Methodology
P-ARC
points per 100 possessions
P‑ARC = [ 0.35 · z(Impact) + 0.65 · z(Box+) ] · s , s = 3.0 / σ(blend)P‑ARC is ARC modified for the low sample variance of the postseason. A playoff run contains between 4 and 28 games. That is too few possessions for the impact term to carry the weight it carries over a full regular season.
The construction is identical to ARC. Impact and Box+ are each standardised across the qualifying pool, combined by weight, then rescaled so that the standard deviation of the pool is exactly 3.0. Only the weights change: impact falls from 0.65 to 0.35.
The weight was chosen by an out-of-sample test. Impact was estimated on one half of the playoff games and used to predict the actual scoring margins of the other half. Across 11 postseasons and 110 random splits, the 0.35 weighting predicted 34 percent better than the 0.65 weighting.
The same effect appears in the regular season. When a full season is used, prediction peaks near 0.35 impact and then declines. When the same season is cut to 85 games, prediction no longer peaks and continues to rise toward pure box production. The cause is sample size, not a difference in how playoff basketball is played.