How it works

Locked before the bell

Each bout gets one pick, published with the time it was locked. It is never edited, swapped or deleted, whatever happens in the fight. Afterwards the result goes next to it: called it, missed, or a push for a draw. Every graded pick counts in our track record, including the ones we get wrong.

Don't take our word for it. The full pick ledger is public, and every pick is committed to a public GitHub repository before its fight, so the commit history shows exactly when each one was made.

The site moved to this address on 17 October 2026. Picks locked before then (8 to 17 October) were committed at its previous address. They are carried over unchanged, with their original lock times, in the ledger above.

How the model picks

Every fighter gets a rating that moves with who they beat, not just how often they win. Beating a contender counts for far more than beating a journeyman. The model combines that rating with nine matchup factors: age, height, reach, activity, layoff, knockout power, chin, experience, and the quality of the opposition each fighter has faced. It learns only from fights that happened before the card it's picking, so it never sees the future.

We tested it on 2,986 past fights it had never seen:

When the model's lean is within five points of a coin flip, we say "too close to call" instead of pretending. In testing, those fights really were close to 50/50.

When we don't pick

The model needs a fighter's whole career, fight by fight. When we can't verify most of a fighter's record, usually a prospect or a journeyman with no public fight history, the bout says "pick pending" instead of guessing. We'd rather say nothing than make something up.

The numbers

Every figure on a dossier is calculated as of the fight date, including each opponent's record at the time they fought. A statistic needs a minimum sample before we show it; below that, you get a dash and the reason. Fight data comes from Wikipedia and Wikidata (CC BY-SA 4.0).

Corrections

If a result, a record or a figure is wrong, tell us at hello@boxingdream.com. We fix it in the open: a corrected result shows what was first published and why it changed. A pick and its grade are never edited.

The drawings

Fighters are drawn, not photographed, so every card looks the same and nobody's likeness is used. Each drawing is to scale with the fighter's listed height, so a size gap you see is a real one, and it stands in their listed stance. A dashed outline means the height isn't listed, and we don't guess.

What we don't claim

We have never tested the model against betting odds, so we don't claim it beats the bookmakers. Picks name the winner only; we don't predict knockouts or decisions yet. This is not betting advice, and we take no bets.