HealthTrend started from a problem I kept running into while cutting. I could weigh myself every day and still have no idea what my weight was actually doing. A single reading could jump up or down because of food, water, timing, or normal day-to-day noise, while what I actually cared about was whether my underlying weight was trending down and how quickly.

I wanted something more useful than a basic weight log or a moving average. I wanted to separate the signal from the noise, quantify how certain that estimate was, and make the result easy to understand rather than hiding everything behind a single smoothed number.

So I built HealthTrend. It estimates the underlying weight trajectory from noisy and irregular measurements, then presents the current estimate, rate of change, uncertainty, and forecast in a form that I would actually want to use myself.

Building it also became a way for me to bring together statistical modelling, software engineering, product design, and rigorous evaluation. An important part of the project was testing where the model does and does not work, rather than assuming that a more sophisticated model must automatically be better.

The estimator, its assumptions and its limitations are set out in full on Method, down to the equations and the code that implements them.

Measurements you enter are sent to the analysis service to produce the analysis on screen, and are not stored. Nothing is saved in your browser either, so a reload starts from an empty page. Every series in the demo scenarios is synthetic and labelled as such.