Publications
Recherches sur les agents IA, backtests et audits de prévisions de marché, et architecture de la plateforme.
Évaluation des prévisions
Log loss vs Brier score explained: how each grades a probabilistic forecast, why log loss punishes confident errors harder, and when to use which metric.
6 min de lecture8 sections
Crypto et marchés
How AI detects crypto volatility regimes using realized volatility, clustering, and time-series models — and why the honest output is uncertainty, not price targets.
6 min de lecture7 sections
Marchés prédictifs
What is Kalshi? A plain-English guide to the CFTC-regulated event exchange, how contract prices become probabilities, and how it differs from Polymarket.
7 min de lecture8 sections
Sport
Expected goals (xG) explained in plain English: what the number really measures, why xG isn't goals in one match, and how it feeds forecasting models.
6 min de lecture7 sections
Sport
How to run thousands of simulated tournaments to turn a single-match model into honest win probabilities — sampling, convergence, and confidence intervals.
6 min de lecture7 sections
Prouvabilité
Anyone can claim they called it. Here's how OpenTimestamps anchors a prediction to Bitcoin so its timing is provable — and how you verify it yourself.
6 min de lecture7 sections
Sport
How a double Poisson model turns attack and defence strength into expected goals, a scoreline matrix, and clean home, draw and away win probabilities.
6 min de lecture6 sections
Crypto et marchés
Football, Binance, Kalshi and Polymarket look like four different worlds — but one probabilistic method reads them all. A plain-English guide to AI forecasting across markets.
5 min de lecture6 sections
Sport
Two forecasts for the same match rarely agree — and the disagreement is the most useful thing on the screen. How to read an AI model against a prediction market.
4 min de lecture5 sections
Évaluation des prévisions
A Brier score is the squared error of a probability — forecasting's lie detector. A plain-English guide with a real worked example, and why calibration beats confidence.
5 min de lecture6 sections
Marchés prédictifs
Prediction market prices are surprisingly accurate — with two documented blind spots. How to read the odds, where they break, and why a losing favourite isn't proof they're wrong.
4 min de lecture5 sections
Évaluation des prévisions
A 70% favourite loses three times in ten — by design. Here's why upsets don't mean the model failed, and how to actually tell a good forecast from a bad one.
4 min de lecture4 sections
