Every morning the watch hands me HRV, sleep score, body battery, training readiness. About 40 numbers before breakfast. And not one answer to the only question I have. What should I do today?
If you own a fitness tracker, you know this feeling. The watch says "HRV status: unbalanced". Okay. Is that bad? Should I train today? The app doesn't say.
So I connected Claude to my training data and let it do the translating. Two decisions turned out to be the whole point.
The first one. It never shows me a raw number. No HRV value, no recovery score, no body battery percentage. Just one sentence. "Your heart is working a bit harder than usual, keep it easy today." The numbers still live underneath, in the rules. They just never reach the screen. I don't want a metric. I want a verdict.
The second one. Claude does not make the call. This is the part I care about. The decision, hard or easy or rest, comes from about 250 lines of plain Python rules I wrote myself. Claude only wraps that decision in words. It can paraphrase the verdict. It cannot invent one.
That split matters more to me than any feature. I don't trust an LLM to decide what I should train. I'm happy to let it explain the decision in plain language.
The apps measure, the code decides, Claude explains, I train. Ten seconds, no dashboards.
This isn't about running. It's about any data you collect religiously and have never once understood. The tools already measure you. The real question is who translates it, and who gets to decide.
Would you trust an LLM to make the call, or only to explain it?