How accurate is the forecast?
We don't just claim to be right — we measure it. Every night, the calculator compares its own forecasts with the DWZ values that have since been officially evaluated.
Current figures (last 12 months)
So far 8 compared players — for reliable percentages we need at least 30. We show the raw values anyway:
| Compared players | 8 |
|---|---|
| Tournaments compared | 3 |
| Median deviation | 23,0 points |
| Mean deviation | 34,9 points |
| Systematic bias (official − forecast) | +17,1 points |
| Within 1 point | 0 % |
| Within 3 points | 0 % |
| Within 10 points | 25 % |
How we measure
The DSB reference list is reloaded daily. Immediately before the old list is replaced, the calculator recalculates the forecasts for all stored tournaments from the last six months using the old list — exactly as every user would have seen them.
If a player's DWZ index in the new list has risen by exactly 1, they have been officially evaluated exactly once since the last comparison. The difference between the official DWZ and our forecast is the measurement error. Only this deviation is stored — no name, no tournament number.
Limitations of the measurement
- An index jump can come from a different tournament we don't know about. Such cases produce large outliers — which is why the median is the honest figure, not the mean.
- The forecast doesn't know about carry-over games, later result corrections, or unrecognised foreign/converted ratings (details).
- Only players with an existing DWZ are compared; initial DWZ estimates are not included.
- If a tournament is evaluated more than six months after it ends, it falls out of the measurement.