How accuracy is calculated
Time-weighted Brier evaluation, community comparison, and event coverage.
Plura evaluates your full probability history rather than a single final answer. A value remains active from its submission time until the next update or the end of the event.
Three ingredients
- Calibration. How well the stated probability matches the actual outcome.
- Time. How long each value remained active.
- Relative value. Whether you were more accurate than community consensus at the same time.
Why an early forecast matters
A forecast submitted during the first half of an event can affect a larger part of your result. A correct value entered one minute before closing covers only that final minute and is not equivalent to an early accurate estimate.
Before your first forecast, the system does not invent a number for you: your relative contribution for that period is zero. The fraction of time covered by your own forecast is called coverage.
The formula for readers who want the details
For an outcome y equal to 0 or 1, the Brier loss of probability p is:
B(p, y) = (p - y)²Your loss and the historical reference loss are integrated over time. The event result is:
points = 100 × (reference loss − your loss)A positive value means your probabilities were more useful than the reference; a negative value means they were weaker. The calculation policy is stored with the published result.
Example
Olena entered 65% early and raised her estimate to 80% after a reliable report. Max entered 90% only shortly before closing. If the outcome is “YES”, both chose the correct side, but Olena has much greater time coverage.
Cancelled events receive no score and do not affect ratings.