What it does
“Model-based matchup simulation using Fight DNA and calibrated fight-state models. Ten thousand simulated fights per matchup, anchored to the PBE Fight Model.” “Outputs are probabilities, not certainties. A 60% favourite loses four fights in ten.”
Two winner numbers: model vs outcome share
The win panel shows the PBE Fight Model probability and the simulation outcome share separately. The simulator's winner split is calibrated toward the Fight Model while keeping its own method and fight-state distribution, so the two are close but not identical — and they never disagree about who is favoured.
| Output | What it is | How to read it |
|---|---|---|
| PBE Fight Model win % | The champion model’s pre-fight probability | The headline win probability; graded in the PBE Algo record |
| Simulation outcome share | Share of 10,000 simulated fights each fighter won | Consistency check against the model; context for method and distance |
| Method shares (KO/TKO, submission, decision) | Share of simulations ending each way | A distribution, not a method pick |
| Goes the distance | Share reaching the judges (includes draws) | Depends heavily on scheduled rounds |
Method vs winner vs distance
These are three different questions with three different reliabilities. Winner probability is the most validated. Method and distance are validated against simple baselines but are harder. The most common method is usually still less likely than all the other outcomes combined.
- A: 58%. Simulated outcomes: A by KO/TKO 24%, A by decision 26%, A by submission 8%, B by any method 42%.
- "A by decision" is the single most common outcome at 26% — and still 74% likely not to happen.
Scheduled rounds and finish time
A 5-round fight has more time for a finish and more time for a decision to swing. The simulator only runs once the official card confirms 3 or 5 rounds; until then it shows rounds unconfirmed rather than guessing a fight's length.
The simulator's model card lists what is not shown because it is not yet validated: finish time, round-by-round winners, point estimates for control time, takedowns landed, knockdowns or submission attempts, and pace-retention claims. Simulations can produce those numbers; they were not accurate enough to publish. Exact finish timing is the least stable quantity any fight simulation produces.
FULL, LIMITED and insufficient data
Before a run, each matchup shows its Fight DNA coverage. After a run, results are labelled FULL or LIMITED (shown with reduced confidence); matchups without enough verified Fight DNA history are not simulated. See Confidence & gates.