Q beats Polymarket on 53.0% of events.

Q beats an event when its forecasts have a lower average Brier score than Polymarket’s prices at the same times.

Polymarket · Forecasts from the last 60 days · Updated November 18, 2026

All Polymarket events

Each event counts once, however many markets it lists.

Events Q beats
53.0%1,669 of 3,151 events
Q average Brier
0.2012Lower is better
Polymarket average Brier
0.1999Prices at forecast time

23,056 forecasts on 4,399 markets · 1 tied event · See the events

Geopolitics

Conflict, diplomacy, and security.

Events Q beats
66.7%58 of 87 events
Q average Brier
0.1303Lower is better
Polymarket average Brier
0.1329Prices at forecast time

1,440 forecasts on 188 markets · See the events

Everything else

Every other topic.

Events Q beats
52.6%1,611 of 3,064 events
Q average Brier
0.2032Lower is better
Polymarket average Brier
0.2018Prices at forecast time

21,616 forecasts on 4,211 markets · 1 tied event · See the events

Largest topics by event count

  • Sports 2,167
  • Equities 377
  • Elections 211

How we measure accuracy

Brier scores measure how close a forecast probability is to the outcome. Lower is better. We compare Q with Polymarket’s price recorded at the time of each forecast.

We average the forecast scores within each market, then average the markets within each event, so an event such as a ladder of dates counts once. Every event weighs the same. Q beats an event when its average score is lower. Ties stay in the total but do not count as wins.

The percentage shows how often Q has the better score. Average Brier shows the size of the errors, so a small win and a large miss affect the two measures differently.

Scoring details

Each Brier score is the squared difference between the forecast probability and the outcome, with YES scored as 1 and NO as 0. We include forecasts made in the last 60 days, Eastern time, on markets marked closed or resolved, with a final YES price above 99% or below 1%. Forecasts made after a market resolved, open markets, and other venues are excluded. The Accuracy page uses the same data and scoring.

Q turns any question into a calibrated probability.

  1. 01Classify the questionResolution rule, horizon, and question type.
  2. 02Map the landscapeActors, assets, linked markets, and related questions.
  3. 03Research the evidenceCheck the sources behind each claim.
  4. 04Build scenariosBase rates, pathways, dependencies, and disconfirming evidence.
  5. 05Produce the forecastA calibrated probability with uncertainty and cited analysis.

Signal selection

Q generates many forecasts. The best ones become Signals. A forecast becomes a Signal when the side Q favors costs under 70and the gap to the market clears its category's minimum, 510 for events and 15 for commodities. The usual limit is 7 days to resolution. Non-sports event favorites can qualify up to 14 days out at 50¢ to under 70¢ with a gap of 10 to under 30 points. Each event gets one Signal, held to resolution.

  1. 500forecasts / day
  2. 7–14 daysto resolution, by category
  3. 70¢ceiling on Q's side
  4. 5–15 ptsminimum gap by category

Q gets better over time.

Learning loop

Score each outcome

Compare Q and the forecast-time market price with the resolved result.

Diagnose the error

Trace the evidence, expert weights, and missing inputs behind the miss.

Feed the next model

Turn repeatable findings into fine-tuning and data changes.

Resolve · score · fine-tune · forecast again