Provenance and Acknowledgements

This book grows from DS 6210 materials developed and taught at the University of Virginia in Spring 2026: lecture derivations, assignments, optimizer cards, journal-club prompts, numerical studies, and the patterns revealed by student work. Those materials supplied many opening phenomena and much of the reveal order. Book prose was rewritten to stand without the classroom, and every load-bearing mathematical or empirical claim was independently derived, executed, or routed to a primary source.

External tutorials are credited as intellectual maps rather than silent technical authorities. In particular, the 2026 ICML optimization-theory and probabilistic-numerics tutorials helped locate useful seams across chapters. The book’s combined phenomenon-first sequence, diagnostic ledgers, and Paper Autopsy Protocol are editorial syntheses; the technical claims inside them remain attributed to their primary literature.

Executable observations carry claim IDs, provenance classes, pinned harness wheels, and content digests. Negative results and rejected explanations are retained when they delimit a claim. Student work remains private editorial evidence unless permission and attribution are recorded explicitly.

The project also owes its production discipline to defects discovered while building the sibling volume, Deep Learning: Making It Learnable. Stable anchors, content-equivalent editions, estimator declarations, source audits, and mechanical promise checks are not ornamental infrastructure; they are the conditions under which a technical book can remain inspectable as it evolves.