GUESSING WITH A SCORE

Guessing with a Score opens the black box around prediction without asking the reader to become a programmer or a statistician first. Starting with ordinary acts of guessing, comparing, and checking outcomes, the book builds—one careful layer at a time—toward the ideas that make machine-learning evaluation useful in the real world.

You will learn why a model score is not automatically a probability, why a threshold is a policy choice rather than a law of nature, why false alarms and missed cases must be counted separately, and why an impressive accuracy number can hide weak performance. Precision, recall, baselines, calibration, ranking, regression error, loss, class imbalance, stress testing, drift, delayed outcomes, feedback loops, robustness, and human review are explained through plain language, dialogues, visual reasoning, and practical thought experiments.

The book keeps returning to one first-principles habit: never admire a score before you understand the decision it is being used to make. That habit becomes a practical method for questioning AI and machine-learning systems without fear of mathematics or technical vocabulary.

The final Value Edition turns the concepts into mental training. Readers reconstruct ideas from scratch, break complex problems into smaller questions, challenge assumptions with counterexamples, practice denominator and error intuition, build reliability boundaries, and create a reusable evaluation canvas.

For non-technical professionals, curious beginners, students, managers, and readers who want to understand how predictive systems are judged, Guessing with a Score offers a calm path from uncertainty to evidence—and from evidence to better decisions.