Pack Dynamics 4 Release [April 2023]

You or Zwift made mention that there is an added “speed coefficient” (my terminology) to a small breakaway group for PD 4.0, in order to help establish separation.

What? I would love to see where that was mentioned… certainly not by me.

I believe Zwift applies draft mechanics for a 4 to 6 man single-file paceline and extrapolates the same power savings that 6th (or final) person receives and applies it to any position beyond 4th/6th in a mass peloton. That’s highly flawed!

No it does not. Where did you get these ideas from? Zwift applies the physics of the player (the one running it) depending on their current surroundings. Slope, rolling resistance and wind resistance, with the wind resistance being subtracted or not, depending on the draft situation. Draft takes into account the currently surrounding players position and the potential draft they provide.
It’s blind to large groups or small groups. Doesn’t do any “magic” based on that.

The problem with that realistic model you posted, is that if we don’t do any draft cap for riders that are deep in the pack doing steady watts, they will fly to the front because there are no collision consequences and because their draft savings are massive, so we need to introduce these unrealistic safeguards to account for the lack of collisions, etc.

  1. Position is crucial in bike racing. Much more important than fitness! Patience is an aspect that absolutely should be rewarded, as should taking chances.

if we get complains as it is now, imagine if you would have to tell people to wait for some action while you can’t steer or move away from a certain position. I believe the experience would be quite frustrating.


Half the comments complain that pack speeds are still too fast for breakaways to succeed. Half the comments complain that pack speeds are faster now because people know that breakaways can succeed. Half the people in each group are also members of the other group. Half the comments approve of PD4. Overall, I think that’s a win for PD4.


To your 2nd and 3rd points (out of 3 total), coupling draft and position/auto-braking:

I made the hypothesis based on required power input settling deeper in a 50+ rider peloton. You have proven that hypothesis — not theory — to be wrong. That said, “draft cap” is very unacceptable, as it then gives bias to rider fitness. I get that this was the framework to date to TRY and circumvent missing auto-braking, but “racing” is not solely a “workout.” Racing is a tactical battle. A game theory of strategies to accomplish goal(s) in competition. Position and luck contribute. Racing is to test your fitness capabilities

Draft cap + pack churn leads to acceleration as the finish placing determinant. It is why in Cat. A the “1 minute specialists” gets brought up frequently. They are the phenotype (puncheur) that wins Zwift races the most. We can argue short formats etc. as factors, but it boils down to Zwift physics flaws as you responded at length to


Thank you for catching this. I can illustrate numbers under the following assumption: drag force at 80% and rolling resistance + gravitational + other losses at 20% power required

Suppose lead rider does 400W. 320W is to overcome drag at their velocity. The rider at 5% in peloton then needs 16W power at their pack position PLUS a 20% of total power input to maintain their momentum/inertia. The former overcomes drag; latter overcomes the other force losses. I reckon it still yields 100W total or lower power required?

David already mentioned the “draft cap,” so we cannot (unfortunately) dive deeper into the analysis. We do know that power required exceeds 100W because of the cap

It is why in Cat. A the “1 minute specialists” gets brought up frequently. They are the phenotype (puncheur) that wins Zwift races the most. We can argue short formats etc. as factors, but it boils down to Zwift physics flaws as you responded at length to

If we want to compare Zwift physics with real life physics then we also would need to make the question, would those same riders be successful if the average Zwift races were between 3 and 5 hours long like regular IRL cycling races?

That said, “draft cap” is very unacceptable, as it then gives bias to rider fitness.

I understand your frustration but I also already explain why we can’t avoid these “fake boundaries” for the time being: steering, manual braking and collision consequences.

1 Like

10 posts were split to a new topic: The success rate is much greater on Zwift than IRL

[quote=“Daniel Jamrozik [+R], post:294, topic:605018, username:Daniel_Jamrozik”]
I reckon it still yields 100W total or lower power required?

I have been riding for many years in big groups and I can say from experience that you need a lot more than 100w to stay in a pack that big.

Flat 8km section of: (more than 50 riders)
Lap1: pulling and riding in the front 41km/h 300w average
Lap2: hiding in the bunch 42km/h 265w average
Lap 3: got dropped and had to TT to the finish 40.2km/h 316w average

@DavidP I’m trying to understand how the auto-braking works. I wonder if this parameter is in use and if it is 40% of 10 sec avg power, 40% of current speed or something else? How long does it last? When is it applied? You have already said that weight/power does not matter, so I wonder what this 0.4 refers to.

Are you referring to IRL group rides? Then I’ll establish conditions to make it clearer:

  1. Velocity is steady state. That pulling power of 400W is steady, so no attacking off the front nor surging. It is an instantenous snapshot that assumes said power to be constant
  2. On the flat. No corners. No wind. No external drafting to vehicles. No slow downs of speed
  3. Where in the bunch were you hiding? Id have to count the number of riders on aforementioned CFD graphic to determine the 5% power required to overcome drag force
  4. Its very important you mention both avg power (AP) and normalized (NP) and/or variability index (NP / AP) for your efforts. Ive seen flat criteriums where riders have AP of about 200W for prolonged periods while the avg speed for that duration is around 49Km/hr. NP suggests that effort typically to be unsteady

(1) greatly implies no acceleration and deceleration effects.

No accel/deceleration of (1) means that VI should theoretically be 1.0 and NP can be ignored from (4)

[quote=“Daniel Jamrozik [+R], post:300, topic:605018, username:Daniel_Jamrozik”]
Are you referring to IRL group rides?

IRL racing. It was not a crit it was a 70km race that had 3 laps and that was a flat section no turns.

There was surges just like any Zwift race.

That is the amount of braking force applied. Imagine you have a brake lever on your hand and you press 40% of the way in, instead of squeezing with full force. So it’s a semi gentle tap to the brake :slight_smile: In the early versions of PD4 it was configured to 60%.


Then my hypothetical comparison is apples to oranges against this particular dataset you share

Perhaps you can narrow your analysis into smaller time windows such as under 5 seconds, where the conditions I specify are satisfied? The goal is to remove the accel/deceleration and surging components in your example dataset

In other words, my 400W example is steady state dynamics. Your current data is transient dynamics

When is steering coming to the Tacx bike? Or the companion app? Or to arrows on a keyboard? Should be trivial.

Under 5sec.

With such a short time I can probably find a sample of anything.

Front doing 1000w and rider in the peleton doing 0w.

Or front doing 5w and rider doing 500w.

IRL we have surges same as in Zwift. If IRL you need to average x to stay in the draft then I would like to experience the same feeling in Zwift.

Zwift is not simulating IRL they make training fun and feel like cycling.

1 Like

I’m with Daniel.

I vote realism.

Use real world cda’s.

At least enable a “Pro” physics mode for racing. That would be a great first step. I still think we should have a “Breakaway” powerup. Each rider would get say 5mins of total time per race. This would simulate riding in the drops or a bent arm aero position. Riders could use it tactical, like we do in real life. All at once, in bursts, in defense, whatever. That would allow the low cda temporarily, while turning regular Zwift cda’s back to something more realistic. It’s not gamification. It’s part of how all road racers ride. It’s missing from Zwift. I believe it would help make racing more tactical, even with the current physics.

Any chance you could get this breakaway powerup into development? @DavidP

Bike racing IRL is more tactical, for many reasons. If Zwift simulated reality better, the racing would be more dynamic.

That doesn’t mean it needs to be a pure reality simulator. Folks could still have their dinosaurs, if that’s what they enjoy.

I would just prefer real world cycling physics. We have the data. Why not make that a target?

In regards to records, I personally couldn’t care less. Why would anybody truly care about a make believe time over a make believe course? I get that it’s a motivation for some, but long historical data… that seems a little extreme. Just set new times under the new more accurate physics. It’s not the end of the universe. In fairness, I imagine some would revolt. However, they’d adjust eventually.

Look up any segment, they’re filled with ridiculous times. They could all do with a reset.

Hit the big red button :grin:


This thread would be a bit easier to follow if either:

A) People removed square brackets from their usernames; or

B) People quoting people with square brackets in their.usernames fixed their quotes.

Thank you and carry on :smiling_face_with_three_hearts:


In the way you mentioned, no. I don’t have the bandwidth for that any time soon.
If it’s of any consolation, the Burrito is going to be “fixed” very soon to work as a real breakaway power-up, but it won’t last 5 minutes :sweat_smile:

At least enable a “Pro ” physics mode for racing.

I do hope to get to work on that…


I did! :stuck_out_tongue:

Didn’t get a prize for it though. :frowning:

1 Like

The problem is no collision and no server side positioning. Unfortunately the idea lives to that no collision can only exist with steering. That is nonsense.

Let server determine positioning based on a ‘simple’ minimization process that takes into account speed/watts of surrounding riders and optimal pack positioning. Moving up and down is based on what intention you signal with your speed/watts and moves you as efficient as possible within the pack. It is not dissimilar to what is attempted (hacked) now but currently one can ride through other riders which despite the hacks currently applied still creates forward momentum for the entire pack which leads to unreasonable speeds.

It also, as Daniel effectively mentions, reduces the game to a simple sit and sprint. Whether its pd3 or pd4. He mentions lead out trains… In Zwift you can sprint yo victory from position 100 within the last 400meters. No tactics required. Which is a detriment to the game. The aforementioned server side positioning which prevents collisions would solve it. Correct definition of pack shape would then also make corners potentially a source of tactics and pack shape would be automatically long so even without corner braking you want to be in front. Anyway, these things can’t be hacked on client side.


Unfortunately the idea lives to that no collision can only exist with steering.

Let’s assume we do have collision detection and some kind of consequences derived from that.
If your only input to the game is power, how would you feel if you get penalised for colliding into someone where you have no control over it because your position is automatically determined?
I think it would be quite a frustrating experience :thinking: