Tracking

What a workout tracker should tell you, and mostly does not

Open any workout tracker and the feature list is roughly the same. Log your sets. See your history. Watch a line go up. Get a rest timer. Maybe a personal record badge when you beat a number.

All of it is recording. Every one of those features is about capturing what happened, and capturing what happened was solved a decade ago, arguably by the spreadsheet.

The question that made you open the app is different, and almost nothing in it addresses that question.

The question is whether it is working

You have two hundred sessions logged. Your bench has been sitting around the same weight for a while. You want to know one thing: is this working, and if not, why not?

What you get is a chart. The chart shows your estimated one-rep max wandering up and down over four months. You look at it and you still do not know, because the chart is not an answer. It is the raw material of an answer, handed back to you with the interpretation left as an exercise.

Which is fine when the trend is obvious. If the line goes clearly up you did not need help. The trouble is that the chart is least useful exactly when your situation is least clear, which is when you actually opened the app.

Why nobody says it

Not because it is technically hard. Comparing two sessions is not a research problem.

It is because saying "you did not progress" requires committing to what progress means, in advance, in a way you can be held to. You would have to decide whether eight reps with two in reserve beats eight to failure. Whether your first set or your best set is the one that counts. Whether a comeback session after a holiday counts as a stall. Whether switching from barbell to dumbbell continues a line or starts a new one.

Every one of those is a real decision with a defensible answer and a cost to getting it wrong. A chart has none of that exposure. A chart cannot be wrong, because it does not claim anything. It is the safest possible thing to ship and among the least useful things to receive.

So the category settled on recording, and the interpretation stayed with the user, who is the person least equipped to do it and the reason the app was installed.

The other failure mode

The newer approach is to hand the interpretation to a language model. Ask why your bench stalled and get a fluent paragraph about sleep, protein, deloads and progressive overload.

The paragraph will be well written and plausible. What you cannot do is check it. You do not know whether it came from a rule the app applies consistently, from something in your actual log, or from the general shape of fitness advice on the internet. And that matters enormously, because you cannot tell a good answer from a bad one without knowing where it came from.

An unverifiable answer is indistinguishable from a wrong one. That is not a complaint about language models, which are useful for a great many things. It is a specific objection to using generation where you needed a decision you could audit.

What the middle looks like

Between a chart that says nothing and a model that says anything is a narrower thing: fixed rules, applied consistently, that you can read.

That means deciding in advance what progress is, saying so, and applying it the same way every time. It means a verdict on the specific exercise you asked about, delivered on the day. It means the app being willing to say "you regressed and I do not know why", which is a real answer and a much better one than a plausible guess.

And it means the reasoning is inspectable. If a verdict looks wrong to you, you should be able to find the rule that produced it and disagree with the rule specifically, rather than being left to argue with a black box that has no visible parts.

That last property is the one worth insisting on. A rule you can read is a rule you can find wrong. A number you cannot trace is one you have to take on faith, and faith is a strange thing to want from a piece of measurement equipment.

The two questions to ask

When you are looking at a tracker, most of the feature list is noise. Two questions separate them.

Will it tell me something I did not already know? If everything it produces is a rearrangement of what you typed in, it is a logbook. Logbooks are fine and you should not pay much for one.

Can I find out why it said that? If the answer arrives with no visible reasoning, you have no way to distinguish a good call from a confident wrong one, and over months of training that difference compounds into real wasted work.

Everything else is a rest timer.

Common questions

What should a good workout tracker do?

Answer the question you actually have, which is whether what you are doing is working. Recording sets is the easy part and every app does it; producing a defensible verdict on a specific exercise is the part almost none of them attempt.

Are graphs in workout apps useful?

Only up to a point. A line chart shows you a number over time and leaves the interpretation entirely to you, which is fine when the trend is obvious and useless when it is not, which is exactly when you needed help.

Why do fitness apps not just say whether you are progressing?

Because saying so requires committing to a definition of progress, and a definition can be wrong in public. A chart cannot be wrong, which makes it the safer thing to ship and the less useful thing to receive.

What is wrong with AI features in workout apps?

Nothing inherently. The problem is when a model generates coaching advice freely, because you cannot check where an answer came from, and an unverifiable answer is indistinguishable from a wrong one.

What should I look for when choosing a tracker?

Ask whether it will tell you something you did not already know, and whether you can find out why it said it. Everything else is logging, which is a solved problem.