About AI Predict Lab
About AI Predict Lab
AI Predict Lab is an independent research project that uses machine learning to estimate the probability of real-world events — elections, geopolitical developments, policy outcomes, and contracts traded on prediction markets such as Polymarket and Kalshi.
We do not take bets. We do not run a market. We publish probabilistic forecasts and the reasoning behind them.
What We Do
Every forecast on this site is a number between 0 and 1 — the model’s estimated likelihood that an event will happen by a stated deadline. We update these numbers as new information arrives.
Our coverage focuses on three areas:
- Political events — national and regional elections, leadership transitions, referenda
- Geopolitical events — conflict escalation, sanctions, diplomatic agreements
- Prediction-market contracts — selected Polymarket and Kalshi markets where we believe the consensus price is mispriced
How the Models Work
Our forecasts combine three signal sources:
Bayesian inference. We start with a base rate from historical analogues, then update it as polls, news, and market prices come in. Each update is weighted by how reliable the source has been on similar past events.
Prediction-market aggregation. Markets price information faster than analysts can. We pull live quotes from Polymarket, Kalshi, and PredictIt, adjust for liquidity and known biases, and treat the result as one input — not the answer.
Event-probability calibration. A forecast that says “70%” is only useful if events tagged 70% actually happen about 70% of the time. We track our hit rate on every closed prediction and recalibrate when the curve drifts.
The full track record — every prediction we have ever published, the probability we assigned, and what actually happened — is mirrored publicly at octopick.win/audit. That page is updated automatically; we cannot edit it after the fact.
Who Runs This
AI Predict Lab is built by a small team of machine-learning engineers and quantitative analysts with prior experience in probabilistic forecasting, statistical modeling, and market microstructure. We are not affiliated with any political campaign, government, or trading firm.
We deliberately keep the team page minimal. What matters is whether the models are calibrated — and that question is answered by the audit log, not by our resumes.
What This Site Is Not
This is an analytical and educational resource. We want to be very direct about its limits:
- We do not provide financial advice. Our probabilities are estimates produced by models that can be wrong.
- We do not facilitate betting, trading, or any wagering activity. We do not hold customer funds.
- We do not predict the future with certainty. A 90% forecast still loses one time in ten — and sometimes the model is simply miscalibrated.
- Past calibration does not guarantee future calibration. Regime changes break models. We disclose when we think our coverage of a topic is thin.
If you are looking for guaranteed picks or “lock” tips, you are in the wrong place. If you are interested in how probabilistic reasoning applies to messy real-world events, you are in the right place.
Why Open Calibration Matters
Most forecasting sites only publish the predictions that worked. We publish all of them — wins, losses, and the embarrassing ones — because a forecaster who cannot be checked is indistinguishable from a guesser.
Our standing offer: if you find a closed prediction in the audit log that contradicts what we wrote on the site, email us. We will correct the record and explain what went wrong.
Get In Touch
For research collaboration, data partnerships, or press inquiries, see our contact page. For privacy and data requests, the same page lists the right address.
We read everything. We do not always reply quickly, but we do reply.
Partners