Scales and the one-SD screen
The calibration columns use the broad reference corpus. Maximum distances are measured from the author fit mean and are shown separately for broad and nonfiction calibration. Multiplication changes the displayed scale; nonlinear transforms change the metric.
Lewis: observed book means
Raw units. Books receive equal weight in the author fit mean. These observations are not author-wide limits.
Individual reference excerpts
Opportunities are the metric-specific denominator or count of usable observations. They are not independent books.
Repeatability and overlap
Half correlations compare the first and second halves of the same Brown sample. They do not measure stability across different books by one author. Correlations have no physical unit.
Closest existing and candidate measurements
The closest field is selected on calibration samples. Reserved-sample correlation is reported separately. High correlation can indicate duplicated information.
Brown baseline: every source sample
Source sample IDs identify released corpus samples. They do not identify distinct authors; one sample can combine several articles.