Introducing ARC v3An Expected Box Value Prior RAPM (EBV RAPM)

Why not just count them up?

Contemporary all-in-one impact metrics often regress box score actions onto RAPM to find their weights.

A fitted model can give a steal about 3 times the value of a point scored leaving us with an opaque number that can't be decomposed into how or the degree they affect the game

why not just count em up?

Introducing ARC v3, an Expected Box Value Prior RAPM (EBV RAPM), that aims to quantify impact using possession by possession accounting

What is an EBV RAPM?

The next best thing to Expected Possession Value

Expected possession value (EPV) is the continuous measurement of the expected point value of a possession, updated with every pass, shot angle and defender distance even down to the prior conversion history of a shooter.

Though it requires granular optical tracking data (Second Spectrum), it is a great accounting of actualized production (a tangible accounting of a player's produced value) with less raw abstraction than on/off data, where one need thousands of samples to attempt to properly allocate credit.

Expected Box Value

Expected Box Value is the next best derivative

A calculation of how each box score action (OReb, STL, screen AST, foul, etc etc.) affects the average possession (2026: 1.16 PPP)

For example, after counting nearly 250,000 possessions from the 2025–26 season, a steal was found to be worth roughly 1.3 points.

It erases the opponent's possession, and the possession that follows averages 1.32 points, as it often creates an advantageous position for the "stealing" team.

EBV as the prior

Whereas the prior renditions of ARC separately calculated and summed Box Values together via z-scored weights outside the solve, v3 uses EBV as a starting estimate of a player's skill, a prior, within the solve.