About FightIQ
Last updated 20 September 2026
FightIQ publishes a prediction for every UFC fight, and then publishes whether it was right.
Who runs it
FightIQ is built and run by Gordon Shepherd. It is a one-person project: the data pipeline, the models, the writing and the site itself are all his work. It is part of the IQ Labs portfolio of independent projects. There is no staff, no newsroom floor, and no institutional backer — which is worth saying out loud, because it is the reason everything on the site is checkable rather than authoritative.
Which is also why the stories on this site are bylined FightIQ and not a person: articles here are drafted by FightIQ's own writing system from the site's own data and the public record, checked by an automated fact gate, and then edited and published by the one person above. How the newsroom works sets that out in full, and lists everything we have retracted.
What it is
Three things. The statistical record of UFC fights going back to 1994, with a page for every bout on it. A page for every fighter on that record. And, for each upcoming card, a probability for every fight — published before the fight, and graded against the result afterwards.
How the numbers are made
Every prediction starts from the official round-by-round statistics of both fighters' previous UFC bouts: strikes landed and absorbed, takedowns and defence, control time, durability, activity, the shape of their finishes, and the quality of the opposition those numbers were built against. Those become features; an ensemble of models trained on three decades of fights turns the features into a probability for each fighter; and a second model, built independently of the first, votes alongside it — where the two agree we say so, and where they disagree we say that too. The betting market's own price is an input rather than something we pretend not to see, and part of the ensemble is deliberately blind to it. The full description — what the confidence tiers mean, how the method is tested, and the places it is weak — is on the methodology page, which is longer than this one and does not flatter us.
The honesty rules
Three of them, and they are more or less the whole product.
- Every pick is graded, win or lose. There is no highlight reel. The losses sit in the same table as the wins, on the same page, sorted the same way, and the toss-ups we got wrong are in there too.
- Nothing is deleted because it aged badly. A call that looks foolish a week later stays up. Pages come down only when the underlying record is corrected — never to tidy away an outcome.
- Corrections are disclosed. When we find a mistake in our data or our arithmetic, we fix it and say what changed and what it changed to. Several of the numbers on this site are lower than they once were for exactly that reason.
Get in touch
Corrections and data errors are the most useful mail we get. Partnerships and press are welcome too. Everything goes to the same address, and a person reads it: gordon@fightiq.win. There is more on the contact page.
This page is regenerated with the site, so the date above is the date this version was built. Questions about anything on it go to gordon@fightiq.win.