MATHEMATICAL MODEL REGISTRY

Tournament Model Variants

Append-only model registry. All parameters, backtest metrics, and calibration scores are immutable and publicly audited.

RESEARCH STATUS:Model AH-dixoncoles-v1.0.0 is NOT VALIDATED. Historical backtest ROI: -2.30%. CLV not statistically significant (p=0.555). All outputs are published strictly for quantitative research transparency, not as betting recommendations.

Scope: AH

CURRENTLY LIVE AUDITED_RESEARCH_BASELINE

AH-dixoncoles-v1.0.0

Dixon-Coles bivariate goal matrix, time-weighted decay, rho per fold

Hypothesis

Baseline champion from EPIC 56 calibration tournament

BACKTEST ROI-2.30%

CLV MEAN-0.0311

CLV P-VALp=0.555

Frozen Parameters (Immutable){"rho":-0.05,"shrinkage":0,"decay_xi":0.0019,"max_goals":6}

Tested: 7,225 betsView Model Audit

Scope: AH

RESEARCH_ONLY

AH-dixoncoles-shrink10-v1.0.1

Dixon-Coles + 10% shrinkage toward market

Hypothesis

Reduce overconfidence via 10% market blend

BACKTEST ROI-2.00%

CLV MEAN+0.0086

CLV P-VALp=0.920

Frozen Parameters (Immutable){"rho":-0.05,"shrinkage":0.1,"decay_xi":0.0019,"max_goals":6}

Tested: 5,831 betsView Model Audit

Scope: AH

RESEARCH_ONLY

AH-dixoncoles-shrink20-v1.0.2

Dixon-Coles + 20% shrinkage toward market

Hypothesis

Reduce overconfidence via 20% market blend

BACKTEST ROI-1.90%

CLV MEAN+0.0431

CLV P-VALp=0.630

Frozen Parameters (Immutable){"rho":-0.05,"shrinkage":0.2,"decay_xi":0.0019,"max_goals":6}

Tested: 5,805 betsView Model Audit

Scope: AH

RESEARCH_ONLY

AH-dixoncoles-shrink30-v1.0.3

Dixon-Coles + 30% shrinkage toward market

Hypothesis

Reduce overconfidence via 30% market blend

BACKTEST ROI-2.23%

CLV MEAN+0.0705

CLV P-VALp=0.430

Frozen Parameters (Immutable){"rho":-0.05,"shrinkage":0.3,"decay_xi":0.0019,"max_goals":6}

Tested: 5,778 betsView Model Audit