Research & Discussion: How do you adjust your training when life gets in the way? (Short survey for self-coached riders)

Training
1

Hi everyone,

Like many of you, I spend a lot of time on Intervals.icu analyzing power curves, decoupling, and training load. It’s an incredible platform for diagnosing what happened in past workouts.

However, as a self-coached rider balancing training with work and family, the hardest part is often the forward-looking decision: what to do when a week gets derailed by overtime, poor sleep, sudden fatigue, or bad weather?

I am conducting research into how self-coached endurance cyclists manage plan disruptions and make compensatory decisions (e.g., when to push through vs. back off, how missed interval sessions are rescheduled, and what recovery metrics matter most).

If you have 3-5 minutes, I would greatly appreciate your input in this anonymous survey:

:backhand_index_pointing_right: Take the 3-5 minutes survey :backhand_index_pointing_left:

Beyond the survey, I would love to hear your personal heuristics:

  • When you miss a key Tuesday/Wednesday intensity session due to work or fatigue, what is your default rule of thumb for the rest of the week?

  • Do you reschedule, shorten the duration, or scrap it completely to protect the weekend ride?

2

Just out of curiosity, could you add a disclosure, if it is an indirect advertisment for yet another AI app? (Sorry if it‘s not the case, but then it also would be nice to know who is doing research) :wink:

Personally I don‘t get the idea of AI coaching - if a key workout is missed, reschedule it. If an easy workout is missed, skip it. If rescheduling is not possible, then this is also not the end of the world.

5 Likes
3

Fair question, and completely understandable skepticism.

To be 100% transparent: no, this is not a stealth promo for “yet another AI coach” app. In fact, I largely share your skepticism - I don’t believe in delegating training decisions to an autonomous AI, primarily because AI fundamentally lacks the necessary context. An algorithm or LLM doesn’t genuinely understand the nuance between systemic work stress, an impending viral infection, poor sleep, or actual physiological overreaching, and usually just serves generic, ungrounded advice.

Your rule of thumb (key workout missed → reschedule, easy workout missed → skip) is solid common sense. The goal of this research isn’t to build a black-box “AI coach” to replace human judgment. It’s to understand where self-coached cyclists actually experience friction - whether that’s the blind spots of classic metrics (like scalar TSS ignoring durability or non-training stress), dealing with multi-day schedule domino effects, or the cognitive overhead of interpreting conflicting data.

I’m a fellow self-coached cyclist and software engineer researching real workflow gaps and decision-making friction, not selling an automated “AI magic button.” Appreciate the direct feedback!

1 Like
4

Just went thru your survey, thanks for the opportunity to share personal experience, Im coming from tech too, been in the field for too long I guess, dont count years anymore lol
And self coached as well, pretty passionate cyclist, love it, love not only riding my bike but work on it too! (not really sure what I like more - upgrading/working on it or riding it).

back to your topic - I build one of these connectors, so my answer is a little biased.

Missed for time then I’ll reschedule if there’s room in the week, and drop it if there isn’t. Missed for fatigue, poor sleep, or a scratchy throat, I don’t reschedule it at all. It’s gone, not owed.

Either way I don’t stack it onto the remaining days. The constraint is the week’s total intensity, not the individual session, and the weekend ride is usually where the adaptation I care about actually lives. Protect that, let the midweek go.

@R2Tom’s version — key missed → reschedule, easy missed → skip — is the same idea

On the AI point, I mostly agree with you, and the disagreement is narrow but I think it matters.

You’re right that a tool which only sees training data can’t tell work stress from an incoming infection. It sees a number that moved and invents a reason. That’s the failure mode, and it’s real.

But that context isn’t missing from the world - it’s missing from that tool. If you’ve been talking to Claude or ChatGPT about your week anyway, it already knows about the deadline, the kid who brought something home from school, and that you mentioned your throat felt off on Sunday. Nobody entered that as structured data. It’s just there, because you said it.

So the honest version isn’t “AI understands your context.” It’s that an assistant you already talk to has heard more of it than any training platform ever will, and the training data is the only half it was missing. That’s memory, not diagnosis, it still doesn’t know you’re getting ill, it knows you said you felt off three days ago. Which is more than a TSS number has, and less than a coach who can look at your face. (my honest take here: dont take TSS as absolute ultimate marker of your load, context matters, at time I can reach half of weekly TSS load just in one sitting, having an intense group ride on a weekend for instance).

The friction you’re researching isn’t really “conflicting data.” It’s that the decision needs information no platform holds, so every tool that only reads the platform is answering a different question than the one you’re asking.

5

When the week blows up I cut the hard days first, not the easy riding. Sleep is off or I’m wrecked, I skip the interval session. I keep the easy volume if the legs are there. The week is not lost if the hard work waits.

1 Like
6

Tbh if you are a scientist or student working on actual research that will be published later, that info needs to go into the form or your post.

If it is just a market study for an app you’re writing which will, like all other such apps, keep the data and algorithms a guarded secret, I’m for sure not interested.

You’re on the forum of the most open tool in the realm of sports… come on, give us something to work with. :wink: just make it a discussion topic here so we all learn something.

4 Likes
7

Fair call, and thanks for being direct.

To be completely transparent: I am neither an academic researcher nor conducting a corporate market survey. I’m a software engineer and cyclist building an algorithmic engine designed to suggest workout adaptations and pacing/training recommendations using telemetry across multiple platforms.

The survey was intended to validate edge cases and common failure modes in how self-coached athletes triage missed or unplanned sessions. However, you’re right about the platform ethos - keeping it to an opaque Google Form defeats the purpose of being here.

Happy to make this an open technical discussion right here. Specifically, I’m tackling:

  1. How to handle conflicting load/strain metrics across platforms without arbitrarily favoring one ecosystem’s model.
  2. Rules vs. probabilistic models for workout adjustments when life disrupts the schedule.
  3. Validating fatigue thresholds against subjective feedback vs. purely power-duration telemetry.

What’s your take on how current adaptive algorithms fail when bridging multi-platform telemetry?

8

The biggest failure I see is treating the metrics as if they’re interchangeable. Garmin readiness, Intervals form, sleep and “I feel wrecked” all answer slightly different questions, and they often update at different times. I’d keep the signals separate, show how fresh they are, and make a smaller change — or ask — when they disagree. A confident new workout is usually the worst response to conflicting data.

9

That’s the exact premise. I’ve lost count of how many times Garmin flashed “Ready for high effort” while my legs felt completely dead - and vice versa. I don’t treat any wellness score as gospel. The goal isn’t to close every gap or build a perfect predictor, but to minimize obvious errors by correlating objective wellness/load data with subjective human feedback.

Every body adapts differently, so the focus has to be on building an individualized context bubble per athlete rather than relying on global assumptions.

10

I also plan my own workouts and I find that this gives me the flexibility to reorganize the week when unexpected things come up. If I miss a key workout, what I try to do is reorganize the week so I don’t miss the key stimulus; sometimes I cut back on a less relevant workout to make room, provided that putting the key workout on that day doesn’t compromise my recovery for the next key workout. Other times, the smartest thing to do is just skip it and stick to the plan. Ultimately, the decision to move it or skip it is multifactorial because it depends on the reason why it was missed (illness, unexpected event, etc.) and the block’s objective

11

I’m actually trying to get this stuff done by AI and it’s been working really well for me. I just load some scientific papers about training schedules. Enter my goals in intervals.icu. Then let AI program my schedule, making sure it stays within the limits. It looks a lot like the premium planning feature of intervals.icu. But the thing I like most about it is that it will fix my schedule to be optimum. I used to train about 10x per week and I would either skip a workout and be completely demotivated afterwards. Or I would very strictly stay on schedule, often burning out because I overtrained and not paying attention to my body. Coaches would often be looking after more then 5 athletes at a time and not know anything about using data to change the schedule.

Right now about every week I’ll have a little chat with AI. Say what I did do, what I did not do, how I am feeling and if there is anything coming up that may put a wrench in my training plans like an all day family event. And it will just go ahead and make an alternative schedule for me, trying to train most optimally towards my goals with all the caveats in mind. It’s amazing. No more feeling guilty about missing trainings. Or feeling annoyed that a party ruins my training schedule etc. etc.

12

Will the algorithmic engine be commercial or open source?

13

I can’t give a definitive yes or no right now, as it depends on several factors. The main issue is that the engine requires more inputs than just a set of FIT files. Extracting it in isolation would result in an incomplete solution.

But I will keep you posted!