ProbGap compares prediction-market consensus with an independent forecast.
We do not attempt to forecast every market.
Instead, we focus on markets where there is enough reliable information to form a meaningful independent view — and where that view differs materially from the market.
Our process is designed around one simple question:
What probability would we assign if we did not know the current market price?
Only after building that independent forecast do we compare it with the market.
1. Market selection
ProbGap focuses primarily on markets within:
AI
Technology
Macro
Business and major product-related events
We generally avoid markets where:
resolution criteria are unclear
reliable external data is limited
liquidity is extremely low
the outcome depends mainly on rumors or unverifiable information
the market cannot be evaluated using a structured forecasting process
Every market must have a clearly defined outcome and resolution date.
2. Independent forecast
Before comparing our view with the prediction market, we build an independent probability estimate.
This forecast considers several layers of information.
Base rates
We begin with the historical likelihood of similar events.
Examples may include:
previous product launch timing
historical policy decisions
previous company behavior
economic base rates
comparable events
Base rates help prevent recent headlines from receiving too much weight.
Current evidence
We then evaluate current information relevant to the event.
This can include:
official statements
company filings
economic data
product roadmaps
government releases
primary sources
credible reporting
other verifiable evidence
Evidence is assessed both by strength and source reliability.
External consensus
Where relevant, we compare multiple external indicators.
These may include:
other prediction markets
analyst consensus
economic forecasts
futures markets
options markets
official guidance
independent forecasting platforms
No single source determines the ProbGap forecast.
3. Market comparison
Only after the independent forecast has been formed do we compare it with the prediction-market probability.
For example:
Market probability: 38%
ProbGap forecast: 54%
Probability gap: +16 percentage points
This difference is what we call the ProbGap.
A large gap does not automatically mean the market is wrong.
It means our independent assessment differs enough from consensus to warrant closer attention.
4. Confidence
Every published forecast also receives an internal confidence assessment.
Confidence reflects factors such as:
quality of available evidence
number of independent sources
strength of historical base rates
clarity of resolution criteria
uncertainty surrounding the event
A large probability gap with weak evidence may not be published.
ProbGap prioritizes quality over quantity.
5. Forecast revisions
Probabilities change when information changes.
When meaningful new evidence emerges, ProbGap may revise a forecast.
Previous forecasts are not deleted or rewritten.
Each forecast version remains part of the historical record.
For example:
September 14
ProbGap: 54%
September 27
ProbGap: 62%
October 8
ProbGap: 47%
This allows readers to see how our view evolved over time.
6. Track record
Every published forecast is timestamped and evaluated once the market resolves.
ProbGap uses probabilistic scoring rather than simply counting whether a forecast was “right” or “wrong.”
Our primary metric is the Brier Score.
For binary events, the Brier Score measures the distance between a forecast probability and the actual outcome.
Lower scores are better.
Over time, we compare:
ProbGap Brier Score
with
Prediction-market Brier Score
using probabilities observed at the same point in time.
This allows us to evaluate whether ProbGap adds forecasting value relative to market consensus.
7. Calibration
A well-calibrated forecasting system should behave consistently over time.
For example:
If ProbGap assigns approximately 70% probability to many different events, roughly 70% of those events should eventually occur.
As the database grows, ProbGap will publish calibration data alongside its broader track record.
8. No cherry-picking
ProbGap follows several principles:
published forecasts remain permanently recorded
forecast revisions are preserved
resolved forecasts are included in the track record
losing forecasts are not removed
performance is evaluated systematically
The goal is not to appear correct.
The goal is to measure whether the forecasting process improves over time.
What ProbGap is not
ProbGap is an independent forecasting and research publication.
It is not a prediction market, bookmaker, broker or trading platform.
ProbGap probabilities are research estimates and should not be interpreted as guarantees of future outcomes.
Readers are responsible for their own decisions and should consider multiple sources of information.
The ProbGap principle
The market has a probability. We have another one.
When the difference is meaningful, we investigate the gap.