 10 minute power differences of 30 to 40% in the same category.
Will it though? See my post here just a couple of posts up from my first experience with my first racing score race knowing my seed was not achieved through any max efforts. How many races do you expect I would need to do before the seed is somewhat where it should be for me?
To be clear I think I’m actually slotted in the correct broad “350-520” category given the current options (barely), but those category boundaries are somewhat arbitrary, so I would still expect seeds to narrow into a place where it makes sense from a results perspective pretty quickly. Also, I am probably not going to be competitive in this category anytime soon, since I’m about mid pack, so if the category boundaries stay the same, and the seed scores move in small increments then I’ll largely have the same racing category and experience not being competitive year round. So I also question why the categories need to be fixed for the core Zwift races and not more flexible.
In my case literally nobody above me in the race had a score within 50 points of mine, and everyone but 5 below had a higher score than I did and my score moved only 4 points - where vELO on ZRA bumped me more than a full category for that same effort.
At most I race once per week, so I don’t race enough to have ±4 point swings make any difference over a season with the current categories personally. My seed is not based on any max efforts since I have not done a max 30second or 10min effort this season yet, so my seed ‘should’ be low.
I have only done one ZRS race so far, but based on what I saw I am questioning if ZRS will quickly move or not. Right now I think the fastest way to improve my ZRS score is to do a max 10min and 30s effort, and not via race results.
Also, I do understand you analyzed a lot of data to find the 30s/10min categories, but, if that data was mostly from races already categorized by zMAP and zFTP, then if you remove those from the equation you need to re-analyze the data. Of course if you look at people with similar zMAP and zFTP ranges then the main distinguishing factor will be 30s power at the line. But if you don’t first group by zFTP then how many of those sprinters will still be at the line to launch a sprint within their ZRS category?
did they ? not saying you’re wrong at all, but i feel like i’d remember that. i can state with some confidence that based entirely on my own observations, and specifically at my own sort of weight/power range (meaning i have no idea its the same for like an 80kg rider or whatever)… 30s doesn’t affect zrs nearly as much as 10 min
You have to consider how variable a 10 minute PR vs 30s will be in vs out of races. We could settle all this if the equation were public like they promised.
If you look at the fields it’s obviously not working right, and “it’ll sort itself out eventually” is a terrible stance since we’re seeing a couple percent move in scores max.
zrCS solved this a long time ago but again ZHQ has to come up with something unique that doesn’t work
If you could look at my zwiftpower profile and explain why I am 470, then maybe I can understand. But I still think I am penalized because of my sprinting in group events
if they really promised it then i think it’s fair to make good on a promise. personally i never saw that promise, but i did take a look at a couple of my fellow light strong brothers around my own level - of which there aren’t too many of us - and saw for myself.
i probably have a higher 30s than most on zwift at my weight by some margin and i am neither over or under seeded compared to people at a similar weight with a similar 10 min. i also don’t think it’ll really matter much in the long run
You shouldn’t be trying to predict the winners of a race. You should be trying to predict the racers that will be there in the closing kilometers of the race with an opportunity to win. That’s how you maximize enjoyment and participation in Zwift racing.
By looking at 10-minute power you’ve increase the number of high-FTP riders that will be around to keep the pace higher and make it harder for the sprinters. And that’s a good thing. But with the heavy focus on 30-second power you’ve pushed the sprinters past the level where they have a chance. And that’s not a good thing.
They did which is why Nick keeps saying “we’re going to keep it a secret for now.” I don’t know why he feels the need for secrecy since we’ve mostly figured it out.
My 30s race PR from this year is 623w and this year’s PR from smashing in a Z1 ride is 748w. That’s enough for me, a low B, to end up in fields of most A riders were I to actually enter a race.
On the other hand every PR of mine of 3+ minutes is from a race and is actually representative of race abilities.
Most people will never get where they should be if, as someone earlier described, they’re seeded at 350, finish around a bunch of people ranked 500, and gain 4 points.
My ZRS is higher than the people who came in 1st and 3rd in my ZRL race this week. I was 18th. Quite a few right around me with scores 150-200 points lower, which is where I was before I smashed some sprints in an otherwise casual ride.
I don’t really believe you. The weighting of 30s power doesn’t seem to have changed at all since you rolled out the testing in July. The formula still overseeds heavier sprinters and drastically underseeds lighter TT-style riders with big FTPs. Results from one of the World Series races today show the top 45 places being claimed by performances with over 4.0 W/kg 20m power in my category/band (520-690). And 40 of those people maintained raw 20m power greater than I can maintain for 20m. So there is no way that anything but 30s power has skewed my seed score as high as it is.
Before the last reset my race results sunk me to a racing score about 14% below my seed. And I’m sure it would have bottomed out if I hadn’t given up on doing RS races. Maybe this is just penance for occasionally finishing in the top half of most of my C races?
I think you’ve found yourself in a sunk cost fallacy situation around the seed score algorithm and the data analysis that went into it.
Your analysis has shown that 30s is highly predictive, and you’re now afraid to try anything else even though it’s plain to see that there are situations where the algorithm is failing.
You’ve tried to make it too predictive based on a biased dataset, so the result shows the same bias.
Racing Score should have a much simpler seed score and more focus on how the race results move the score after it.
@Nick_ZHQ why is it so hard to recognize that your data sets, biased by pre-sorting riders based on zFTP, don’t work here? Why haven’t you tried analyzing against mass start races which eliminate that bias? What do you think you’re gaining by maintaining a tiny shred of opacity on the algorithm? The sooner you share it the sooner we can tell you exactly what you’re doing wrong with it.
This is what I have been contemplating. It is a different thing to model “who would win” vs. “who can be there in the closing kms of a race”. For me the thing that sucks the most is knowing you’re going to get dropped at the first hill every time and seeing folks around you drop out, so you end up with a small, splintered group, or a solo TT for a large portion of the race.
To me it makes sense that the ‘seed’ does not have to accurately predict the winner, but it should group people quickly such that folks aren’t continuously being dropped off the bat. Of course the implication of that is it’s completely a sprinter’s game at that point and anyone that doesn’t have a huge sprint won’t get a podium unless they have some heroic breakaway (which doesn’t tend to happen given the current drafting dynamics), but I’m coming from the perspective of someone that has never won anything, so my concern typically is trying my best to not get dropped in any race I’m in rather than winning, ha.
As others have mentioned, I believe your data analysis is based on a flawed premise. While it’s true that most winners need a strong sprint, the fixed CE category limits mean that the winner often has both a high zFTP and enough energy left for a 30-second sprint.
However, the ZRS seeding isn’t just about predicting the winner. It’s designed to assess overall ability, providing a starting point. The seeding formula also plays an important role in setting the score floor. It’s not just about identifying the winner, but also the rider who lacks the one-minute power to compete, or the endurance to stay in the race, along with everyone in between. It covers all positions from 1st to 100th. A machine learning model would be ideal for this.
What you should focus on is finding the metric(s) that best correlate with finishing position. Using CE category-biased data may not be perfect, but determining the correlation with finishing position would be far more effective than trying to predict just the winner.
If you do your data analysis on ZRS race results, you’ll likely discover that a high 30-second power doesn’t strongly correlate with finishing position. Riders with impressive 30-second power can still be matched with those who have much better endurance. A high 30-second power, whether achieved in a group ride or a half-hearted sprint to win your “loser” group, doesn’t correlate with victory.
Endurance is key—riders need the stamina to keep up with multiple breakaway attempts, surges on climbs, and the ramp-up before the sprint. While 10-minute power covers some of that, without 3-5 minute power, riders can still be dropped during attacks.
This is what we see in ZRS races now. Fair? Fun? Well…
If you have to select just two power metrics, my suggestion is to scrap 30 sec and replace it with a power range that is more important and used more often during the race. Not just if you happen to be in the mix.
Just an idea: do a survey where you ask people if they are happy with their scores, if they think they should be in a higher/lower bracket, etc. Then you’ll have enough data to confirm that your calculations are completely wrong, and you should scrap this formula completely.
Explain to me how I have a score of 527 and should compete with high B and A riders. When I am a middle (at best) cat C rider, I can’t do hills or endure long rides (more than 10-15 minutes for me is a “long” ride), but I can do relatively above average short sprints, so I should be punished and placed in groups where I’ll be dead last to finish, alone and ashamed with my finish times. Guess what!? I just won’t participate in this!
And this leads up to this :
Zwift World Series Race 2 at 18.30 BST. This race has a pretty good entry so gives a fair idea of the spread of riders within each category and relevant start point dependent variables Zwift consider important to group riders into a pen giving a better and fairer race experience for all.
OK so you may be shocked to see actual 20 min wkg comparisons within each pen.
Pen A - Ranges from 3.7wkg to 6.0 wkg
Pen B - Ranges from 3.1 wkg to 4.7 wkg
Pen C - Ranges from 2.8 wkg to 4.5 wkg
Pen D - Ranges from 2.4 wkg to 4.0 wkg
Pen E - Ranges from 1.7 wkg to 2.5 wkg
That 30s wont mean anything ecept those at the top of those w/kg since all other will be back minutes and dont have any need for sprinting anymore.
If you want to make a valid argument, perhaps use an example that is not there today. There are plenty of people who are bottom of category and dont have that opportunity today.
It might now be different people competing for the win and experiencing the back of the race, but there is no difference with people being dropped in a CE race & people being dropped in a ZRS race.
There will always be a back of the race… Variable boundaries would give the opportunity to ensure this is mixed around a wider group of users than fixed boundaries. The current change has just shaken up who is competitive and who is not in a fixed boundary system.
There are people who are forgetting a small detail: Compound score for 10 minutes is not the same as w/kg for 10 minutes.
You have to multiply w/kg by raw power.
