We Cut Almost Everything From Menteora. One Action Is Left

Here’s what Menteora did in August. A five-question check-in in the morning. Another one in the evening, about the day and the workout. The answers produced a score, the score produced a personal baseline, the baseline produced a forecast for tomorrow. Plus a weekly report, an AI breakdown, streaks, correlations.

All of it worked and was carefully calculated. And all of it is now switched off.

Why

The main screen was tomorrow’s forecast. It has a problem you don’t fix by improving the math: there’s nothing the user can do with it. You get a number about a day that doesn’t exist yet. There’s no use for it today, and tomorrow morning a new one arrives and nobody remembers yesterday’s.

The second problem was worse. I checked how often the bot actually said “ease off today” on days when the person was planning to train. It came out to 1.6% of days. So the feature the whole thing was built around fired roughly once every two months. The rest of the time the bot was a daily questionnaire with no consequences.

The scale was too coarse as well. Three levels per question means almost every day lands in the middle.

What did hold up

Before breaking it, I tested the product’s core assumption: do our answers about stress describe something that isn’t already in the watch data?

I took 545 days where both sets of data exist. There’s almost no relationship between our stress level and Garmin’s metrics. Which means we really are measuring a separate thing, not duplicating what’s already sitting in the watch app — the idea the product started from.

The original idea was right. The implementation wasn’t.

What it does now

Menteora now has one action: a check before your workout.

The bot messages you on the days and at the hour you set during onboarding. Three questions, a 1-to-7 scale: tension, pressure from circumstances, muscle condition. Twenty seconds.

What you get back isn’t a score. It’s where today sits among your own previous check-ins. Three possible answers:

calmer than your usual;
an ordinary day;
today is in the top 20% of your hardest days — bring the intensity down.

That’s it. No forecast, no reports, no evening questionnaire.

A personal scale instead of a shared threshold

This is the main difference from the old version, and from everything else I’ve seen.

A shared threshold doesn’t work, because people answer differently. A calm person lives in a 3–12 range, and a 17 for them is an emergency. An anxious person lives in a 12–25 range, and a 17 for them is a Tuesday.

Menteora calculates the threshold from your own answer history. For the calm person it lands around 11, for the anxious one around 19. The same total gives them different answers, and both answers are correct.

That’s why the first ten check-ins run on a generic scale, and the bot says so plainly on screen. From the tenth check-in the scale becomes yours. The trial is twenty check-ins, so you spend the second half seeing how the tool behaves once it’s calibrated to you.

Why we don’t show you the score

Keeping the number hidden is a deliberate decision.

The moment someone sees a number, they start comparing: to themselves yesterday, to themselves last week, to other people. Three questions can’t carry that kind of comparison — a one-point difference means nothing. Displaying it would be a lie told with a straight face.

We hand over the part we can stand behind: whether today is above your normal or not.

Muscles are counted separately

The latest rebuild, done this week.

All three answers used to add up into a single total. Which meant legs trashed by yesterday’s session produced the same signal as a stressful week. Those are different things and they call for different decisions. Stress is a reason to pull intensity across the board. Sore muscles are a reason to leave yesterday’s target alone today and train everything else as planned.

The verdict is now two-layered: background load separately, muscles separately. The answer shows you which of the two is the limit today and why. A 7 on muscles is no longer read as stress.

What’s in progress

As of September 2026, one thing: Garmin data. Right now it syncs on a schedule and doesn’t reach everyone who connected a watch. We’re rebuilding it so the bot pulls fresh data at the moment of the check-in and shows it in the reply: Body Battery, training readiness, stress — each number stamped with the time the watch recorded it.

The timestamp matters. Body Battery 78 at six in the morning and Body Battery 78 at noon describe two different people, and without the time the number tells you nothing.

After that the product answers one question completely: what the watch says, what you say, and what intensity to take today.

What I’m after

I want people to open the bot before training, not because a reminder arrived. Twenty seconds, three questions, an answer they understand — and they go train differently than they would have.

If the check doesn’t change the decision about the workout, the product isn’t needed. Everything else is already gone.

Menteora is a Telegram bot: @menteora_bot, and it speaks English. More on how it works on the product page.

It’s a tool for planning training load, not medical advice: pain, injury or illness is a doctor’s call, not a bot’s.

If you try it, write to me at vesninrunner@gmail.com and tell me what your first ten check-ins looked like.