How the calculation works

For decimal odds d, the raw implied probability is 1/d. For each complete 1X2 market, we divide each of the three implied probabilities by their sum. This proportional method removes the overround mechanically; it does not establish the true chance of winning. We then average the normalised probabilities of the included bookmakers with equal weight.

The current consensus uses only football matches with a future resolved kickoff, a recognised full-match period and at least two bookmakers. Each bookmaker contributes one complete latest snapshot. Captures must be no older than 24 hours, at most 6 hours apart, and before kickoff. These are editorial freshness rules, not measured confidence thresholds. A display-context full-match label is accepted as context, not independent verification of settlement rules.

An explicit capture timestamp takes precedence. The first recorded import time is used only when capture time is absent, and is labelled. Timestamps with an offset keep that offset; unzoned owner captures use UTC+7. Kickoff time is separate. A later import never refreshes an existing capture.

A large favourite probability, small bookmaker disagreement and recent observations are different facts. We show them separately. Similar prices may come from shared data feeds. No extra confidence is awarded merely because prices repeat. We do not select the best price from different bookmakers to create a synthetic probability market.

When the two leading outcomes are within 3 percentage points, the summary says no clear favourite. Otherwise the leading outcome is named; 70% is the display threshold for a strong favourite. These labels describe the estimate, not a betting recommendation.

A scheduled capture stores eligible estimates before kickoff, preserving the method and included observations. Calibration uses the last stored pre-kickoff estimate per match and only results entered with a source. The three-outcome Brier score ranges from 0 to 2. Small samples cannot establish reliability.