Start with the exact question.
Read the resolution rules, deadline, and source. Forecast the contract's outcome, not a loosely related headline.
Our research standardConviction should come
Conviction should come
with a paper trail.
We are building research that helps you understand a market, examine its assumptions, and recognize when the evidence is insufficient.
Separate probability from value.
An outcome can be more likely to win and still be expensive to buy. Compare an estimated probability with the actual entry price and costs.
Make the case against the answer.
Use identifiable sources and include conflicting evidence. A lack of sufficient evidence should lead to uncertainty, not a confident recommendation.
Keep a record of what changed.
Timestamp the evidence and preserve the original forecast. Assess predictions after resolution instead of rewriting them in hindsight.
A hypothetical example
More likely does not always mean better value.
Suppose an estimate puts Yes at 65%, while a Yes share costs 80¢. The expected payout is 65¢, so the expected net result is −15¢ per share before costs.
No has a 35% estimated chance. At a hypothetical price of 20¢, its expected net result is +15¢ before costs. The estimate itself may be wrong.
Estimated Yes probability65%
Hypothetical Yes entry price80¢
Expected net result · before costs−15¢
What you can use today
The explorer uses real Polymarket data. Signed-in users can sync watchlists; invited beta members can request DeepSeek research where current evidence is sufficient. Catalogue probabilities are market estimates. Reports clearly label model estimates, and no advantage over market prices has been established.
Research coverage
You can browse the public prediction-event catalogue across all categories. AI research currently supports open, liquid Yes/No markets in politics and the economy with clear resolution rules. Evidence must be sufficient; sports, crypto, paused and resolved markets are available to explore but are not automatically eligible for AI research.
A fresh completed report uses one credit. Valid shared reports and repeat views do not use fresh-report credits. Insufficient evidence can return no estimate.
A record, not an accuracy claim
Original forecasts and same-time market baselines are saved before resolution. After results are recorded, we compare calibration and Brier scores against the market baseline. A small sample is not evidence of an advantage; no accuracy or win-rate claim is made here.
Your private forecast history is available in your account. The open sample preserves a real forecast with its original date, evidence and prices.
Reports use dated sources collected independently of DeepSeek, a second editorial model review, and deterministic comparisons with executable prices and taker fees. Model review can still miss mistakes. Sources and rules are rechecked on reuse; historical forecasts are preserved. Performance claims require prospective evaluation against market baselines.
Explore the market facts