The Evolution of Testing: Continuous AeT

Steve House:
The problem has not been effort. It’s been uncertainty. And this is just one tool that we’re developing to help close that uncertainty gap.

Steve House:
Join me live on our YouTube channel June 30th, where I’m going to talk about continuous heart rate monitoring and how you can set your training zones with confidence. If you’ve ever wondered whether your training zones are right, come join me — the link is in today’s show notes.

Steve House:
Every endurance athlete who trains with a heart rate monitor has had the same experience. You head out for your easy aerobic run, you settle into what you think is the right effort, and you finish the workout maybe not sure that the work you did matched the work you were meant to do. You had a number, but you just weren’t sure if the number was right. Today we’re going to talk about how we’re trying to close that gap. We built a tool inside Uphill Athlete that reads your training every week and tells you where your aerobic threshold actually is — based on the training you’ve done, not on a special test. And then we did something equally important: we tested it against ourselves. We ran a statistical analysis of the programmatic way of determining aerobic threshold, the manual way of doing it as a coach, versus having the athletes do it themselves. So we’re going to tell you exactly what we found — including the parts that maybe aren’t that flattering. And I’m joined today by Mr. Will Zittlau, who led the build of this on the software side. He’s a technical project manager by trade. Will, thanks for being here.

Will Zittlau:
Yes, thanks for having me. I’m excited to talk about this. It’ll be fun.

Steve House:
I think it’s a pretty important problem that we’re solving, because the problem is sort of the basis for so much of what we do. It’s something I’ve been working toward for a long time, and it’s pretty fun to finally have the first parts of this be visible to people. One of the insights I’ve had from my last ten years at Uphill Athlete is that athletes do care about the work they do. They put in the training, the long slow distance; they’re consistent, they’re patient. But what they often can’t do on their own is know whether the effort they put in is the right effort. The one number this whole thing turns on is your aerobic threshold, which we call the AeT. That’s the point — just a little refresher — where your body’s reliance on fat as a fuel source starts to give way to carbohydrate. Your aerobic and glycolytic systems, as we sometimes call them, are always working together; there’s no switch that flips hard from one system to the other. What changes as you go harder — train harder, run harder, climb harder — is the balance. Below your aerobic threshold, fat oxidation is dominating and lactate production and clearance are in balance. When you’re training in that zone, you’re building the adaptations that make long, slow endurance possible. If you go into the glycolytic — into carbohydrate burning, where carbohydrate has taken over as the dominant fuel source — lactate starts to outpace clearance, and fatigue builds faster than it did at a lower heart rate. So here’s the problem we set out to solve. You want to do a workout that builds your aerobic engine, but if your aerobic threshold number isn’t set correctly — especially if it’s set too high — the easy workout you thought was building your aerobic system is actually pulling you into anaerobic metabolism. You’re taxing your body in a different way, and that has a different recovery cost. You didn’t train the part of your physiology you intended to. The whole system is off balance because of this one fundamental miscalculation. So the problem isn’t effort — it’s this uncertainty. We’ve solved this in the past with lab testing and aerobic heart rate drift tests, like the MAF system developed by Phil Maffetone, who we’ve had here on the podcast. And for years I’ve been wondering if there’s a better way to solve this. And you helped me do that.

Will Zittlau:
Yeah — and echoing what you said, everyone kind of knows upper Zone 2 is maybe the approximation people use for that. But depending on how often you’re updating your heart rate zones in whatever health-metrics platform you’re using, that can be out of date. And like you said, it’s super important to nail, because if you’ve subscribed to the Uphill Athlete training methodology, you know you’re spending a lot of time in those lower zones — so you want to make sure you’re actually doing that accurately. So this is kind of where that problem came up: how do we close the gap on that uncertainty?

Steve House:
Yeah. And it’s important to pause there for a moment and say it doesn’t matter, in a way, whether you’re using RPE or strict zone-based training — you still need to figure out what exertion level you’re running aerobic, and what you’re running when you start to become more glycolytic, more anaerobic. And again, it’s a continuum, not a black-and-white situation. This was the gap that, as a coach, I filled for a long time by simply going into athletes’ training files, looking at places where they’re aerobic, and extrapolating their aerobic thresholds from workouts as they go. Because, as you said, the aerobic threshold changes over time — as you get fitter the heart rate goes up. This is where RPE actually really shines: your RPE doesn’t change. It still feels like you’re running an RPE of 2 or 3, but you’re going faster, because your heart is beating faster and your metabolism is too. So that’s one of the advantages of that system — it avoids the aerobic threshold test you do by running around a track, figuring out your pace versus heart rate and whether that’s decoupled or not. And for each of those tests you have to taper a little bit, fuel properly, have a good day, be in a particular environment — and then do it again every six or eight weeks. So this is going to take care of all that and make it a thing of the past. Will, I want to turn this over to you. You’re an athlete in the system, not just the person who helped build it. Before any of this existed, what was this uncertainty actually like for you?

Will Zittlau:
Totally — this definitely hits home. Last summer I was training for my first 100-miler, so a lot of time in Zones 1 and 2. As part of that, I had the opportunity to hike the GR5 across the Alps with my partner, which was awesome — hiking with a heavy pack, lots of lower Zone 1 and 2 work. That was a month, a couple hundred kilometers, 30,000 meters of vert, and I came out of that trip with a lot more fitness than I came into it. That’s right where my big volume weeks started to build, and to be honest I was kind of just guessing — back to having to do these controlled tests. I had an idea of where my AeT was, but I was just going off RPE and winging it for those big volume weeks. So having something like this that I could’ve checked to see that progression would’ve been awesome — to then hit that transition. Because sometimes you do those really big build periods and you come out feeling like you have a lot more fitness, but you don’t actually know how that translates.

Steve House:
Yeah, this is spot on. The problem isn’t that people aren’t doing the training — it’s that people are uncertain whether they’re training at the right intensities. That’s what we’re trying to help them fix. So let’s talk about what we actually built. You want to give us a walkthrough of what continuous aerobic threshold is and what it’s doing each week? Maybe pull up the dashboard in Training Groups and give us a little tour.

Will Zittlau:
Yeah. So what you’re seeing here, if you’re a member of Training Groups — this is the Training Groups dashboard you get to see. Right front and center, we’ve put those thresholds we’ve talked about, with AeT being the dominant one of this conversation so far. But we also have AnT, and then your AeT–AnT delta. The goal is to actually show your progression. As anyone who’s done fitness training knows, you can go weeks without feeling like you’re making any progress, and then one day you stick with it and a switch flips and suddenly that grade starts to feel easier. We’re hoping this bridges the gap so you can actually see that progress — even small, subtle changes week over week that you might not notice physiologically, we’re hoping we can pick them up and show you that trend line of your fitness growing. So that’s what continuous AeT is: a four-week moving average, showing your progression over time based on the workouts you’re loading. On the other side — a lot of fitness platforms will show you this number; AnT is a bit easier to calculate — AnT is going to be the upper end of your Zone 3, that threshold heart rate. And then the third value we’re showing is the actual AeT–AnT delta, which has some interesting implications that Steve can talk to a bit more.

Steve House:
Yeah, thanks Will. This has been the vision I’ve had for a long time — to make the aerobic threshold visible to people. And you helped me realize we could actually calculate AeT on a daily basis. But there are a few reasons we don’t want to do that. One is that it changes a little from day to day depending on your rest and fatigue, hydration or dehydration, heat, altitude — all kinds of variables could affect your heart rate and your threshold that day. And we have to remember — we’ve talked about this from day one of Uphill Athlete — we’re not inventing anything new. We’re just taking conventional endurance-training understanding and translating it for non-conventional sports. Since we’re not running around a track where we can easily use pace to measure, and we’re not riding bikes where we can easily use watts to measure, we need this tool to make our fitness progress visible. And if we look at it every day, it’s going to drive us crazy — and by default, because we’re human, we’re going to give it too much importance. So we need it to be a little bit fuzzier on purpose, so we’re not over-obsessing about the number and how it’s changing day to day, because that’s not the reality of how our bodies behave and respond to training.

Will Zittlau:
So Steve, we’ve talked a lot about AeT and what we’re trying to do with it — but how would athletes actually get this value before? Like with your coached athletes, what were you prescribing for them to update this manually?

Steve House:
Well, with some athletes we’ll go do a test in a lab — that’s still going to be the best, and we’re not replacing that accuracy. But in a way we’re doing that on purpose; I want to make that point. What we want is a field test — going out and doing the heart rate drift test. While it works and it’s good, and you can still do it… What we found was, when people were doing it — and I’ll talk about that; we’re going to get into the statistics in a minute, but this is really interesting — in Training Groups, people use the heart rate drift test and then set their own training zones from it. I went in and hand-calculated — I think that’s the best word — hand-calculated training zones for people, including looking at the heart rate drift tests. And in my experience they were consistently four to six beats too high for what they’d decided their top of Zone 2 was. So this points out the problem: it’s just hard and complicated to do, it’s not easy. So I wanted to find another way to make this less difficult.

AD READ:
If you’ve been wondering whether one-on-one coaching is the right next step for your training, this is the month to find out. Sign up for coaching and receive a free 30-minute call with one of our specialists. Visit uphillathlete.com to learn more. Now back to the show.

Steve House:
I think the important thing to talk about is just how we validated this.

Will Zittlau:
Once we developed this algorithm, we needed to test it. What did that actually look like in the validation process? Steve, I know you did a lot of manual calculations on your end — maybe just run everyone through how we actually validated our work.

Steve House:
Yeah, great question. There’s a pretty long history I won’t get into — I’d be happy to tell people if they’re ever interested, but I just don’t think they are. I’ve been working on this for almost two years now. Before you came on the scene I’d been working on it for a year and had the calculation piece sort of figured out, using a bunch of our athlete data — figuring out a way to calculate this, what the filters are, things like that. But I wanted to figure out a way to replicate how I would score athletes if I sat down with their data myself. To the point about lab testing: there is lab testing, there’s ventilatory-threshold work we’ve talked about — especially in Training for the New Alpinism, like the first threshold being able to speak while you train, and so on. There are well-run field tests like the heart rate drift test. They can all give you good data, but coach judgment is the closest thing we have to what the athlete is actually experiencing. So that’s what I wanted to test it against — actual coach judgment.

Will Zittlau:
Yeah. Once we had this in place, we pulled the training data for 128 of the Training Groups athletes and ran this continuous AeT against six months of that data. Steve then sat down with 65 of those files and hand-scored them in a blind test — he didn’t know any of the scoring coming into it. He was just provided with the data and did what a coach does, validating his own approximation based on what he was seeing.

Steve House:
And it’s important to make a couple of points. One, I didn’t do all of them, because it just takes a really long time — we got about halfway through the data, and statistically we had enough to make this work. There are some learnings in there we’ll talk about in a second. I again want to say the hand-scoring is a better yardstick, in my opinion, than a laboratory number — not because it’s more accurate, but because it’s more timely. The problem is, unless you’re a professional athlete, you’re not going to be able to get lab-tested every week, or every two weeks, or every month. That’s just unrealistic. So we want to keep this based in the lived reality of being an athlete — being an uphill athlete. Let me talk about the numbers a little bit; we’ll link to the blog I wrote about this so people who want to can go see the actual data and all the math behind it. Of the 65 athletes I scored, the average difference between the tool’s number and the number I chose was one-tenth of a beat per athlete. So that means — almost zero — it was basically the same as what I scored, on average. As a group, the tool isn’t running systematically high or low; it’s centered right on top of where a coach lands. But there’s an important caveat, and I want to be transparent about it: not all the data is equal. Some was actually quite poor, some pretty good, and some high quality. So what I ended up doing — I didn’t foresee this, but it became a natural part of the process — was ranking each athlete as low, medium, or high quality data. When we split the athletes out according to how clean — how high quality, or well recorded, or well organized — their data is, a couple of different stories came out. For the 34 athletes whose data was clean enough for me to score with very high confidence, the tool read two heart-rate beats lower than my number. So if I’d said the aerobic threshold was 130, the tool would’ve said 128, on average, for those 34. For the 17 athletes whose data was sparse or poor quality, the tool read about six beats per minute higher than mine — which kind of makes sense: the formulaic approach, when the data is more variable, just doesn’t produce as good a result. Garbage in, garbage out. The mixed-data athletes were in between. On average those two errors sort of cancel out — but for an individual athlete they don’t. And the direction of this matters. We’ve also seen this in reality; Will and I have both had a bunch of messages about this from athletes in the Training Groups, because they’ve been using it for a while now. When you get to actual athletes, the people who have clean data — they’re recording it, marking everything correctly, using either a chest strap or an armband, or if they’re using a wrist-based monitor they’re getting good readings from it — not everybody does. If they’re following these basic good-data-hygiene protocols, they’re getting pretty good results. It’s the people switching between different modalities a lot — especially, unfortunately, cycling, especially mountain biking — that we’ve had to just rule out. We just don’t look at mountain bike data, because think about it: mountain bike data is completely the inverse of what we want. The heart rate is highest when people are going slowest, and lowest when they’re going fastest, going downhill. So there are some things we learned in the process that helped us improve our filtering. It taught us a lot. But I want to be very clear: it’s still very much a process. We’re still working it out. There are still a ton of cases, and I’m confident it’s just going to get better over time as we keep working on it.

Will Zittlau:
Yeah, 100%. It’s good to be clear that we’re not claiming we’ve solved this problem outright — we’re going to continue to learn, and our process is going to keep evolving. Once we got it in the wild and actually got user feedback, as opposed to just data sets we were running our own experiments against, it’s been good to see the general feedback. As you mentioned, that data filtering — I knew that was going to be an area of challenge, but in real-world use there’s so much variety. The real world is very dynamic, so it’s hard to have a one-size-fits-all solution, but we’re trying to get there to fit the average athlete. The biggest problem we were seeing is heart rate drop-off in the middle of a workout, or big spikes. If you use a wrist-based watch, that’s a lot more likely to happen — so, back to the data-cleaning thing, a chest-strap heart rate monitor is going to give cleaner data. And depending on the workout — obviously, running on a track or a road is going to be more representative than a trail run, just because of terrain variations that can skew the stats. Once again, the real world is very dynamic, and solving that from a data perspective has been a fun challenge. Every week we learn more, get more feedback, revisit, and say, actually we need to look in this direction or adjust this. So I do expect it to get more accurate over time.

Steve House:
Yeah. When I first did the statistical analysis and it came back that the average was only 0.1 beats different from my hand-scoring, I was flabbergasted — at first overjoyed. But my confidence was short-lived, because I actually got into it. Another way to break this down: 62% of the athletes had a number within five beats of my hand-score, and 77% were within ten beats. So that’s actually, to be honest, for me, not that great. It’s the right direction — we’re going to get there — but I want 99% of the athletes to be within five beats of my hand-score. That’s ultimately where we want to go. And we’re seeing some anomalies — people 17 beats up or down from my hand-score. That’s just not reliable, not actionable. So one of the reasons we’re running this in the Training Groups with real-world athletes is that it’s a big enough group to have statistical relevance, but small enough that we can still really communicate with them. I can still go in and hand-score athletes from time to time and help figure out where they’re off. It’s really helped inform our improvements — especially filtering; the data filtering is one of the places where it’s really useful. And teaching the group — I hate to say it, but we’re kind of having to teach the group some data hygiene as well.

Will Zittlau:
And for anyone confused — because we led by saying we were within one-tenth of a beat — that’s the average. Obviously that disagreement range is going to be a lot wider per athlete. The average takes into account positive or negative variance. When you take the average it’s actually one-tenth, which is awesome — but then on a single athlete, plus or minus five would give you an average of zero, even though both of those athletes were off.

Steve House:
And that’s where we’re trying to continue to improve — continue to run this in the real world with real athletes and figure it out. So yeah, we’ve still got some ways to go. I think we’ve had a really great response from the Training Group athletes — they’re really engaged with it, and several of them are truly geeking out with us on the data: how to filter it, how to calculate it, all that stuff. That’s been really fun. We’re going to get there, and we’ve actually made huge improvements in a really short time, so I’m pretty excited about it. Anybody who wants to look at the statistics — the full write-up is in the blog post, which we’ll link to from this podcast episode, with all the numbers there.

Will Zittlau:
And so once we have this data computed, is this just being applied to athletes, or what’s the vision there? Okay, the system’s found this value — then what?

Steve House:
Yeah, great question. This is actually very fundamental to our whole approach with data within Uphill Athlete: the program suggests, and the athlete always gets to decide. Ultimately, if we’re able to get this good enough that it’s useful for one-on-one coaching as well — where it’s very, very precise and coaches can rely on it — even then it’s going to be very much a suggestion to the coach, an additional data point. It’s never going to replace them, and that’s super important. You can accept this and then adjust your zones — or not. We’ve done this; I did it with two Training Group athletes just today, said, “Hey, that number’s not good, I don’t think it’s accurate, I think you should use something else.” And then we build their zones off of that. Along the way we’re learning what all these edge cases are. So I think that’s the right approach. The wrong approach is to have too much confidence in this. It’s not magic — it’s still just a process we’re applying.

Will Zittlau:
Definitely, and I think we both fundamentally agree on that. As software development evolves and gets potentially easier to coordinate vast amounts of data — at the end of the day there’s still decades of expertise under each of your coaches. These tools should be used to surface insights, not be used deterministically.

Steve House:
And that’s fundamental to the whole philosophy we have toward data as a coaching team. So one of the things some of you have foreseen is that you could have your training zones calculated for you, basically — because now you’re going to have your aerobic threshold suggested, your anaerobic threshold suggested from TrainingPeaks or Garmin or whoever’s originating that number. And then you can build your whole zone system, and convert that to RPE if you wish, or stick with the heart rate zones. That’s kind of the next step, and we are working on it. We talked about it in the livestream last week, so it’s not a secret. There’s still more to do, and it’s still really interesting to work on — but we have to be very realistic about how accurate this is. That’s one of the reasons we just want to be transparent: hey, this is working, it’s working really well, and it’s going to continue to get better over time. My prediction is that by the end of this year we’re in territory where it’s very accurate, we’re very confident in it, and we’ll be able to show that through these kinds of statistical analyses.

Will Zittlau:
Yeah, I’m excited for it. Just to close off — what does this mean for the average athlete, if you don’t have access to those lab tests? What does this really unlock? You mentioned you’ve wanted this for quite a long time.

Steve House:
Yeah. Once we have continuous aerobic threshold suggesting to you every week what you’re at, you can adjust your zones and train that way. It’s important to understand that at the end of the day, when you go into a training block — let’s take a week, for example — what you’re trying to do most weeks, except for recovery periods, is sort of embarrass your system just a little bit. Just enough to show it, “Hey, you can’t quite do this. This is what I want you to do.” So by having continually updated aerobic thresholds, you’re more able to keep nudging that aerobic threshold up — embarrass your system, so to speak, a little bit more every week. So you’re going to be more effective in your training, and you’re going to have this visual tracker of your aerobic development. You could see it — because this doesn’t happen day to day. It’s not like strength training, where you notice you can lift five pounds more this week than last. It takes longer and it’s harder to see, so making this visible is going to be really important. And if we zoom out, this feature is just one visible piece of what’s going to become a larger picture, where we’re trying to scale clear, science-based human coaching to more athletes than one-on-one coaching can reach — without diluting the one-on-one coaching. There are sort of three layers to this system in my vision. One is the coaching — the high-touch, one-on-one, where it’s a one-to-one relationship in terms of determining what’s best for a given athlete on a given day, week, training cycle. Versus a one-to-many system, which is what we have in Training Groups. Training Groups, in a lot of ways, I see as a continuity layer — people want to train for typically three or four months, they do their climb or run their race, and then they want to continue, at least maintain. They maybe don’t need that full one-on-one dedicated high attention — plus, sometimes, a lot of pressure — and they can drop back into Training Groups that have that continuity. And then we’re building a tool we call Maria, which I think of as the connective intelligence layer that supports all of that — helps us understand what athletes need and when they need it, and helps the coaches better support our athletes. It holds together the athlete methodology, holds the context of training, and it’s always in a proposal role — never in a deciding role. Whether you’re in Training Groups, where you’re the governor of your own numbers and accepting suggestions, or you’re a coach getting the suggestions — your work is able to be made faster and more accurate. The principle under all of this is that, just as we’ve always done with training, as we get new tools, those better tools should make us better athletes and better coaches — help us scale human expertise, protect that one-on-one relationship and mentorship, and provide more structure with less uncertainty for more people.

Will Zittlau:
100%. I think it’s a really exciting time to be an athlete right now, actually, with all the new tools and data we have access to these days.

Steve House:
Yeah. And I’m so grateful — I just want to say I’m so grateful for the Training Group athletes. I was a little hesitant at first, like, “Oh man, they’re going to feel like guinea pigs or whatever.” It’s been the opposite. They feel like they’re on the cutting edge of something really exciting; they’re fired up and really contributing. If it wasn’t for them out in the field doing their workouts every day — and adjusting their training zones, in some cases every week as their threshold changes, and continuing to progress — we wouldn’t have this laboratory in which to build this. So I’m really grateful for them.

Will Zittlau:
100%. I know it’s only been a couple of months, but the community we’ve built working alongside them has been awesome. Everyone’s been so great to deal with and talk to. It’s pretty cool.

Steve House:
I appreciate all of them. So any Training Group members listening — hats off to you guys. If there’s one thing that captures all of this that I want to close on, it’s this: the problem has not been effort. It’s been uncertainty. And this is just one tool we’re developing to help close that uncertainty gap. We also want to be transparent — we want to show our work while we do it. So again, the full analysis, with all the numbers and all the parts, including those that are not so flattering, is on the blog; we’ll link to it in the show notes. If you want to see the dashboard yourself, live, we’re doing a live walkthrough on YouTube on June 30th — seven days after this drops — and we can show you then. And if not, come join us in Training Groups; we’ll have a sign-up link in the show notes. It’s a really dynamic community and we’re making it better basically every week at this point. So, hey Will — thank you so much for all your work on this. I couldn’t have brought this to life without you, that’s for sure. More to come. Thank you, everyone.

STEVE HOUSE:
Hey, real quick before you go — everything we publish, the articles, new podcast episodes, and the live webinars, are now first in our newsletter. You can elect to receive between one and three newsletters a month, written by myself and the coaching team. If you want them, sign up at uphillathlete.com. Thanks for listening, and we’ll see you in the next one.