Intensity vs Volume - maximising XSS might not be optimal?

Having followed XERT now for about 2-3 months of training, a few things have changed significantly from programs supplied by Trainerroad and Zwift.

One of them is very little threshold work, but a lot of significantly suprathreshold work.

The second is a large amount of time spent at the very top end of the endurance zone.

The ‘so what’s’ here are that mitochondrial endurance adaptations correlate with volume but do not correlate with meaningful intensity.

The conclusion they draw from the meta-analysis is that you should probably just work at a lower intensity for endurance.

What I mean by this is: 100 XSS over 1.5hrs is not the same as 100XSS over 2hrs. 100XSS over 2hrs is superior in terms of endurance signalling.

I feel like in XERT, the engine takes an “Area under the curve” approach. I.e. 100xss done in 90mins IS considered equivalent to 100XSS done over 2hrs - but this practice means I find myself more fatigued by my endurance sessions than I would like to be.

e.g. comparing two endurance workouts,

(1) Xert - Workout Details (


(1) Xert - Workout Details (

Have similar XSS, but if I asked you: which one would you rather do the day before Ronnestads, I think the answer would be clear.

In fact, this effect in the paper seemed relatively consistent down to pretty low levels of Intensity. They joked but suggested you might just be able to noodle about at like 100w and still get most of the adaptations.

I’ve started to find the high LTP workouts having a real impact on my level of fatigue - so I’m going to try and rebalance and do easier endurance and harder hard days.


Xert treats all XSS, divided into low, high and peak the same. We don’t go a level deeper to say this XSS is better than that XSS. If we actually knew some intensities were better than others, we’d likely change how XSS was calculated to reflect that.

Riding at or just below LTP is a staple in Xert workouts and the training advice as a way to add low XSS without having to cope with the challenge of depleted glycogen from threshold work. If your LTP has been determined properly by Xert, it should not feel easy to ride at LTP but just on the brink of discomfort. This creates the most efficient low XSS training so that you don’t have to ride for more than you need to. Having said that, if you have time then, by all means, go easier. Xert is not going to treat the workouts differently and the marginal differences that may be observed in your improvements from riding at LTP or LTP-10W or LTP+10W is still unclear. In the polarized training model they say it should be below LT1 but there are 101 ways to establish LT1 so precision isn’t really going to make a sufficient difference. Perhaps over the course of a long training period and you’re looking for a extra couple of watts, it might be prudent to be more precise on this.

So in the end we just say do X low intensity XSS and if you ask for a workout, most times it’ll have intervals in and around LTP. But not always…If you have time you can just ride at the intensity that makes you want to do the low intensity XSS. You don’t need to do a workout. Just ride. What I often say to people is that if the training says low intensity and there is an outdoor ride that you did that meets that training requirement, go out and do the same route. You don’t have to worry about intensity other than stay away from high and peak. If you simply do the mileage, you’ll likely have a final XSS that should be pretty close to what the previous outdoor ride had, independent of whether you rode at LTP or well below or above even. It’ll be more a function of the mileage in the end. Grab a coffee and treat while you’re at it.

Hope this helps and thanks the for thought-provoking question.


Hi Armando. Appreciate we end up just using the forum sometimes to exchange messages, but I do hope these threads interest other people into thinking about it.

This creates the most efficient low XSS training so that you don’t have to ride for more than you need to. Having said that, if you have time then, by all means, go easier. Xert is not going to treat the workouts differently and the marginal differences that may be observed in your improvements from riding at LTP or LTP-10W or LTP+10W is still unclear

My understanding (and I’m happy to be challenged on my interpretation) of the systematic review is to conclude the following:

  1. Riding at the top end of Endurance costs non-linearly increasing fuel than riding at the lower end, and so has the risk of putting a rider into a fuelling hole, risking poor quality intervals / intensity and burnout.

  2. Riding at the top end of endurance costs non-linearly increased fatigue, and so has the risk of putting a rider into a fatigue hole, risking poor quality intervals and burnout.

  3. Endurance signalling is remarkably flat across intensity. That is to say, for someone with a 250W LT1, the endurance adaptations from riding 4 hours at 200W are basically the same as riding 4 hours at 220W.

  4. Establishing a clean number for LTP, or LT1, or Z2, or Fatmax is hard, and it is a lot easier to overdo this number than it is to underdo this number.

Since the consequences of slightly overdoing it are 10 times as bad as the negative consequences of slightly underdoing it, and the endurance adaptations are remarkably similar, I conclude that even if volume does not increase, doing endurance work at an easier pace is superior to doing it at a pace closer to LTP, because you will simply be better able to absorb the load and have more capacity for quality in your intervals when you need to work intensity.

Moreover, you might even find you have the capacity for more volume, or more intensity as a result.

If you were to consider this in the XERT model, how might it be reflected?

Do you think that changing the rate of Low accumulation to be nonlinear with intensity (like the opposite of normalised - denormalised?) might - and this is the ideal end result - make people faster?

Either way - from my perspective I’m just going to trial picking the easiest endurance workout of the options (even if I miss my XSS target) and see if it gives me the energy and fatigue capacity to invest more in my high and peak training, or the motivation to do an extra 15-30 minutes of endurance.

1 Like

I’ve not read the whole paper (paywall) but from the figures that are visible on the podcast website I have a different takeaway. It shows that performance increases with volume, however volume is the product of duration * intensity (to give the ‘arbitrary units’).

They then show a chart showing that there is no difference between the different intensities, but I understood that to be a simple correlation of that factor in isolation. That is effectively keeping all else constant, including volume (so e.g. doubling hours if riding at half the intensity)… this makes sense to me as otherwise the first conclusion that volume = intensity * duration drives adaptation wouldn’t hold.

So 4 hours at 220w would be more volume than 4 hours at 200w and provide greater adaptation according to their analysis. More is still more (assuming you can recover, and your point re fueling / digging a hole is relevant, though I suspect highly individual). That’s in line with my n=1 experience - when I tried riding endurance for the same hours at lower intensity, I lost fitness.

That said, if you have unlimited time it is likely you can ride more than 10% longer at 10% lower intensity and so build more volume for greater adaptation… if you are time constrained I see it as more of a balancing act of getting as much volume as possible in the limited time, subject to ability to recover (can you still go hard on the hard days).

In your case, if you feel like you are not recovering and have time to ride longer at lower intensity, it makes sense to do so. You can of course also try reducing low intensity volume (same hours, less intensity), and I’d be curious what happens in your case

Happy to be corrected in case someone has the whole paper and discussion though


Thanks @wescaine for the detailed explanation. It’s in line with what our algorithm implements. As I’ve stated many times, we haven’t explicitly tried to model adaptations and physiology but rather we follow the data as it relates to breakthroughs and changes in your fitness signatures parameters and the relationship to training loads across multiple dimensions. It’s certainly not perfect but it’s 1000x better than the old way of doing things that you’ll find elsewhere. We continue to look at opportunites to improve things and have many ideas in the works.


That is in line with what xert prescribe, but very much in contrary to the way they discuss the paper. The key discussion revolves around the P value for correlation between intensity and citrate synthase development, which for volume is like .85 and for intensity is indistinguishable from placebo.

1 Like

i’m only a year or so in to the endurance world and I’ve heard this several times…not just this paper

just find it so hard to believe. it just doesn’t make sense to me

wouldn’t you after a while get so used to 220W that that becomes your “floor” rather than 200?

I don’t get it and these guys can show me any data and complex biochemical reactions but I will probably never believe that 200 watts = 220 watts. :confused:

Haven’t listened but read the transcript and at 1:08:52 they say the takeaway is that volume is more important than intensity… however as above, volume = duration * intensity (as they also call out earlier in the podcast, and as noted in the explanation of the figures from the paper)

I agree in other parts they say you can ride endurance at lower intensity, and what’s not explicitly stated by them is the need to increase duration to offset… but it is what the paper implies, and what their takeaway implies

1 Like

Just for fun 200 and 220 watts are pretty close to where i ride my endurance. It winds up being around 200 when I’m tired and 220 when fresh. (300 TP / 240 LTP)

I made a workout that was an hour of each and 200 was 61 XSS for me and 220 was 67 XSS

That just makes sense to me. How could it not be something like that?

The problem is studies like that are basically impossible…and I dunno ins and outs but I am thinking they are harder to find funding for than something more profitable (materials, pharmaceuticals, etc etc). so when I hear things that defy common sense i just don’t pay attention. I dunno…maybe I’m close minded.


You’re right, it can be confusing to wonder why ‘more is not more’, especially if you’ve not studied science much. Cliffs - I’m not a research academic, just a thick physician.

We don’t know the exact mechanism for this mitochondrial signalling yet, but a huge amount of processes in the body are rate limited or governed by zero order kinetics. Let me give you an example.

Caffeine has a half life of ~5 hours. If I drink 1 cup of coffee and ingest 100mg of caffeine, in 5 hours I will have 50mg of caffeine in my blood. If I drink 10 cups and ingest 1000mg, in 5 hours I will have 500mg of caffeiene. The caffeine exits my system in proportion to how much I ingested (this is how you think it works with training - the result is proportional to what goes in)

Ethanol also has a half life of 5 hours. If I drink 1 shot of vodka and ingest 10g of ethanol, in 5 hours I will have 5g of ethanol in my system. If I drink 10 shots, what happens is that the processes that get rid of the ethanol are maxed out at 1 shot. Ethanol is processed at a constant, not proportional rate. The other 9 shots are just sitting there waiting for that first shot to be processed. So in 5 hours I might still have 8 shots worth of alcohol in my system (and now I have a hangover).

The point is that it might be that the signalling that tells your mitochondria ‘get more efficient’ can only do it at a constant rate. A bigger dose just gives you the same constant rate. So 200w for 1h is no different to 220w for 1 hour, except that it makes you more tired and costs you more fuel.

I’m not saying that this is the exact mechanism, but it might be.

There’s also some misreading going on here in the study. Yes, initially volume is defined as intensity x load, but the point is that when these are categorised into intensity distributions, the higher categories show no more citrate synthase than the lower.

So it’s not just 220w for 1h makes no more citrate synthase than 200w, it’s FTP for 1h makes no more than 200w. Neither does sprints at VO2 - same rate of accumulation.

The fine print is obviously that citrate synthase (aka mitochondrial efficiency) isn’t the only adaptation we want as cyclists. But - it might be the best marker of ENDURANCE adaptations. So - the conclusion - do your easy work easier so you can do your hard work harder or your easy work longer.

But of course - this may all be totally wrong, but that is at least what the study is saying. I do, however, find the “it doesn’t match our model so it must be wrong” attitude curious!


One thing I am missing with Xert is the distinction between low endurance vs threshold/sweetspot. To my understanding there will be different adaptations. By the Xert model, threshold (as long as you stay just under) is not different from low endurance. You only need to do enough of it.

Also, I believe the XSS numbers are used for two things: to calculate the training signal and to calculate the recovery needs. In Xert they are the same, but it is possible that it would be more appropriate to use different curves/calculations.


Yes, however this doesn’t negate the first conclusion that there is a correlation between volume (duration * intensity) and citrate synthase. They don’t change their definition of volume throughout the paper.

For that to be true you need to ignore the first conclusion that there is a correlation between volume and citrate synthase - those three are not the same volume (duration is the same, intensity different), and so would not have the same citrate synthase levels according to the paper

None of this is to say that including easier rides as part of a structured program is a bad idea, but looking at the big picture, intensity is an important part of volume

1 Like

Thanks for an interesting thread @josephhlbusby and great insights (as always) @wescaine ! I enjoy threads like these!

This is the same as old approaches too, no? TSS is used to calculate training load & recovery needs too. Unless I’m mistaken, 1 TSS of “Z4” would mathematically take the same recovery time as 1 TSS of “Z2”?

This is (sorta) my approach. On low-intensity days, I basically assume a ~60 XSS/hr pace, which is pretty moderate (~85-90% LTP). This makes it relatively easy to plan my week, knowing that a ~120 low XSS ride will take me ~2 hrs, for example. I suspect the higher LTP endurance workouts are higher recommended since they allow the XSS to be completed in a shorter duration. If you want the time/want to ride easier, feel free! :slight_smile:

Absolutely! As @josephhlbusby mentioned, I think riding easier (which means less XSS given a fixed training duration) might help users better complete their high-intensity training. And since high-intensity workouts often accumulate strain (XSS) at much higher rates, the big picture might still be more total XSS across all training sessions, even if the low-intensity sessions are done at a lower intensity. At least that’s (kinda sorta) been my approach on things & seems to nicely mesh the suggestions from the literature with Xert’s approach that ‘more XSS is better’. Hope that makes sense…

I believe this is the paper they are talking about:

I was able to download the PSF here:

Perhaps you don’t have access to the entire article - I forgot I have academic access. These lines might be helpful?

Pooling results from 49 training studies indicated no significant correlation between relative exercise intensity and training-induced changes in CS activity (r = - 0.13;95% CI [-0.37, 0.12]; P = 0.315; Fig. 2c), not even when studies employing SIT were removed (r = - 0.01; 95% CI [-0.31, 0.30]; P = 0.971; Fig. 2d), or when the effects of training duration were ‘‘partialed out’’ (i.e. controlled for [77]) (ry1.2 = 0.23; 95% CI; [-0.02, 0.45]; P = 0.085). This further suggests exercise intensity may not be a key determinant of training-induced changes in CS activity

  1. When the impact of increased duration (not volume) is statistically accounted for, changes in CS activity flatten. I.e. it is the duration component of “volume” that is driving the adaptation, not volume as a product of intensity and duration.

  2. Higher intensities result in non-linear increases in fatigue.

  3. Higher intensities result in non-linear increases in glycolytic cost.

  4. ???

(Spend that extra fatigue and fuel on your harder sessions or making your easier sessions longer or taking more rest or enjoying other bits of your life more)

  1. Profit


1 Like

Great points all around!

@josephhlbusby First things first, we’ve never said that if something doesn’t fit our model, it must be wrong. Quite the contrary. We say “This is how the algorithm works and how we add things up.” It’s not likely to be totally accurate but it is far more useful than anything else we’ve seen. We follow where the MPA algorithm leads us rather than attempting to model a physiological process. As we see in the discussion, interpreting the science can be easily misleading or misunderstood, sometimes contradictory even. There are no misunderstandings or contradictions in being able to predict how much power you have and how long you can sustain it when we plot MPA with your power data. The better we do it, the more visible it becomes and the better we demonstrate our understanding of athletic fitness. We believe that how much power you can sustain at any given point in time, with or without fatigue, is true determinant of athletic fitness.

@hpbieker We indeed see some model improvements in how low, high and peak systems are involved in the production of power and we also see this as affecting MPA. As an example, the current model does not show any reduction in power for efforts just below TP whereas there are real reductions in MPA that exist in the data when efforts just below TP are sustained. In the Xert 2.0 model, there are changes in how we model fatigue near TP that we think improve upon the calculation of MPA and, as a result of that, will improve how XSS gets calculated (since it is a function of MPA). Indeed, this may even indicate precisely what @josephhlbusby is describing where the rate of low XSS accumulation is capped at intensities much lower than TP. In 1.0, LTP is a useful metric. In 2.0, it becomes a core parameter that enables this to be modeled directly.

I’m not sure I entirely agree with you Armando - and with respect to all of the work you put into the software - as above - I feel like you were very quick to jump on a quick reading of the paper that misunderstood its meaning, as long as it agreed with the model - when in fact the paper says the exact opposite.

Xert treats all XSS, divided into low, high and peak the same. We don’t go a level deeper to say this XSS is better than that XS

But what if it was?

There are no misunderstandings or contradictions in being able to predict how much power you have and how long you can sustain it when we plot MPA with your power data

Well, as you’ve said yourself - this isn’t quite true. If I rode at 1 watt below my TP for 2 hours (lord alive I wish I had that kind of robustness), the model would state that my MPA would still be 1400w, just as if I’d ridden at 50w for 2 hours.

It’s this difficulty of interpreting the impact of sub-TP and +/- LTP work that I wanted to discuss in this thread.

The aim being - to understand and better prescribe training so people get faster (this should be the only goal, right).

interpreting the science can be easily misleading or misunderstood, sometimes contradictory even

Yes, a fair point as wes has shown, but does that not make you want to work harder to understand it so that you can improve your model? I suppose a simpler question is, at what point would you adjust the model to reflect robust scientific evidence of a physiological process that supports a certain approach to training?

I do look forward to Xert 2.0 and appreciate the engaging conversation. If anything, it might be a great way to have more empirical data to understand questions like this.

You might have enough data to stratify training on Xert into “people who do a lot of +/- LTP work” and “people who do a their endurance at like 50% TP” and see if they are better or worse at performing intervals and/or their overall fitness profile improves and if so at a faster rate? This is something that TR are getting very good at with their analysis of big data sets (although a simpler model).


“It’s in line” refers to his explanation rather to what the paper concluded. Perhaps that’s where the mixup was? :smile:

1 Like

I can appreciate where you’re coming from, and I think it’s a pretty valid point! I also think Xert becomes easier to nit-pick on this than other platforms since we have MPA so front & center in our analysis. But as far as I know no other platform adjusts your FTP and/or power zones as you ride.

So to provide an similar example, let’s say I start a ride at 95% FTP (Z4/Threshold) and ride that for 2 hours straight (which I also agree, I WISH I had the ability to do that)… is my TSS calculated any differently 2 hours in as it would be at the start? As far as I know, FTP zones stay fixed throughout a ride, regardless of the work they’ve performed (unless I’ve missed something).

As a side note, something I think that Xert can offer that can be helpful in these situations is estimated carb burn. This year I’m working to focus more on my on-bike nutrition and trying to replace my carbs as they’re burned (up to ~90 g/hr). In my n=1 experience, I feel totally different riding near LTP after 3-4 hours when I’ve been eating enough versus when I have been under-eating (or fasted) for a few hours.

1 Like

to to provide an similar example, let’s say I start a ride at 95% FTP (Z4/Threshold) and ride that for 2 hours straight (which I also agree, I WISH I had the ability to do that)… is my TSS calculated any differently 2 hours in as it would be at the start?

I agree entirely. It is sort of an easy punching bag; because it has set itself up in some ways to get around the problems that are raised by using %FTP and TSS but this has opened up other complexities!

You quite rightly highlight one of the issues with TSS as a measure of either load or adaptation or, well, anything. It sort of measures a non stochastic effort very well in that you can compare it to another non stochastic effort well (which is harder, 5min at 120% or 10mins at 100%?) but is similarly a poor reflection of adaptive load that I think can guide wooly thinking (how many people obsess about their Trainingpeaks CSS?).

Deep down my gut says that a nonlinear accumulation coefficient for XSS might be the magic sauce here. Perhaps low XSS could accumulate a rate that reduces logarithmically with intensity (encouraging volume), with peak accumulating at an exponential rate (encouraging and rewarding people to ride in slope mode and hit quality).