Kelix Lines / Public research

How Kelix Lines estimates a game.

How Kelix Lines builds sports probabilities using opponent-adjusted ratings, 10,000 simulations and historical calibration, and how to interpret model-versus-market gaps.

Ratings and adjustments

Opponent-adjusted offense, defense and pace ratings use regularized regression, weight recent games and blend in a preseason prior. Available adjustments include home advantage, travel, time zones, rest and altitude. The scheduled venue determines home advantage. Player adjustments depend on the sport and available data; college injuries are not covered.

10,000 simulations

Basketball is simulated possession by possession and football drive by drive, including league-specific overtime rules. A seeded simulation makes a given matchup, ratings version and input set reproducible. The number of simulations describes computation, not model accuracy.

Calibration and historical evidence

The pipeline evaluates earlier seasons with walk-forward backtests and calibrates raw probabilities. The interactive Lines workspace displays available historical Brier score, log loss, accuracy and spread error alongside market results, including periods when the market does better. These metrics are historical simulations, not live trading returns.

Comparing the same outcome

Market sources include Kalshi, Polymarket, DraftKings and FanDuel when matched data exists. Sportsbook probabilities have the margin removed. The comparison reports a probability difference in percentage points; it is not an executable return. Inspect the underlying contract rules, bid and ask, costs and liquidity before treating two markets as equivalent.

Freshness and limitations

Public forecasts and schedules are cached for five minutes. Quotes show their own collection time; quotes over 30 minutes old are marked stale. Missing sources stay missing. Forecasts stop at the scheduled start, and completed pages do not reconstruct pregame predictions using later ratings. A persistent archive of frozen pregame forecasts is not yet available.

Sources and corrections

Underlying pipelines use CollegeFootballData, CollegeBasketballData, nflverse, balldontlie and The Odds API where configured, plus exchange market data. Availability varies. For a data or matching issue, send the exact game page to rocco@kelix.dev. Kelix is independent of these providers.