Model Learning
After each match the model compares its predictions against the real result and computes a calibration factor per market. Future probabilities are automatically adjusted so the model gets smarter over time.
13808
Resolved predictions
79.0%
Overall accuracy
—
Brier score ( better)
0
Markets calibrated
How factors work
× 0.85
Over-estimating → deflate
× 1.00
Well calibrated
× 1.18
Under-estimating → inflate
Calibrated prob = clamp(raw × factor, 5 %, 95 %)
No calibration data yet
Calibration data is built from resolved predictions (matches that have ended). Save predictions for upcoming matches, wait for them to finish, then click Retrain Model Now.
How the learning works
Predict
For each upcoming match the Poisson model computes probabilities for every market
(Over 1.5, Over 2.5, BTTS, Match Result, Corners…).
Observe
After the match ends the system compares each prediction against the real result
and marks it correct / incorrect.
Adjust
The ratio actual success rate ÷ mean expected probability becomes the
calibration factor. All future probabilities for that market are multiplied by it,
so systematic biases are corrected automatically.
Brier Score measures overall accuracy:
0 = perfect, 0.25 = no better than random guessing, 1 = always wrong.
Lower is better. A good sports model typically achieves 0.15–0.20.