THE ESTIMATE
A winner probability.
A clear pick and the engine’s raw estimate of that fighter’s chance to win.
INDEPENDENT MMA RESEARCH / ENGINE 3
A PICK SHOULD
COME WITH
A CASE.
Final Bell brings historical fight data, statistical modelling and written matchup analysis together, so you can examine the thinking behind a forecast.
See our latest picks ↗WHAT WE ARE
Fight intelligence.
With the evidence attached.
We’re an independent MMA research project, built around a straightforward idea: a forecast becomes more useful when you can see the evidence, the uncertainty and the reasons it might be wrong.
Our current engine is the third iteration of that work. It uses recorded fight history to estimate a winner, then sits alongside a written assessment of the matchup. We keep the probability, the supporting history and the method judgement clearly labelled.
For people who want to look more closely at a fight.
THE CASE
The supporting history.
Relevant performances, opposition and matchup evidence, with a likely route if the pick wins.
THE RESERVATION
What could overturn it.
Counterevidence, thin records, changes in weight class and other limits that deserve attention.
HOW THE WORK FITS TOGETHER
From fight history to a forecast.
- 01 / HISTORY
Build the picture.
Use earlier results and available fight statistics: striking, takedowns, control, finishing history, recent form and strength of opposition. Age and physical dimensions add context.
- 02 / ESTIMATE
Make the call.
The statistical model combines the recorded inputs into a winner probability. Missing history and small samples limit how much confidence a number deserves.
- 03 / EXAMINE
Put the evidence beside it.
The written briefing considers support for the pick, a possible winning route and the strongest reservation. These histories do not establish what caused the model’s output.
- 04 / RECORD
Keep the original.
For the current pilot, forecasts were saved before the event alongside comparison models. Corrections are recorded separately, with the original issue preserved.
WHERE WE ARE TODAY
A research engine.
A test still to come.
Engine 3 was developed after examining earlier test errors. Its later historical replay is exploratory. We need new, future results to find out how well it holds up.
Our first recorded future card is Noche UFC on 12 September 2026. As of this edition, it is unscored. One card cannot establish a stable success rate or prove improvement.
Read the evaluation rules ↗A FEW USEFUL DISTINCTIONS
How to read Final Bell.
What does a percentage mean?+
It is the model’s raw estimate of the selected fighter winning. A higher number does not make the evidence complete or the outcome certain. The estimates are experimental, and some rest on only one or two captured UFC appearances.
Where does the method lean come from?+
It is a separate judgement based on the available histories, conditional on the picked fighter winning. A 63% winner estimate beside “Submission” does not mean a 63% chance of winning by submission. The winner model has not tested these method predictions.
Does the written case explain the model’s reasoning?+
The case presents relevant matchup evidence. It is not a feature attribution or a causal explanation of the model’s output. We show supporting evidence and counterevidence so readers can examine the forecast for themselves.
What happens when the history needs correcting?+
We label the update and retain the original. In the current briefing, one additional Rongzhu result changes Garcia’s estimate from 64.0% to 60.5% in a separate calculation. The original sealed forecast remains available, and the incomplete statistics are disclosed.
See the correction and calculation ↗MEET THE CURRENT EDITION
Thirteen matchups.
Thirteen cases to examine.
See the picks, the likely routes and the reservations.
