Universal Discussion on the Universal Intensity Descriptor Known Universally as Critical Speed

This should be fun. For the record, nobody uses it because it's not actually practical. Look at the Norwegian lower mileage thread and guys and gals success there. Did anyone need it? No. It's a nice scientific theory. But practicality wise, almost zero.

@hard2find your PDFs got lost in that thread. But I found them about 10x more useful than the 32 initial and then the 21 follow up time trials @lexel wanted everyone to run and that was based on 5k time I believe. Job done. Move on. Works for 99% of people.

Also, let's hope Coggan returns to take big cr*p on lexel from a high.

I like the set-up. I can post this and lexel still won't see it coming, 👏 As he's wrapped up and swallowed up by his own ego.

This is probably the most niche but clever troll thread title I've ever seen on LRC. I hope the chaos you are hoping for plays out. I'm all for it.

The title has killed me. I don't even need to read the rest mate ha ha

@ pass the dutchie So do I. It would compliment the build up to the Olympics nicely haha

@ badger troll Thanks for the compliment!

I like to stay open to different ideas and approaches, but do agree that it seems more of an academic construction meant to delineate a “true” steady-state than a value with any practical advantage over current practice. Performance wise, the speed-duration relationship is still maintained across the different exercise intensity domains, so it’s not like something suddenly breaks down once you are running faster than CS, nor is there a sudden improvement when running slightly slower than CS.

I thought CP was more of a cycling metric, since cyclists can better handle the numerous time trials needed to optimally fit the model. Plus, they have “real” power meters. However, having seen it brought up quite a few times on the Norwegian thread, specifically by lexel, LetsDoIt, and most recently lolerskates, it seems well embraced? I'd be interested to hear specifics though from the aforementioned proponents (or opponents)...

It's ridiculous as a concept. Ludicrous actually, that it could be anymore useful in a practical sense than knowing even something like your VDOT or 5k pace etc etc. Literally anything that you can use rather than going out and wasting your time running however many time trials lexel wants you to run today. It would be laughable, if he were not actually serious.

It's a metric or starting point. Just like any other. Most of which are almost all more useful to 99.99% of the population. None of it's proponents will explain how the average hobby jogger could make use of it more than anything else. Let alone how they use it to shape their training, because, well they don't.

Lexel will come and link his old thread. Attach some papers that absolutely do not say what he thinks they say. Then he will run off again and the whole thing will be repeated in 6 months time. Remember this post for prediction accuracy.

This is funny. CV/CP is just math. I don't see why it's so hard to understand. All humans have decaying power as the distance increases. CV/CP give an idea of how much the speed or power decays. Depending on the method of determination, CV/CP can be an extremely close approximation of mlss (Or Daniels T pace).

By the way, Daniels %VO2 vs. time plot in "Oxygen Power" is basically a CV decay plot! So VDOT is essentially pace tables based on a "VO2 vs velocity" curve divided by "%VO2 vs time" curve. And then Daniels gives training paces based on some % of CV (or threshold) and some rules (6 second, etc). Both curves are based on generalized data trends from "Well-trained" runners. That's why the paces are too fast for A LOT of runners.

And you complain about time trials? How hard is it to go run 1 mile and 3 miles all-out? DO you race? That's one data point. You need to observe your limits to define them. I use 0/5mile, 1.5mile, 3.1miles as my data points.

I use like 6+ models, but the basic inverse time (1/T) model works great for me.

One of Coggan's 7 deadly sins is CP testing :)

Well, two things I did not expect from this thread, starting it off with a discussion on Daniels and being told if only I was mathematically competent, I could understand CS haha.

I was more interested in how people apply CS and/or perceive it to be an improved approach to establishing training and race pace estimates, as this seems to be the implication when people promote it. However, l’ll just reply to your comments instead and you can let me know if I have the appropriate maths base to comprehend this, fair?

You mentioned the standard 2 parameter model works great, out of the 6+ you employ. I agree, it’s direct, interpretable, and avoids overcomplicating the process. However, the speed-duration curve you refer to is only accurate for race times in the ~3-30 minute range. Essentially, whatever the duration of time one can sustain CS, that’s the duration beyond which the curve is irrelevant. Sure, you can extend the range of TT/race times you use when fitting the model, but that would then produce an inaccurate estimate of CS, which is the point of the whole process, right? Also, even if you did that, the hyperbolic function is not a very good fit of race performance across a broader spectrum, say 3-180 minutes (i.e. not a very good predictor of how speed decays with time).

You brought up Daniels fractional utilization of VO2max curve as one of the functions, along with the economy curve, that he uses in calculating VDOT, training paces, and race predictions. I’m sure you know the method by which the race predictions are made. That method is what would be required in order to extend CS so that it could be used for estimating race performances beyond the ~3-30 minute range. So basically a fractional utilization of CS curve and then applying the iterative algorithm to solve for the desirable race distance, based on a runners CS. Any system modeling speed vs duration will go through a similar process, I’m not sure what your point was in bringing that up?

I don’t know that I was complaining about the time trials. Rather I was saying that 4-5 seems unreasonable for runners, no? 2-3 seems much more realistic, but given how CS and D’ can be sensitive to what durations are chosen, I thought it reasonable to see what those with experience find are optimal durations to use for a two time trials (or race) approach.

So, how’d I do? Did that make sense? I hope so. Afterall, it’s just math, I don’t see why it’d be so hard to understand : )

For starters, I never implied YOU didn't understand math. I said that the CP/CV concept is just math and shouldn't be that hard to understand WHY people use it. Honestly, your thread title is a troll at Lexel and others who prefer to bark up the CV/CP tree, rather than piss into the wind with 10K pace +XX seconds or HMP+XX seconds, etc.

If using some of the earlier CP models, there are limitations to how long of trials you should use, but models like extendedCP (Golden Cheetah), Peronnet-Thibault, and Pinot and Grappe extend the domain considerably. For example, an improved version of Peronnet-Thibault model uses world track records from 100m to 200,000m for calibration. But these models are not really supposed to be used for prediction. They are mathematical descriptions of an athlete's physiology. And you can get a fairly good description from 2 data points (efforts to exhaustion). Skiba recommends 3minutes and 20 minutes....easy. My main use of CP/CV is to get a ballpark for mlss and practice running at mlss (or threshold) to know the sensations in order to be able to know when I'm near or over threshold to a very fine degree. There are other benefits, like workout building and race predicting, but those are also "arts" based in science.

My point in bring up Daniel's VDOT system is because it's usually referred to as an alternative to CP/CV. "Just use VDOT"....but VDOT is based on a hyperbolic curve (%VO2 utilization is hyperbolic with an asymptote at 80% VO2max) that goes out to 3h:30m. That %VO2max curve is just another way of plotting pace vs time only using O2. The curve is wrong for most people anyways, since the curve is based on a fixed economy (which we all already know). So VDOT is ok for ballpark estimates, but should not be used directly unless you know your body matches the physiology of the sample that Daniel's et al used to develop Oxygen Power/VDOT. That's all.

RE: Time trials. I don't really agree that 4-5 trials is unreasonable. But I'm a hobbyjogger who doesn't really race. I do, however, regularly do all-out efforts from 15 seconds - 30 minutes to keep my power-duration curve relevant. If you read Dr. Skiba's book, he recommends two trials (as mentioned earlier). I like two trials, but you always get R-square = 1 with two points. I find that a short trial is important for fast-twitchers with a hump-shaped power-duration curve (like me) since it pushes CP down to a more reasonable level. My partner is all slow-twitch (low W'/FRC compared to CP/FTP) and his short trials don't move CP down nearly as much as mine do. Optimal testing strategies are an iterative process and are highly individual.

How'd you do? I don't know, to be honest. You seem to have a basic understand of the concepts, but it seems you don't have practical knowledge from using the models. Do you use Golden Cheetah or WKO5? Have you tried fitting your data to a model just to see how your data fits? Have you compared your CV data to Daniel's VDOT (or whatever system you use) to see how they compare? If you want to be a better skeptic...employ CV and then come tell us why it's not working for you.

The hyperbolic paraboloid theory is the only 100% proven method of all of the imaginary training algorithms there is. /EOT

That was the type of input I was interested in, so thanks. Despite the whimsical thread title, I’m genuinely interested in the topic, meaning the case for using CS as a training basis. If I have to take a shellacking from proponents like you to explore it, I’m here for it. So too is lexel, even if he doesn’t reply haha

I’m going to Tarantino my response and start with your last paragraph first. Yes, I’ve fit my data to different models. I’ve used friends data, elite athletes data, even lexel’s data, seriously. So I’m not just coming at this from a narrow and theoretical perspective. There are a lot of different platforms and models out there, all of which I can’t possibly account for. I have however, over the years, done what you suggested. I’ve compared performance using MAS, CS, LT, 10K pace, 5K pace, 400m pace, 200m pace as a single basis and in combination for a multi-basis approach. Point being, I’ve not just used what exists but also tried non-standard approaches and derivations. If you’ve the curiosity (and in some instances, skillset), you can create almost any type of method you want for analyzing/tracking training and performance.

I’m not familiar with all those names you mentioned in your second paragraph. I do know P&T and am familiar with extended CP though, and while I can’t say I’ve seen their algorithm based on the function/fitting, it seems to be P&T (i.e. a piecewise regression method), correct? Knowing P&T, they are not using CS, defined as the steady-state speed determined by the 2 parameter hyperbolic model, they use MAS (defined as 7 minute speed), correct? So, using MAS and a few bio-energetic parameters, they estimate the speed/power one can maintain for durations of seconds out to about 4 hours. I’ve personally modeled it out to 3 hours, as my marathon time is sub 3. It’s indeed an interesting approach. It would also seem to support that one doesn’t need to use CS to accurately model the speed-duration relationship across a broad spectrum.

Bringing it back to you advising me on how to be a better skeptic… I wouldn’t say CS doesn’t work, I would say it doesn’t work any better. For example, a simple 2 parameter power function (one parameter representing your speed, the other parameter representing your endurance) would be more accurate at capturing performance across the spectrum you’re time trialing for and place none of the time restrictions on the range of durations you use, than the 2 parameter hyperbolic model. It will also more accurately extend to estimate performance across a broad range of distances, 100m-marathon. Drawing on the example of you and your partner, those differences (one being more slow-twitch than the other) would be captured by the speed and endurance parameters and not require CS as a basis.

P.S. Not that I’m inclined to defend Daniels VDOT, but do you know if he used the %VO2max as calculated from the average economy curve or the individuals economy curve from testing to develop the fractional utilization curve? Regardless, the bi-exponential function that he uses can be modified, just as extendedCP or the standard hyperbolic curve would be, with updated performance data. The result being more customized training paces and more accurate performance predictions across the 1500m-marathon. I’ve personally done this so can speak to it, practically.