About the model

LoLPredict estimates the probability that a team wins a pro League of Legends series. The current build uses a chronological team Elo baseline trained from a bounded Oracle's Elixir public match-data crawl. Forecasts are generated only from source rows available before each match; no shuffled splits or future results are used.

Data source

Direct Oracle's Elixir CSV downloads were blocked or quota-limited in this run, so the ingest uses the public Datalisk/frontend API exposed by the Oracle's Elixir match-data catalog. The checked-in sample covers recent 2024+ pro series and upcoming fixtures.

Model

Teams carry an Elo rating. The backtest grid selected K=40, temperature=0.9 and annual carryover=0.8 by chronological log loss with Brier as the secondary check. There is no home advantage and no draw outcome for LoL series.

Performance

The performance page reports Brier score, log loss and reliability buckets from the generated walk-forward predictions. These metrics describe the current bounded sample, not a complete all-history production model.

These are model probabilities for entertainment and analysis, not betting advice.