About These Predictions
This page publishes AI-generated forecasts on Polymarket prediction markets. Each prediction includes a probability estimate, reasoning, and key evidence — published before the outcome is known.
How It Works
An agentic forecasting system analyzes each question. The approach draws inspiration from methodologies used by previous winners of forecasting tournaments on Metaculus. Each prediction is the output of a multi-step pipeline that researches, reasons, and synthesizes a final probability.
Track Record
Open markets — has the market price moved in our predicted direction?1
| Predictions tracked | 60 |
|---|---|
| Correct direction | 39/60 |
| Accuracy | 65.0% |
| Avg edge2 | 12.5pp |
| Annualized return3 | +86.7% |
| p-value4 | 0.0273 |
Resolved markets — final outcome known.5
| Correct | 2/2 |
|---|---|
| Accuracy | 100.0% |
| Brier score6 | — |
| p-value | 0.157 |
These numbers will improve or worsen honestly. Every prediction stays published regardless of outcome.
Proving Predictions Are Real
Every prediction is timestamped on the Bitcoin blockchain using OpenTimestamps. Here's how it works:
- When a prediction is published, we compute the SHA-256 hash of the content
- This hash is submitted to OpenTimestamps calendar servers
- The calendar aggregates hashes into a Merkle tree and anchors the root into a Bitcoin transaction
- The
.otsproof file is saved alongside the prediction
To verify: download the .ots file and the content.md from any prediction. Compute sha256sum content.md and check it against the proof using the OpenTimestamps verifier. The proof traces back to a specific Bitcoin block, proving the prediction existed before that block was mined.
No trust required. The Bitcoin blockchain is the timestamp authority.
This page is for informational and research purposes only. Nothing here constitutes financial advice. Do not make investment decisions based on these predictions.
-
For each prediction, we compare the market price at publication to the current live price. If we predicted higher than the market and the price has since risen (or vice versa), that counts as correct. ↩
-
Average absolute difference between our predicted probability and the market price at publication, in percentage points. Measures how much we disagree with the market on average. ↩
-
Unrealized return per prediction, time-normalized to a yearly rate. Computed from mark-to-market P&L on open predictions only, divided by each prediction's age. ↩
-
Probability of achieving this accuracy or better by chance alone, assuming 50/50 random guessing. Below 0.05 is statistically significant. ↩
-
The market question has been answered and Polymarket has paid out. This is the definitive measure — directional accuracy is just a leading indicator. ↩
-
Mean squared error between our predicted probability and the actual outcome (0 or 1). Lower is better. A Brier score of 0.25 corresponds to random guessing. ↩
This page is for informational and research purposes only. Nothing here constitutes financial advice. Do not make investment decisions based on these predictions.