How WeCalledIt works

Before a call is made, the fixture is simulated 40,000 times. Each run is a full ninety minutes played one second at a time, with the ball at a real position on a 105 by 68 metre pitch, moving according to how each side actually plays: how quickly they move it, how direct they are, how high they press.

That is 216,000,000 simulated seconds per fixture. Goals, shots, corners, cards and fouls all fall out of the same run, which is why the markets agree with each other rather than being guessed one at a time.

What we do not publish is the weights, the priors and the tuning. That part is ours.

The engine resolves 23 markets from a single simulation. The desk then publishes only the ones it has real conviction on, usually seven to ten a game, because a short board of strong reads beats a padded one.

A simulation cannot see everything. The desk runs two AI analysts: Tom reads the numbers PRISM returns, form, the matchup, the shape of the scorelines it produced. Sheila reads what it cannot: pressure, motivation, who actually wants it tonight, what is at stake.

Six layers of reasoning go into a call, and nothing ships unless both lenses land on the same side. Where they disagree, we publish nothing, which is why some matches carry two calls and others carry ten. Every argument is published on the match page, unedited.

Calls are committed 24 hours out and timestamped. There is no hindsight and no moving the goalposts once the teams are announced. A call that did not exist before the whistle has never counted towards our record and never will.

When the signal on a match is thin, the desk says so and lowers its confidence rather than fill the gap. A confident call looks different from a cautious one, on purpose. Anything it cannot evidence, it does not claim.

The desk was funded with ₦200,000 to put behind its calls. Every stake and every return is a row on an append-only ledger, so the balance is the sum of what actually happened rather than a claim we are making.

A prediction that costs nothing to get wrong is worth what you paid for it. This is the part that makes the record mean something.

78% right across 3,999 graded calls in 503 matches. That figure includes pre-season friendlies, which grade a little kinder; competitive fixtures only, it is 77% across 269 matches. Both are on the record, win or miss.

Join the channel for every call before every match.

We are not a bookmaker. We take no bets, we hold no money for anyone, and nothing here is financial advice. What we publish is our own analysis and the positions we have taken with our own bankroll. Anyone who follows them does so at their own risk.