The coach that knows you

3 September 2026 · Written by the founder

A real coach is the best thing that can happen to a runner. Someone who knows what you did last Tuesday, knows you always go out too fast on the first rep, knows you have been quiet for a week and asks why. Someone beside you saying the one thing that helps, at the moment it helps.

Almost nobody has one. They are expensive, and there are not enough of them.

So build one that knows you. For a fraction of what a real one costs.

That is only possible now, and only because of what you already carry. Every run is recorded in more detail than a coach could ever observe from a bike: every heart beat, every metre climbed, how you slept, how your pace fell apart in the last third, how the same session felt three weeks ago. You have been collecting it for years, and most apps hand it back to you as a number.

That is the gap. The data is already there, it is already yours, and it is sitting unused. We can use it. That is the whole premise of what I have been building this week.

What I went looking for

Not new features. I went through Klivet's own code looking for things it already works out and then throws away. I found a hundred and ninety-six.

Four, so this is not abstract:

• It decides several times a run whether your execution is improving, holding or falling — and only ever uses "falling".
• Your race fuelling plan reaches your watch. Nothing on the watch reads it.
• The wind advice knows you should run your loop backwards so the wind is behind you coming home. Only the sentence about the wind ever reaches the screen.
• There is a five-level progression, Building through Elite, with a progress bar. Crossing a level does nothing.

Then I found the thing underneath all of it. Klivet decides when to speak using thirteen triggers — pace drift, heart rate climbing, a climb, going out too hot, fading on a rep. I read the list properly for the first time in months.

Every one of them fires because something is wrong.

You could run a perfect session — every rep on target, the easy parts genuinely easy, a negative split — and the app would say nothing. Not because it did not notice. Because there was no reason in the system for it to speak when things were going well.

No coach worth having behaves like that.

What broke

Two things I found on the way, and both were live.

The effort score. The watch asks how hard that was, one to five, and writes it to Apple Health. Apple's scale goes to ten. So your hardest ever session was recorded as five out of ten — "moderate" — in a store other apps read. Every hard run in your history was written down as half of what it was.

CUT SHORT. When Klivet decides you should do less today, it shortens the session — then judged you against the original target and stamped CUT SHORT on your run. You did exactly what the coach asked, and the coach called it short.

What shipped

1. Effort is recorded as what it wasFive buttons on the wrist, mapped to Apple's ten. The top one now means maximal.
2. Doing what you were told is not failingA shortened session is judged against the session you were actually asked for.
3. The watch stopped scrollingEvery screen is a snap-to-page card. On a 40 mm watch it used to shrink text to fit and cut sentences in half; now a test fails if a line is too long.
4. The coach stopped repeating itselfThe lines you hear most now have several wordings, and both devices rotate them.
5. It can tell you when you are doing wellA new kind of cue that fires because something went right. Four a run, maximum, and never straight after a correction.
6. Your fuelling plan speaksThe race fuel point that had been reaching your wrist and sitting there now says when to eat, and why there.

That fifth one is the one I care about. It sits at the bottom of the priority ladder, so it can never push a warning out of the way, and the whole run shares one budget. Restraint is the point. A coach who says everything is not a coach, it is a podcast.

What it will sound like

Four from the plan. None of these are shipped — they are what the next months are for, and every one of them is built from something the app already knows about you.

• Race day, approaching 21 km Fuel point — take it now, you’ve got a hill at seven.
• An easy run, flagged as run with company You ran that 40 seconds a kilometre slower than your easy pace. You were talking. It still counts as the run.
• Three days of effort scores against the numbers You called that an 8. The numbers say 5. Third time this week — something is going on that isn’t training.
• Planning a route, twelve weeks out from a race Found you a loop with 180 m of climbing in the last third. Same shape as the finish of your race.

What that was checked against

• Three blocks of work, all merged and pushed
• App, engine, health-layer and watch suites green on each
• A new test that fails if a coaching line will not fit a 40 mm screen
• 26 checks on real hardware — owed, not yet walked

Passing tests are not a good run. I want to finish a hard session, tap five, and read a ten back out of the Health app before I call that one fixed.

Where this leaves it

The remaining findings are written up as sixteen build plans, in the order they have to happen. Each one names the file, the function, and how we will know it worked.

Klivet has been a very well-informed app with a narrow mouth. It knew you had just run your longest in three months. It knew your second half was faster than your first. It knew you had eaten nothing in ninety minutes and had a hill coming. It said none of it, because the only reason it had to speak was that you were doing something wrong.

That is what is being fixed. Not more talking — about a third of the plan is the app choosing to say less, in the right places. Running with a friend should get you a quieter coach, not a louder one.

But depth is the promise. Something that knows what you did, remembers what you said, notices when you are doing well, and says the one thing that helps at the moment it helps.

Next entry goes out with those 26 checks walked. I will tell you what they find.

Klivet reads your training and your recovery signals and gives you a readiness estimate. It is an estimate, not a diagnosis. It does not know that you are ill, and it is not a substitute for a doctor or a coach who can see you.