Built on the OpenPowerlifting corpus

OpenPowerlifting is one of the quiet triumphs of sports data: a volunteer-maintained archive of millions of competition results across federations, decades and continents, released into the public domain. Every meet result carries the things that matter for prediction — the lifter's age, sex, body weight, equipment class, the federation's testing status, and what actually happened on the platform, attempt by attempt.

For chalk.bar, this corpus is the foundation. Competition results are the rare kind of strength data that is both abundant and trustworthy: performed to a standard, judged by referees, recorded publicly. Gym data is plentiful but soft — self-reported, uncalibrated, collected under wildly varying conditions. Meet data is the opposite: sparse per lifter but hard, and across millions of results the sparsity stops being a problem for learning what populations do.

A prior, not a verdict

We use the corpus to build population priors: what lifters of a given age, sex, body weight and competition history tend to total, how strength tends to move across a career, how much meet-day performance varies. That prior is the starting point for every individual model — deliberately weak, ready to be updated by the lifter's own results. The corpus tells us what is typical; the individual's data tells us where they sit relative to typical, and with how much certainty.

This division of labour matters. A model trained only on population data can never tell you about yourself — it can only tell you about people shaped like you. A model trained only on your own handful of meets drowns in noise. The combination — a population prior updated by individual evidence — is the standard Bayesian answer, and competition data is unusually well suited to it.

Giving back

The corpus is public domain (CC0), which is a gift, and gifts of that kind survive on reciprocity. OpenPowerlifting runs on volunteer effort and contributions. If chalk.bar is useful to you, some of the credit belongs to the people who spent their evenings transcribing federation scoresheets — consider supporting the project.