Three different statements
| Statement | Meaning |
|---|---|
| “A wins with 70%” | If calibrated, sides like this lose about 3 times in 10. |
| “High confidence” | The inputs behind the number are complete and well-sampled. It is a statement about the evidence, not a promise about the result. |
| “A will win” | A certainty claim. No honest sports model makes it. |
How confidence tiers work on the network
Lean below 60%, Medium from 60%, High from 70% — but High only when both fighters have at least 3 prior UFC bouts and at least 31 of 33 features are available. Thin samples are capped at Medium however large the probability.
Notice the design: the probability alone never earns a High label. Evidence quality does. The same idea appears as FULL / LIMITED / INSUFFICIENT DATA on the Fight Simulator and as sample-size tiers on Fight DNA — see Confidence & gates.
A no-call is information
Well-run models refuse to answer when they should: too close to call, a debut fighter, missing features, a bout that changed. A no-call is the model telling you its number would not be trustworthy. Treat it as a result, not an absence.
Even a High-confidence, calibrated 75% loses one time in four. Confidence labels describe the inputs; they never remove the randomness of the sport.