The signal others will see tomorrow,
you have it now
A statistical ensemble audited by adversarial AI. No black boxes: every number is reproducible and every prediction is challenged.
Real data
124K+ historical matches (scores, xG, corners, cards) + live market odds.
HPIE ensemble
Six models combined with weights learned by the auto-calibration loop.
NEMESIS audit
Every prediction goes through an adversarial critic that exposes risks and recalibrates confidence.
Auditable metrics
Brier, calibration, CLV and drift — a transparent track record, not promises.
Six voices, one consensus
Poisson
Models each team's goals from attack/defense strength.
Dixon-Coles
Corrects low scorelines + applies time decay.
Dynamic Elo
Rating updated by result and home advantage.
Bayesian
Gamma-Poisson with a conjugate prior.
Market consensus
De-vigs odds → fair implied probability.
XGBoost
900 trees trained on 124K real historical matches.
ATHENA proposes
Builds a complete mathematical representation of the match and derives ~73 markets with their probability and expected value. It doesn't bet: it calculates.
AnalyzingNEMESIS audits
An adversarial critic reviews every selection, exposes overconfidence and risk, and recalibrates confidence before showing it to you.
AuditingMetrics, not promises
Explore the engine's real track record and how it recalibrates with every result.