Skip to content

SupeRR Selector · Season 2 · Rajasthan Royals × Unstop

Backing players under pressure

Definition · Data as of 18 Nov 2026

Pressure is the swing available to one ball

Pressure is not a slot in the batting order. It is the win-probability swing available to one ball. Before each ball of a chase we ask the model two questions: what is the win probability if the batter hits six and survives, and what is it if the ball is a dot that takes a wicket. The gap between those outcomes is the pressure score of the ball.

1

The formula

WP is the win probability of the chasing side from a match state (runs needed, balls left, wickets lost). High pressure is PS ≥ 0.3; low pressure is PS < 0.12. About 6.6% of chase balls clear the high-pressure bar, which is why pressure samples are small and why shrinkage exists.

The full ball-by-ball trace lives in the report PDF, drawn from the formula rather than the private model file. The explorer below carries the same semantics: pick any state of a chase and see the win probability and the six-versus-wicket swing.

Figure 1. The report's Figure 1 traces a full illustrative chase (a 120-ball chase of 172, wickets at balls 40, 72 and 96): pink dots with PS ≥ 0.30 cluster where the chase is genuinely live and vanish once the equation is decided. See the methodology report PDF.
2

The model behind WP

The model behind WP is a gradient-boosted tree classifier constrained to be monotonic: needing fewer runs raises the win probability, more balls remaining raises it, more wickets lost lowers it. It was trained on the 842 decided chases of the pack; grouped 5-fold cross-validation (whole matches held out, never a single match split across train and test) gives AUC 0.889. Reported That is good but not perfect, which is what a noisy sport should produce.

One artefact needed a fix. Raw PS ran hot in the first over of an innings (mean 0.159) and the last (0.174) against 0.126 in the middle overs, an artefact of state extremes rather than a real swing. Each over's mean PS was normalised to the overall mean (0.141), which flattens the bias without touching variation within an over.

3

Chase explorer

Set a match state. The explorer shows the chasing side's win probability, the two outcomes of the next ball, and their gap: the pressure score of that ball.

Chase explorer

Illustration · not the report's model (see note below)

Loading the win-probability grid…