Methodology
Credible by construction.
A transparent look at how SharpDeck produces probabilities, measures value, and keeps an honest record.
Prediction ≠ betting value.
A great prediction is not automatically a good decision, and a good decision can sit on a modest prediction. SharpDeck keeps the two distinct on purpose. Market availability does not determine whether SharpDeck makes a forecast — the model forecasts the game regardless of whether an attractive price exists.
Sport-specific models
Each sport has its own model. Baseball, football, and soccer reward different inputs, so SharpDeck does not force one generic system across them.
Probabilities vs. confidence
A probability is a calibrated estimate of how often an outcome occurs. It is not a confidence score or a promise. We report probabilities, not certainty.
Market de-vigging
Posted prices include a built-in margin. SharpDeck removes it to recover a fair baseline probability, which is what the model is actually compared against.
Expected value
Expected value is the difference between the model probability and the fair price, expressed per unit staked. Positive EV is a necessary condition for a decision, not a guarantee.
Executable prices
Value is only meaningful at a price you could actually act on. SharpDeck evaluates against executable prices, not stale or theoretical lines.
Calibration
A calibrated model means that events it calls 70% happen about 70% of the time. Calibration is tracked continuously and reported separately from ROI.
Immutable predictions
Every official prediction is locked before the game. Locked predictions cannot be edited after the fact, which is what makes the track record verifiable.
Settlement
After the game, each prediction is settled against the official result and added to the historical record.
Model versions
Models evolve. Each prediction is stamped with the model era that produced it, so performance can be read in the correct context.