Kovura's design principles: Two questions, never blended
Can the body train today?
Readiness composes your systemic recovery score with average muscle readiness into one forward-looking number, weighted by a persona that self-calibrates as you log.
How hard has it actually been worked recently?
Load compares a fast-moving acute average of your training load against a slow-moving chronic average. The same acute:chronic method professional sport uses to estimate injury risk.
Designed for athletes. Powered by data.
Five rules, enforced in code. Not just in copy.
These aren't marketing claims. They're constraints the app can't violate. Hover any card to see it work.
Predicted and measured never merge
A routine's simulated “Predicted Load” and a session's real logged-set load are always two visibly separate numbers. Never quietly averaged into one.
Explainability over accuracy at all costs
The soreness-personalization loop is a small, hand-written correction rule rather than a fitted model, so every adjustment traces back to one check-in you can remember.
AI explains; it never computes
The on-device language model is only ever handed numbers the deterministic engines already produced. Every AI feature has a rules-based fallback that still answers.
A workout is a sandbox until you confirm
Every substitution, extra set, or reorder made mid-workout stays local to that session unless you explicitly accept updating the saved routine.
Confidence is shown, not assumed
A reading built from six signals and one built from zero are never presented as equally certain. A visible confidence tier adjusts what's asked of you next.