There is a Live Slow Ride Fast podcast dedicated to training with Join and how it was developed. Jim explained that he started with programming a couple of rules based on how he trained elite athletes. This list then became longer and longer. I can imagine that a machine learning component is combined with these rules, rather than using machine learning to develop a program from scratch.
My endurance workouts are normally in the range of 60%-80% FTP and keep me nice and ready for the harder workout on Friday.
I can imagine that if you just switch to Join for 1 workout, it can’t exactly pick up the difference between you being able to do a recovery or endurance ride.
Is it in English? Or Dutch?
ETA: it’s in Dutch
Yup, podcast is in Dutch
Looking forward to Friday the workout is a 2:30 one, 105min at Z2 endurance, then 1x30min at Threshold.
Definitely different. Very intriguing.
That sounds like training for real world performance rather than trying to maximize a FTP test result.
Absolutely, and aligns with stuff I’ve seen from really excellent coaches. It’s not a normal structure for the trainer though so it’s different - to me - in that respect .
I’m looking forward to you doing that as well, cos if you’re doing it I don’t have to.
I’m happy about it too
Fridays my day off
I meant the other type of “looking forward” i.e. future gazing
Can I interest you in some vo2max pyramids instead?
Described the calculation in post #30 above, it is a combination of CTL and FTP/kg. For a level of 37 different combinations are possible, but would expect it to be around CTL of 100 and ~4W/kg. Close?
Yes, very close. Could be higher on FTP or lower on CTL (obviously either of those would have the same result).
Woah. My brain hurts, but that’s super clever and also - very relevant to a lot of the discussions recently on this forum and EC podcast etc.
Join must do some analysis of the ride files you upload. I don’t know exactly what, but twice now it has removed the planned VO2max session the day after my cross race. I had those in as planned rides with RPE 9 and race duration, and they came in with pretty much the predicted time and I rated them 9 too. Both times, the next workout changed to endurance, but only after I uploaded the file.
To be honest, I’m only testing Join to see if the promised FTP increase will actually materialise. I’ve never had an FTP increase from training, so really interested to see if it’ll happen. So far, it has predicted a 2W eFTP increase, which is within the error of my power meter, but it is early days.
It looks at time in (power) zones and adjusts future workouts to get the desired distribution for the plan.
Nicely done! My current CTL of ~60, W/kg of 3.2 and Join score of 19.3 fits perfectly in the function.
I was a bit puzzled what the Normal distribution had to do with this, but I see now that you were only looking for something with that shape and could just as well been something like arctan instead, no stochasticity involved in the Join score.
How did you plug in the CTLs into Join, did you just manually input one ride with a huge TSS to get the right CTL?
One of the LSRF podcasts was about fatigue resistance and Jim mentioned how that was a much better indicator for performance than the regular powerprofile that takes the maximum power ever for each interval.
He discussed how you can construct a powerprofile where each interval is measured after having done a specific amount of work and that is a much better indicator. But he also mentioned that they had not programmed that yet into Join. So eventually that could be a much better input to determine a score than the FTP/kg which effectively is only 1 point on the regular powerprofile.
I’ve been playing around with the settings myself. I’ve setup available hours as 10 hours a week as …
Mon - off
Tue - 1hr
Wed - 1hr30
Thu - 1hr30
Fri - off
Sat -2hr
Sun - 4hr
But it doesn’t seem to plan anything above 9hr and 5min for a week.
What weekly hours are you guys using?