Ex-ante and post-mortem: how to know whether the prediction was right

To trust a model that predicts whether a signing will work, it is not enough for it to look good. You have to prove that it gets things right before the outcome happens. Two ideas hold this up: ex-ante, which means before the fact, and post-mortem, which means after the fact. When the two meet consistently, we know the accuracy is real.

What ex-ante and post-mortem mean

Ex-ante comes from Latin and means before the event. It is the evaluation made before the signing happens, using only the information available at that moment. It is the prediction itself.

Post-mortem also comes from Latin and means after the event. It is the analysis made later, looking at what actually happened during the season. It is the real result.

The honesty of the model lies in comparing the two. On one side, what it predicted, the ex-ante. On the other, what actually occurred, the post-mortem. If the two match consistently, across many cases and many seasons, the model truly gets it right.

How we measure accuracy

The method is straightforward. We take years that have already passed. We put the model back into that moment, with one simple rule: it may use only what was known up to then, nothing from the future. Then it makes its prediction, the ex-ante. Finally, we compare that prediction with what actually happened, the post-mortem.

Accuracy comes from that comparison. It is the correct predictions divided by the total. Because we already know the real outcome, we can measure it precisely.

An analogy helps. It is like asking someone to predict the results of a tournament that has already finished, but hiding the scores from them. If the person gets it right without seeing the result, the method has real value. The model is put through exactly that test.

What leakage is, and how we avoid it

Leakage is when the model accidentally peeks at information from the future, data that would only exist after the signing. This inflates accuracy artificially. The model looks brilliant, but only because it saw what it should not have.

We avoid this problem in several layers of checking. The bases are isolated: each league's model is developed on a separate base, with none of the data from the years being evaluated. There is no future data: the prediction uses only what existed at the moment of the signing, and nothing that happened afterward enters the calculation. The validation is forward-looking and real: the prediction is recorded before and only later compared with the outcome. Because the outcome is a verifiable fact and the prediction came before it, there is no way for the model to have cheated. And there is continuous auditing: we review features and time windows to detect any leak, and if something appears, the model is corrected.

That is where the confidence comes from. Real data about what actually happened, compared with a prediction made before it happened.

How this method helps surpass the market

Most signings in the market come from a mix of reputation, recommendation and recent impression. It is a legitimate process, but one exposed to bias: a player who shone in the last few rounds looks safer than he is, and the judgment shifts once the result is already in. That is why the market gets, on average, about one signing in three right.

The discipline described here targets exactly those blind spots. Recording the prediction ex-ante removes the bias of deciding with hindsight. Comparing it with the post-mortem, season after season, separates what truly works from what only seemed to. Preventing leakage ensures the measured accuracy is real, not an inflated number. And models calibrated league by league, over years of data, capture patterns a single spreadsheet cannot see.

This method also speeds up the evolution of the models themselves. Without it, testing each new model would mean waiting for a whole season to end just to see whether the prediction held. By applying ex-ante and post-mortem to seasons that have already finished, SigningLab validates a new model and measures its accuracy immediately, without waiting for the period to pass, and keeps in use only the best model for each league.

The cumulative effect of these techniques is a hit rate far above the market's: about 81 percent on average in 2025, against the roughly 36 percent clubs reach on their own, and higher than the clubs in every league we track. It is not luck or opinion: it is the result of measuring correctly, recognizing that in the evaluation of complex returns every prediction carries a margin of error, and never relying on information from the future.

Why this matters for a club

An accuracy number is only worth something if it is honest. Ex-ante, plus post-mortem, plus the absence of leakage, is what separates a trustworthy model from a mere promise.

It is also what makes the result auditable. Anyone can check, after the season, whether the prediction recorded beforehand came true. There is no way to adjust the answer once the match is over, because the prediction was already written.

SigningLab publishes this confrontation season after season. It is not about looking right. It is about proving it was right before the fact.