Public accountability

Model performance

The original probability. The actual result. Wins and losses, measured on the same terms. Our performance record is open to everyone.

Total predictions: 0
Accuracy: —
Brier score: —
Average confidence: —

Published UFC picks

0 wins · 0 losses · 0 draws · 0 no contests

One original published pick per matchup, scored against its latest confirmed result. Draws and no contests are excluded from accuracy.

No audited probabilistic predictions have been recorded yet. Legacy editorial picks are preserved in the archive, but lack verified publication history and probability estimates. They are not model results.

A dash means insufficient data, not 0% accuracy.

0 original publications0 awaiting results0 draws · 0 no contests · 0 cancellationsMean scored confidence: — / 10

Cumulative performance over time

Original predictions, ordered by event start. All results remain in the cumulative score; the latest 12 event checkpoints are displayed.

No scored history yet

Time-series performance begins after decisive fight results are recorded.

Calibration / predicted vs observed

For a well-calibrated model, 70% forecasts should win roughly 70% of the time over a sufficiently large sample.

50–55%Predicted —
Observed — · n=0

55–60%Predicted —
Observed — · n=0

60–65%Predicted —
Observed — · n=0

65–70%Predicted —
Observed — · n=0

70–75%Predicted —
Observed — · n=0

75–80%Predicted —
Observed — · n=0

80–90%Predicted —
Observed — · n=0

90–100%Predicted —
Observed — · n=0

Blue: predicted win rate. Green: observed win rate. Empty buckets have no scored observations.

How to read this record

Accuracy is correct picks divided by decisive results. Brier score is the mean squared difference between Fighter A’s probability and their binary outcome (1 for a win, 0 for a loss). Lower is better; a constant 50% forecast scores 0.25.

Calibration compares the picked fighter’s average probability with their observed win rate. A 70% forecast should win about seven times in ten over a large, comparable sample. Bucket upper boundaries are exclusive, except 100%. Small samples can be noisy.

Confidence is an assessment from 0–10, not another win probability. Draws, no contests, cancellations and pending results are disclosed above and excluded from binary accuracy and Brier score. Each fight contributes only its first published version. Result corrections are append-only and update evaluation transparently.

Performance by confidence
Group	Scored / published	Wins / losses	Accuracy	Brier
No predictions in this cohort.

Performance by weight class
Group	Scored / published	Wins / losses	Accuracy	Brier
No predictions in this cohort.

Performance by predicted method (winner accuracy)
Group	Scored / published	Wins / losses	Accuracy	Brier
No predictions in this cohort.

Performance by model version
Group	Scored / published	Wins / losses	Accuracy	Brier
No predictions in this cohort.

Historical events
Group	Scored / published	Wins / losses	Accuracy	Brier
No predictions in this cohort.

Inspect the prediction archive Read the methodology