Model Accuracy & Track Record

Every prediction frozen at registration time. Every result shown — including where the market wins.

Pre-match model

Team-signal predictions, frozen at registration time — published uncalibrated and resolved automatically

League	Month	N	Acc %	Brier(lower = better)	Mkt Brier
LES	Jul 2026	10	90.0%	0.1705	0.1277✓
LFL	Jul 2026	15	53.3%	0.2687	—
LPL	Jul 2026	12	58.3%	0.2360	0.1892✓
MSI	Jul 2026	15	53.3%	0.2442	0.1828✓
Prime League 1st Division	Jul 2026	45	60.0%	0.2383✓	0.2667
EMEA Masters	Jun 2026	13	46.2%	0.2679	0.2639✓
MSI	Jun 2026	5	40.0%	0.2266	0.0030✓
Prime League 1st Division	Jun 2026	5	40.0%	0.3119✓	0.3192
Overall totals		137	60.6%	0.2337	0.2085

At-draft AI agent (Hermes)

Predicts at draft lock-in, compared against Polymarket odds captured at the same moment. Rows where the market wins are shown — that transparency is the credibility.

League	N	Acc %	Agent Brier(lower = better)	Mkt Brier
Prime Leag	70	67.1%	0.2136	0.1819✓
LCK	67	71.6%	0.1885✓	0.1922
MSI	65	66.1%	0.2052	0.1978✓
LJL	55	63.6%	0.2272	0.1709✓
North Amer	55	58.2%	0.2571✓	0.2935
LES	51	66.7%	0.1971	0.1759✓
LCP	40	67.5%	0.2019	0.1908✓
LFL	39	71.8%	0.2047	0.1754✓
LEC	37	70.3%	0.2122	0.1872✓
LCS	33	42.4%	0.2723	0.2659✓
VCS	24	54.2%	0.2549✓	0.3700
CBLOL	23	65.2%	0.2358	0.2251✓
LCK CL	19	73.7%	0.2145	0.1730✓
TCL	17	64.7%	0.2190	0.2099✓
EWC	9	44.4%	0.3067	0.2723✓
All leagues	610	64.3%	0.2211	0.2007

✓ ★ agent Brier lower✓ ★ market Brier lower0.25 = coin flip

Methodology

The pre-match model uses team-level signals only: Elo ratings, season win rate, H2H record, and recent form — no champion data, frozen at prediction time. Hermes (the at-draft agent) adds seven signals including champion win rates and duo synergies. Brier score measures probabilistic accuracy: 0 = perfect, 1 = maximally wrong, 0.25 = equivalent to a 50/50 coin flip. Lower is better. Predictions are pre-registered — they cannot be edited after creation.

About the model·Our prediction game's append-only public ledger