UFC DFS projections vs ceiling: the average is not the outcome

An average can be a score that never happens

Suppose a fictional fighter scores 110 points in 60% of outcomes and 20 points in the remaining 40%. The mean is 0.60 × 110 + 0.40 × 20 = 74. In that two-outcome model, the fighter never actually scores 74.

The example explains an average, not a real fighter's range. Real forecasts can include many scores, different finish timings and different paths through the bout. The two bars let you see the arithmetic without mistaking a smooth chart for evidence from actual fights.

A 100-point chance answers a different question

In the default example, the chance of at least 100 points is 60%. Lower the high score to 95 and that chance becomes zero, even though the mean can still be relatively high. Change the threshold question and you change what is being measured.

A mean and a threshold probability are useful for different comparisons. The average summarizes every outcome with its probability. The threshold probability counts only outcomes on one side of a chosen line. One cannot be recovered from the other without additional assumptions.

Always ask how ceiling is defined

The word ceiling can refer to a percentile, a high-scoring scenario or an observed historical maximum. Those are different quantities. A 90th percentile is a position in a forecast distribution; a career high is one past observation.

In this tool, the high score is simply a score you type into a fictional two-outcome model. We do not call it a measured 90th percentile. When a product shows a ceiling, look for the definition, the data behind it and the date the estimate was made.

Six individual upper outcomes do not make a lineup ceiling

Lineup points add. Percentiles generally do not. Adding six fighters' 90th-percentile scores does not calculate the 90th percentile of the lineup, because outcomes across fighters can depend on one another.

MMA makes the issue especially clear when two fighters share a bout. A lineup distribution needs coherent joint outcomes. The single-fighter toy model here cannot answer that joint question and is not an optimizer input recommendation.

Evaluate a forecast with the question it asked

A high-probability outcome can fail once without disproving the forecast. A useful review preserves the prediction before the event and compares many such predictions with outcomes. For example, predictions near 60% can be grouped to ask whether the event occurred about six times in ten across a sufficiently large comparable set.

Avoid fitting the explanation only after the result. Write the question first: expected score, chance of a threshold or a particular finish path. Then use the matching observation when reviewing the forecast.

Common questions