Release notes (last 2 weeks):
Enhanced Bulk File Upload Feature - improved handling of bulk upload zip files that contain directories for onboarding users from other platforms.Improved handling of “empty values” in recovery profile charts (HR, HRV, Sleep, etc).Significant updates and improvements to our metric calculations to enhance overall system accuracy.Introduced Swimming Pace Profile charts to enhance data visualization for swimming activities.Optimized training plan and calendar interfaces to load on session-by-session basis, improving user experience by reducing load times and preventing disruptions from individual session timeouts.Implemented a tiered pricing model for coaches to provide more flexible and equitable payment options.
.
Maybe I’m weak, but how is anyone going to tally up 147 TSS in a 75 minute workout? I’ve been loosely using the program for the past few months, and I plan to 100% rely on it going forward to peak for some late season racing, but opening up my account today to see some of these sessions kind of turns me off a bit. How is that even feasible? And the program seems to love 30†vO2 max intervals - is that just the way it works?
The program absolutely loves 30/30s but if you use the workout wizard there’s usually a 45/15 option to switch it up a bit.
I was so happy when the Workout Wizard launched so I could get away from 30/30s on run workouts.
Maybe I’m weak, but how is anyone going to tally up 147 TSS in a 75 minute workout? I’ve been loosely using the program for the past few months, and I plan to 100% rely on it going forward to peak for some late season racing, but opening up my account today to see some of these sessions kind of turns me off a bit. How is that even feasible? And the program seems to love 30†vO2 max intervals - is that just the way it works?
Thanks for posting cloy. This has been a good exercise for the team to look into.
First - you are far from weak and that is evident from all three of your power and pace profiles, swim, bike and run.
Second - you mention loosely using the program for the past few months. This is likely key to the issue you’ve spotted for us. There are quite a few missed sessions and your Performance Potential (i.e., PMC chart) reveals relatively (for you) low fitness. The key aim of the Athletica algorithm is to raise your fitness and maximize freshness before your key event. It’s always working to do that no matter what you do or don’t do. What seems to have happened in your case is that Athletica is trying to raise your fitness by applying its loading methods (the 147 you mention), and that has clashed with a VO2 HIIT session that you changed (via WW) to a 40/20 session, thus creating an unrealistic session. This outlier situation has been noted and we’ll be working hard to refine the logic going forward.
Fortunately, you had the wisdom to recognize this and the ability to listen to your body, and you changed the session to a 30/30 short interval set that was perfectly suitable to your ability (about half the reps and load), as evidenced by your Workout Reserve approaching zero after the last set. You hit the target and walked away unscathed, and progressed your fitness.
Again, appreciate the feedback that was spot on and we’ll work to improve this for the future.
This is good information, Paul. I’ve noticed that when the system shows a bit of craziness like this due to missed workouts, illness, etc, just do one of the normal workouts and it settles down after a day or two.
This is good information, Paul. I’ve noticed that when the system shows a bit of craziness like this due to missed workouts, illness, etc, just do one of the normal workouts and it settles down after a day or two.
Thanks M----n. When I shared my response to Cloy to our backend developers I included your reply. The ‘craziness’ comment post missing values prompted what could be a promising solution to this… let’s see.
Keep the feedback coming all…
This week’s releases:
Limiting Maximal Mean Power (MMP) Insight
We’ve added MMP limiting to session chart analysis, highlighting its relationship to Workout Reserve (WR).Full discussion with examples here.
Enhanced Annual Training Plan Functionality
Ability to add any number of ‘A’ priority races during the year.‘A’ race no longer needs to be the last race, i.e. another (B or C) race can be scheduled after an ‘A’ race.Particularly helpful for coaches building full year annual plans for athletes (no limits).
The program absolutely loves 30/30s but if you use the workout wizard there’s usually a 45/15 option to switch it up a bit.
The program does love 30/30s (and other short interval options via the Workout Wizard) for good reason. The studies from Nicki Almquist and Bent Ronnestad (blog here) tend to show that short intervals (SI) are superior to long intervals (LI) in both well trained and elite cyclists. There will be individual differences of course, but on average there is an indication that the SI option tends to make you better faster, which is why SIs appear routinely in the training diet. Key objectives at Athletica is to help you train smarter using evidenced-informed prescription, and this is one example.
Hi Paul - I’ve tried Athletica AI in the past and have been following along as you continue to develop the system. I have a use case which I’m interested to get your take on if you wouldn’t mind.
I currently train with a triathlon squad but only do some of the squad sessions as most are early AM and I don’t have to do my training so early so I prioritise a bit more sleep 
I typically do 3 sessions with the squad (Wed run intervals and a swim on Thurs and Fri) and sometimes we go for a group ride on Sat. I figure out the rest of my training myself.
I quite like the idea of Athletica AI being able to programme the rest of my training alongside the squad sessions. Training with the squad works for me in lots of ways … and the sessions/programme gets the squad fit in a general sense … but its not individualised in terms of session design nor annual race plan … so that leaves quite a bit of knowledge/planning required to build successfully for a season/races.
I appreciate it probably wasn’t designed with this use case in mind. But curious to know how you think it would handle it?
Thanks
Hi Paul - I’ve tried Athletica AI in the past and have been following along as you continue to develop the system. I have a use case which I’m interested to get your take on if you wouldn’t mind.
I currently train with a triathlon squad but only do some of the squad sessions as most are early AM and I don’t have to do my training so early so I prioritise a bit more sleep I typically do 3 sessions with the squad (Wed run intervals and a swim on Thurs and Fri) and sometimes we go for a group ride on Sat. I figure out the rest of my training myself.
I quite like the idea of Athletica AI being able to programme the rest of my training alongside the squad sessions. Training with the squad works for me in lots of ways … and the sessions/programme gets the squad fit in a general sense … but its not individualised in terms of session design nor annual race plan … so that leaves quite a bit of knowledge/planning required to build successfully for a season/races.
I appreciate it probably wasn’t designed with this use case in mind. But curious to know how you think it would handle it?
Thanks
Hi Chowders,
Thanks for following along. Yours is the perfect use case for Athletica. All the squad and group sessions are 100% beneficial for the social and group dynamic factors. But to your point, schedules don’t usually align to do every single session. Athletica was definitely designed for your use case as each session is moveable in the plan. So to your situation (3 swim squad sessions, Wed run, and group ride Sat) I would do the following.
align the key Athletica swim sets with your squad days. Don’t worry too much about the content match, just follow the squad session per your coach on deck and record what you did on your watch so Athletica takes notice for subsequent program loading.move the key HIIT run session of the week in Athletica to Wednesday to cover your run interval group workout. Like the swim, don’t worry too much about the content, follow your coach, but record the whole session.group ride on Saturday will typically match with long ride in Athletica. Just ride with your group and record the workout.
The remaining sessions in Athletica can now be tweaked and moved to your schedule with the loading adjusted according to the algorithm (see your Performance Profile chart).
Edit: Our user-time constraints feature (available for beta users) might also be used for this use case by placing the known days and durations as constraints so that those days are always populated according to above.
Hope that makes sense and shout out with any other queries.
Hi Chowders,
Thanks for following along. Yours is the perfect use case for Athletica. All the squad and group sessions are 100% beneficial for the social and group dynamic factors. But to your point, schedules don’t usually align to do every single session. Athletica was definitely designed for your use case as each session is moveable in the plan. So to your situation (3 swim squad sessions, Wed run, and group ride Sat) I would do the following.
align the key Athletica swim sets with your squad days. Don’t worry too much about the content match, just follow the squad session per your coach on deck and record what you did on your watch so Athletica takes notice for subsequent program loading.move the key HIIT run session of the week in Athletica to Wednesday to cover your run interval group workout. Like the swim, don’t worry too much about the content, follow your coach, but record the whole session.group ride on Saturday will typically match with long ride in Athletica. Just ride with your group and record the workout.
The remaining sessions in Athletica can now be tweaked and moved to your schedule with the loading adjusted according to the algorithm (see your Performance Profile chart).
Edit: Our user-time constraints feature (available for beta users) might also be used for this use case by placing the known days and durations as constraints so that those days are always populated according to above.
Hope that makes sense and shout out with any other queries.
Hi Paul - been thinking about this some more and others questions that occur:
Does the system use an overarching CTL/ATL/Form model? If so, what role (if any) does that model play in scheduling (e.g. are there ‘guard rails’ around ramp rate for each modality, or other)? Or does it just report CTL etc after the fact for info and instead base scheduling on e.g. individual session feedback and/or other factors?
If such a model is used, is the calculation of load for a session anchored on FTP/CP for each modality? Context for this question is that over time I’ve drifted away from using TSS and the PMC in training peaks because it seems to reward a lot of TSS for long sessions and rather underplay the load of shorter/hard sessions (and in general swimming TSS is a mess!). I wouldn’t want a system to adjust/delay my planned intervals session in 2 days because the sun was out and I did 5 rather than 4hrs endurance ride today!
However load is calculated, assume that for bike it uses power as key input (and failing that HR)? In absence of sensors (let’s say I do my endurance ride on my gravel bike which doesn’t have a PM and I forget my HRM) how is load calculated?
For run can it use power also?
How is load calculated for swimming? This seems a tricky area to me (!): my CSS pace is 1:38/100m just now but I do a lot of squad swimming with pull/paddles zooming along at about 1:30/100m (with my HR at high Z3 / low Z4) … hard for a system to interpret that, unless I tell it I’m using toys? HR (I use FORM goggles with a HRM) maybe helpful here for longer sets but of course isn’t going to pick up load for 25m sprints very well.
I’m not keen on testing - it stresses me out!Is this going to be a problem?
Does programming/periodisation take account of differing distances for ‘A races? For example, I have 2x ‘A’ races this year - one is a 70.3 and the other is a standard distance tri. Broad approach for the 70.3 might be an emphasis on highest intensity work further away from race day (maybe a VO2max focus block, into a threshold focus block, into a 70.3 race-pace focus block); whereas for the standard distance race I might reverse that order and/or take account of e.g. how much VO2max style work I’ve done earlier in the year (would another block be helpful or not vs more threshold or sweet spot style work).
Is there a standard tapering approach for A/B/C races? Is there scope to individualise this? Is this something the system can learn over time? How to travel plans figure into the algorithm for tapering?
How is the load of weight training accounted for? I am in my fifties and have concluded that a decent amount of weight training is essential. I have also realised that it’s tricky to programme it in a triathlon plan without something else having to give!
Does the system give any consideration to whether a single sport block might be the best programming solution for an athlete?
Thanks in advance for considering these points and apologies if any have already been asked/answered elsewhere.
Thanks, Chowders. We appreciate the thoughtful questions you’ve posed. There’s a fair few questions in there so instead of making a long winded reply that becomes hard for anyone else reading to digest, we’ll address your questions point-by-point over the coming days so please be patient as we consider each one before rolling them out. Please note that responses are made in collaboration with our backend modeller and data scientist, Dr. Andrea Zignoli.
Does the system use an overarching CTL/ATL/Form model? If so, what role (if any) does that model play in scheduling (e.g. are there ‘guard rails’ around ramp rate for each modality, or other)? Or does it just report CTL etc after the fact for info and instead base scheduling on e.g. individual session feedback and/or other factors?
Yes, it does. Specifically, we use a modified version of Banister’s 1975 TRIMP model. As you mentioned, there are guardrails around ramp rate, fatigue-to-fitness ratios, and other parameters, including some load response modifiers. This model plays an integral role in scheduling by ensuring that training load increases are gradual and sustainable, preventing overtraining and injury.
Additionally, we report these metrics in your Performance Profile chart, offering a comprehensive view of your fitness levels over time. The Performance Potential chart reflects the chronic increase in fitness due to progressively increasing training loads. Example of successful run to Kona for a female athlete here.
The program absolutely loves 30/30s but if you use the workout wizard there’s usually a 45/15 option to switch it up a bit.
I’ve paid for but effectively have stopped using the program because of how repetitive it is.
If such a model is used, is the calculation of load for a session anchored on FTP/CP for each modality? Context for this question is that over time I’ve drifted away from using TSS and the PMC in training peaks because it seems to reward a lot of TSS for long sessions and rather underplay the load of shorter/hard sessions (and in general swimming TSS is a mess!). I wouldn’t want a system to adjust/delay my planned intervals session in 2 days because the sun was out and I did 5 rather than 4hrs endurance ride today!
Yes, our calculation of training load is anchored on your mode-specific critical power or speed, which is displayed on your Power and Pace Profiles (example below). We employ a model similar to normalized power, where we raise the power values to the fourth power, average these values, and then take the fourth root to return to Watts (MP4). This method enhances the influence of high power outputs while minimizing the impact of lower outputs in the average calculation. It’s more of a mathematical manipulation to better capture the intensity of sessions with high power bursts over short durations, correlating more effectively with session RPE (Rating of Perceived Exertion) compared to simple average power.
We acknowledge your concern about translating load increases directly into performance gains, as the same training load can be achieved with various combinations of intensity and duration. Indeed, improved performance can still be achieved with a maintenance of training load, and might be arising from better efficiency and neuromuscular adaptation over time. Monitoring real changes in power output or pace across your profile (i.e., graph below) allows us to track these adaptations more accurately. Nevertheless, and in my experience with elites, when comparing athletes with different training load regimens, those with higher loads (and theoretical fitness development) often see a translation into real performance outcomes.
To your point about the intervals, Athletica also puts a premium on given “key†sessions during the week, and the logic tries to get you ready for those. Those are the sessions where focus is required and typically the ones that include high-intensity bouts. To be clear, the Athletica logic wants you to be in your best position to tackle those “key†sessions, and it is unlikely that it will “adjust/delay planned interval session in 2 days because the sun was out and did 5 rather than 4 hrs endurance rideâ€.
Regarding our system’s adjustments based on the model, we aim to provide the best-educated guess. Athletica is highly user-adjustable to cater to your personal feelings and preferences. You have the flexibility to modify, move, or delete sessions and switch up training routines using the Workout Wizard. Athletica serves as a valuable tool, but you remain in control, similar to any successful coach-athlete relationship.
It’s also important to note that if you impose training duration constraints on your plan in settings, the Performance Potential chart might not appear optimal to you (i.e., not trending upward). Thread here if interested. This highlights a limitation of the Banister model, which has its duration bias you mention. Once again, this doesn’t necessarily indicate that your performance potential (fitness) is suboptimal.
The workout wizard is actually pretty good, and the system adapts to whatever you decide to do anyway. I usually follow it pretty closely because it’s given me good results, but I’ll be certainly be leaning to the alternative workouts in the wizard a bit more in the next cycle.
My challenge would be that this shows the inherent flaw of the program itself.
If the program essentially gives you the same workout every week, 30/30, and never deviates, and you have to go replace it, then at some point, you’re the actual coach making the decisions.
I hadn’t looked at Athletica in several weeks, but I just checked, and it’s 30/30 this week and the same for next week. Even then, the workout wizard only offers one workout above 40 seconds, and that’s jumping up to 5-minute intervals. (I can choose from 10/30/40-second sprints or a 5-minute interval). That selection hasn’t changed in the 5 months I’ve done the program.
As was mentioned, there’s an over-reliance on 30/30’s that tend to go against the guidance of Matt Fitzgerald, who is part of their team: https://www.trainingpeaks.com/blog/what-is-the-best-workout-for-increasing-vo2-max/
Science is great, but it has certain limitations. The conditions that scientists are required to create in order to make head-to-head comparisons between workouts are artificially simple compared to the real world. If one workout increases VO2max more than another in a study lasting three weeks, who’s to say the opposite wouldn’t be true over nine weeks? And who’s to say that the greater increase in VO2max that one workout confers compared to another will translate to a bigger improvement in real-world competitive performance? It’s impossible to properly assess the effectiveness of a given workout type outside of the context in which its benefits are manifested, and that’s exactly what studies like the ones just described lack: context.
The question addressed in this article also assumes that, once the most effective workout for increasing VO2max has been identified, athletes should do that workout exclusively going forward. But this would be inadvisable. Whether your goal is to increase your VO2max, your lactate threshold, your endurance, or any other facet of overall fitness, it’s best to use a variety of workout formats to get the job done.
So all that said, it gave me a good base structure for a workout week, but after a month or so, you could more or less do it yourself and be just fine.
However load is calculated, assume that for bike it uses power as key input (and failing that HR)? In absence of sensors (let’s say I do my endurance ride on my gravel bike which doesn’t have a PM and I forget my HRM) how is load calculated?
For run can it use power also?
How is load calculated for swimming? This seems a tricky area to me (!): my CSS pace is 1:38/100m just now but I do a lot of squad swimming with pull/paddles zooming along at about 1:30/100m (with my HR at high Z3 / low Z4) … hard for a system to interpret that, unless I tell it I’m using toys? HR (I use FORM goggles with a HRM) maybe helpful here for longer sets but of course isn’t going to pick up load for 25m sprints very well.
Training load for cycling is primarily calculated using power output. If power data isn’t available, heart rate is used as the fallback, and speed is used if neither power nor heart rate data is available. In the case of your gravel bike ride where you might lack both power and heart rate data, speed would be the next best metric we can use to estimate load (note that we are close to a solution for cycling speed conversion to power output).
For running, we prioritize using Normalized Graded Pace (NGP). If NGP isn’t available, we use heart rate data. In situations where your device isn’t functioning and you know the time and distance of your run, you can manually input this data to calculate the load. Running power analysis is on the roadmap but has not yet been implemented.
Swimming load is calculated using pace first and heart rate second. We acknowledge the complexities and limitations here, especially given variations in swim training conditions like using pull/paddles, drills, different strokes, or doing sprints. Although heart rate data from various devices can be helpful, it too is not always perfect, especially for short, intense efforts. Our thinking with swimming for the triathlete in Athletica is to get some level of load indication but as per posts above, let’s not be crazy concerned with precision because our devices aren’t yet to the same level we see with running and cycling modes. But they are getting better for swimming and we will continue to iterate as the tech in this space develops.
In addition to these primary metrics for swimming, cycling, and running, our system includes Foster’s Rating of Perceived Exertion (RPE) under ‘Session Analysis.’ This provides a fallback method to calculate load when other data isn’t available or reliable.
While no system is perfect, maintaining consistent and relatively accurate inputs yield the most reliable outputs, keeping your training prescriptions sound and within the guardrails described above.
Paul - it felt a little rude waiting in silence while you worked through my questions, so just a quick note to say thanks for the responses so far and for the considered approach.
Much appreciated,
Paul
Paul -
Maybe I’m in the minority worrying about something so stupid, but any plans to change the interface? Or having a dedicated app?
I, like many, are used to trainingpeaks. Granted, TP isn’t creating anything magical by any means, but the current Athletica AI interface is a few steps down from some other competitors.
Overall, I’m enjoying it so far!