How the coach makes decisions

Why planning works from the season down to the day

Because a daily workout needs to fit into the week as a whole. The coach first establishes what you're working toward, then decides what to do tomorrow.

There are four planning levels, each building on the previous one.

• Season. Your main goal and races: what you're preparing for, the target date, and how much time you have.
• Month. The focus of the training block: building volume, developing speed, tapering, or recovery.
• Week. A day-by-day schedule that accounts for your availability, constraints, and completed workouts.
• Day. The specific workout: type, volume, pace or zone, and its purpose.

If a higher-level plan hasn't been generated, the service completes it before producing the daily plan: without that foundation, a workout is just a random choice. The daily plan is prepared overnight so it's ready for morning delivery.

What data the coach takes into account

All the data from your watch and dashboard, plus your messages. Numbers don't override how you feel: if you say "I'm struggling", that carries more weight than a readiness score.

Workouts come from Garmin, Polar, COROS, Strava and Apple Watch, with zone and split breakdowns wherever the watch provides them. Sleep, resting heart rate and heart rate variability come from the watches that send them; the readiness score is taken from Garmin only. Training load — CTL, ATL and TSB — is calculated by the service itself from all workouts on a single scale where “100 = one hour at threshold”, rather than taken ready-made from the watch. From the dashboard — the season goal, available days and minutes, blocked dates, and health constraints that you specified yourself. From the chat — what you told us: fatigue, a trip, a missed session, an inconvenient time.

There are five intensity zones, based on the lactate threshold (LT2). The threshold is taken in order of priority: a confirmed test, a value you entered yourself, the threshold from Garmin or Polar, and if there is nothing else, the service's own estimate, which is then labelled as such.

Below — how individual signals turn into a decision.

How training load increases and when it's reduced

Load grows in steps: several loading weeks are followed by an easier one. There is no rigid “three-to-one” template — the recovery week is set by your condition, and the calendar only serves as a safeguard.

Before each week, the service decides for itself whether a recovery week is needed. The week becomes a recovery week if the training load model shows overload, or if two signs coincide: low Garmin readiness, poor sleep, missed key sessions, accumulated fatigue in the model. A single sign is a reason to cut back the problem area, not the whole week: with poor sleep, intensity is reduced first, and volume too if needed.

Overload in the model means specific numbers:

• ATL/CTL above 1.3 — acute load more than 30% above what you have adapted to; this is the same as TSB/CTL below −30%. The rule applies when CTL is 20 or more and ATL exceeds CTL by at least 10 points: at a low CTL a single long run pushes the ratio above 1.3, and a recovery week would be assigned for nothing.
• TSB below −30 — only when CTL is 30 or more. At a lower CTL the absolute number does not scale: for CTL 15 such a TSB means ATL is three times CTL.
• With CTL below 20 we look at growth instead of the ratio: CTL rose by 5 or more within a week.

Accumulated fatigue as a sign also depends on your fitness level: TSB below −10 with CTL of 30 or more, and TSB below −20% of CTL with a lower CTL. All these thresholds are heuristics of the training load model, not an injury prediction: the acute-to-chronic workload ratio has been challenged as an injury predictor (Impellizzeri et al., 2020). We use it as a signal that “load is growing faster than you are adapting to it”.

The model is sensitive to the threshold: load is inversely proportional to the square of the threshold. A threshold set 5% too low inflates load by about 11%, 7% too low — by about 16%; for heart-rate-based load the effect is stronger. A constant threshold error scales CTL, ATL and TSB equally, so ratios are robust to it, while absolute numbers are not. The dashboard shows where your threshold comes from and how reliable it is, and with an estimated threshold the service offers a field test.

The safeguard comes from the methodologies: no more than three loading weeks in a row. If there are no signs of fatigue, an easier week still comes after the third.

The taper before a race works differently: its start is taken from the methodology table for your distance — for example, about 2–3 weeks before a marathon and 3–4 weeks before an ultra of 100 km or more in the mountains or 120 km or more on flat terrain. Before a trip to the mountains, volume is also reduced in advance. Read more in our blog articles on the first days at altitude and heat adaptation.

What happens when you miss a session, get sick, or take a flight

The plan is rebuilt, not "caught up on." Missed workouts are not piled into the remaining days — that's the fastest way to get injured.

Just explain what happened in the chat.

• You missed a day or a week. The coach revises your upcoming plans based on your current fitness, rather than the plan you couldn't complete.
• Illness. The plan is paused, with a gradual return to training. The coach makes it clear that a doctor should decide whether you're fit to train.
• Flights and time zone changes. Travel days are kept free, and demanding workouts are moved to days after you've adjusted.
• Your schedule changed. Just write "three workouts this week" or "move the long run to Sunday."

Where the AI's limits are and when the coach sends you to a doctor

AI is good at keeping training structured and calculating training load, but it can't see you or assess your health. Pain and illness are outside its scope.

The coach explicitly advises you to see a specialist for persistent pain, injuries, symptoms of illness, or any other medical concern. Medical assessment, treatment, medication, and therapeutic diets are outside its scope.

If a message shows signs of a crisis — thoughts of self-harm or suicide, acute despair — the conversation about training stops. The model does not handle such a conversation: the service replies with a prepared text, says that the AI coach is not an emergency service, and gives the number 112.

When a request goes beyond the training plan — nutrition and weight as a goal in their own right, personal guidance towards an important race, technique review from video, pre-race anxiety — the coach may offer a consultation with a live 1trAIner specialist: a coach, a nutritionist or a sports psychologist, from the areas that are currently open. The consultation is paid, and the request is sent to the specialist only after you press “Send”. If there are signs of an eating disorder, for medical questions, and in a crisis, the service does not offer its specialists.

There are technical limits too: the model can get a number wrong or misunderstand a question. That's why you can always view the full plan in the dashboard and mini app and request changes in everyday language. If an answer seems strange, email support@1trainer.ru: the message timestamp helps us trace what happened.

Where the methodologies come from

From a methodology base organised by discipline. The coach does not get the whole base as context, only the sections that match your discipline, distance, level and current training phase.

The base contains 10 methodology guides:

• road running from 5 km to the marathon and beyond;
• road running for beginners: the first 5, 10 and 15 km;
• the Norwegian system: intensity control;
• flat trail running;
• mountain trail running and mountain ultras;
• flat ultra;
• strength training for trail runners;
• swimming in the pool and open water;
• cycling: gran fondos, stage races, gravel, indoor;
• triathlon: building a week from 3 sports.

7 of them are written by authors; 3 — swimming, cycling and triathlon week building — were prepared by a language model following the authors' guides and went through review. Based on the goal in your profile, the service selects the methodology and its sections: the cycle for your distance, week structure, key sessions, strength training, readiness criteria, revision rules, taper. In the plan, the coach refers to the section it relies on.

Separate topics — heat adaptation, altitude, sleep, nutrition on long runs — are covered in blog articles: they are reading material and are not added to the coach's context. Some rules are built directly into the code: clothing and footwear for the weather, cross-training alongside a running plan, elevation gain and descents.