Smashrs predictions come from a diverse ensemble of six submodels, each built on a different view of the game (form, surface, head-to-head history, and more). Model consensus measures how many of those six agree on the same pick for a given match, written as n/6.
Why an ensemble
Any single model has blind spots. Running several independent models and only acting when they line up filters out matches where the signal is shaky. A 6/6 pick is one where every angle points the same way; a 4/6 pick is a genuine call but a closer one.
The consensus gate varies by tour
The bar for acting on a pick is not the same everywhere:
- ATP requires at least 4 of 6 submodels to agree.
- WTA and ITF require the full 6 of 6.
The stricter gate on WTA and the lower-tier ITF events reflects that those fields are noisier and the cheap, high-volume edges only hold up at full agreement. In backtests, loosening the gate there gives back the edge to the vig.
Consensus pairs with model edge: the edge says the price is wrong, consensus says the model is confident about which way. Both are recorded in the backtest track record.