What a Score Is and Is Not - Logistic Scoring

The finale separates exact score mechanics from named probability and loss boundaries. It closes with the explicit honesty boundary for this book.

highlighted = computed this step

What is exact

The score z and the threshold decision are exact. The boundary is z=0.

z=w x+b,\quad z=0

The probability column names sigmoid except at the exact midpoint 1/2. Log-loss also names logarithm.

\sigma,\ln\text{ are named boundary ops}

This is not calibration, not accuracy, and not learning; it pins the decision mechanics on toy data.

\text{toy-data decision mechanics only}