How the evidence is compared.
These are the common rules used across Beyond the Box's public basketball metrics. They make an estimate easier to read, not more certain than the data allow.
Possession context
Rate statistics are easier to compare when pace and opportunity are part of the denominator.
League normalization
Season-relative values help separate player performance from the environment in which it was recorded.
Percentiles
A percentile locates an observed value within its comparison distribution; it is not a probability of future success.
Shrinkage
Small samples are pulled toward a defined prior so unstable observations do not receive false precision.
Recency weighting
Recent seasons can carry more weight when the question is current quality, but the weighting rule remains explicit.
Descriptive versus predictive
Descriptive metrics summarize observed performance. Predictive models estimate an outcome beyond the observed record.
Uncertainty
Reliability, intervals, and missingness describe what the evidence can support; none turn an estimate into certainty.
Role context
Peer comparisons should respect position, minutes, and functional role when those definitions are available and defensible.
Missing data
Unknown fields stay unknown. The system does not manufacture tracking, contract, or validation values to complete a display.
Versioning
Metric and model versions make changes traceable across ranking editions, data periods, and published notes.