Probability, not a promise
A 60% probability is the model’s estimate before match start. It does not guarantee the outcome of a single match.
BESTORAMA PREDICTION
How Bestorama Prediction works
Bestorama publishes statistical probability estimates before a match starts. The public prediction is stored as a pre-match snapshot and is not rewritten after the match begins.
PRINCIPLES
What Bestorama publishes
Pre-match snapshot
The published pre-match snapshot is locked in advance. Once the match begins, the original probability remains unchanged.
Probability ≠ Pick
The Pick gate is game-specific: Dota 2 / CS2 / LoL use 70%+, while StarCraft II uses 65%+. Brood War currently publishes probability only, without a Bestorama Pick.
PRODUCTION MODEL
How the prediction is generated
For Dota 2, CS2 and LoL, public Winner predictions use the production model native-team-player-v6-production-v1: team signal and Native Player Elo are used together. StarCraft uses separate player-vs-player models.
A visitor page does not trigger external data collection. It reads already stored matches and locked Prediction snapshots.
Internal experiments do not replace public probabilities. This page describes only the published production layer.
1Local match history
→
2Opponent strength and context assessment
→
3Series probability distribution
→
4Quality / publication gate
→
5Locked pre-match snapshot
VALIDATION
Prediction history →How model quality is evaluated
Pick Accuracy
Share of correct outcomes among actionable Winner signals that passed their game-specific gate. Probability-only Brood War is not included.
n=129 · 101 correct / 28 wrongModel Winner accuracy
Diagnostic directional accuracy across all scored snapshots, including probabilities around 50%. This is not the same as Pick Accuracy.
n=349Brier score
Evaluates the probabilities themselves across all scored snapshots. Probabilities such as 51–52% remain part of model validation even when no public Pick is published.
log loss 0.6028Out-of-sample backtest
A separate historical validation helps compare model behavior on data outside the current live history.
n=368Important:A single win or loss proves nothing about model quality. Bestorama therefore shows sample size alongside the metrics and preserves snapshot history.
PUBLICATION STATUS
Publication by game
Dota 2
81.25%
Pick Accuracy 70%+ · n=16
WinnerLengthScenarios
Open predictions →
Counter-Strike 2
78.26%
Pick Accuracy 70%+ · n=92
WinnerLengthScenarios
Open predictions →
League of Legends
76.19%
Pick Accuracy 70%+ · n=21
WinnerLengthScenarios
Open predictions →
StarCraft II
—
Pick Accuracy 65%+ · n=0
WinnerLengthScenarios
Open predictions →
StarCraft: Brood War
PROB
Probability-only · Pick disabled
WinnerLengthScenarios
Open predictions →
FAQ
Frequently asked questions
Does the prediction change after the match starts?
No. The public pre-match snapshot is locked before the start and is not rewritten. If a separate LIVE recalculation is shown, it does not replace the original snapshot.
Why do some matches or markets have no prediction?
Bestorama publishes only markets that pass the current publication gate and have sufficient local context. A missing estimate is never replaced with an invented number.
Why does a probability not always become a Bestorama Pick?
Because Bestorama separates probability from an actionable signal. Dota 2 / CS2 / LoL use a 70%+ gate, while StarCraft II uses 65%+. Brood War currently publishes probability without a Pick; those estimates still remain part of Brier validation and model statistics.
What does “confidence” mean?
It is an internal confidence grade for a specific snapshot, not a guarantee of the result. Probability, sample size and accumulated metrics matter more when evaluating quality.
Does 70%+ guarantee a team win?
No. Probability describes uncertainty before the match. Even highly rated outcomes sometimes do not occur, so quality is evaluated across a series of matches.
Is an experimental model used in public predictions?
Public Dota 2 / CS2 / LoL cards use the current Native v6 production model. Internal experiments do not replace published probabilities.
VERIFY WITH DATA
Go to predictions →Look at the history, not promises
Prediction Center provides current predictions, real metrics and completed snapshots with actual results.