Large Universe Models in sports analytics

What it ingests

Wearable and GPS load data, training session output, sleep and recovery metrics, medical and physiotherapy notes, match event data, opposition scouting feeds, and transfer and contract news.

The belief that matters

Readiness and injury risk per athlete, revised on every training session rather than assessed before each fixture.

Load is cumulative and individual

Injury risk depends on accumulated load relative to an athlete's own baseline, which drifts with fitness and age. This is a maintained-state problem: yesterday's belief is the correct prior for today's session.

A model retrained periodically loses exactly the continuity that makes acute-to-chronic load meaningful.

Multi-source reconciliation

Wearables, subjective wellness reports and physiotherapy notes disagree routinely. Weighting them by demonstrated reliability for that individual — some athletes systematically under-report — is standard evidence weighting.

Limits

Injury prediction has a poor track record and small samples per athlete. The defensible output is risk that has moved and why, informing a medical decision rather than replacing one.