Quartz performance statistics

Follow performance in real time

Strategy backtesting

The chart below presents back testing results of the model against bookmakers’ odds, starting from late August 2025 and is updated daily.

This simulation uses actual quartz-based calculations performed on each race day (not calculations performed afterwards). It assumes a bet is placed 10 minutes before race start (configurable), using either the best or worst odds available from bookmakers at that time—but only if the bet meets a minimum threshold for profitability, defined by its expected profit.

Bet profitability is a theoretical metric used to detect when bookmakers offer inflated odds relative to a runner’s true chance of winning.

Example:

    • Suppose a runner has a 20% chance to win, and bookmakers offer odds of 5.5 (decimal format).
    • If the race were repeated infinitely under identical conditions:
      • A £1 bet would win 20% of the time, paying £5.5 → net gain: £4.5
      • It would lose 80% of the time, losing £1
    • So the average return is:

(5.5 – 1) x 0.2 – 1 x 0.8 = 0.9 – 0.8 = 0.1

  • This implies an expected profit of 10% per £1 bet

Model Key Metrics

The two metrics are designed to benchmark the accuracy of Quartz calculations relative to bookmakers’ odds at various times before each race: strike rate and odds accuracy. These statistics are based on actual Quartz outputs and observed bookmaker prices since late August 2025, and are updated daily.

This measures how often Quartz correctly identifies the winner—specifically, how frequently the runner with the highest predicted probability wins, compared to how often the bookmakers’ lowest-priced runner wins.

It provides a direct comparison of predictive power:

  • Quartz strike rate reflects the model’s ability to pick winners.
  • Bookmakers’ strike rate reflects market consensus.

Predicting winners is only part of the challenge—accurately estimating winning probabilities is equally critical.

A well-calibrated model should assign probabilities that match real-world outcomes. For example, runners given a 20% chance should win approximately 20% of the time over a large sample.

This graph compares:

  • Quartz probabilities vs. actual win rates
  • Bookmakers’ implied probabilities (from average odds) vs. actual win rates

The closer the plotted points are to the diagonal line (where model probability = actual probability), the more accurate the model.