2  Preface

This book began as a smaller one — a book of principles — and grew into the question its readers actually asked: how do you build the thing? The thing goes by many names. A statistical computing environment. A clinical reporting pipeline. A platform product. This is a book about building one that can survive contact with regulation, with auditors, and with the truth.

It is written from three sources, and the mix is the method. First, a literature: the regulatory frame, the risk-based validation movement, the open-source ecosystems, and the pipeline tooling — reviewed in Part II with the citations you would expect of a review. Second, practice: two systems run in production by a team you could seat at one table — a twenty-one-service clinical review platform and a validation gateway whose engine combines metamorphic, property, fuzz, and mutation testing under human governance — reported honestly, including where they stop short. Third, principles: nine of them, stated in Part III, each ending with a test you can run on any system, including your own.

The argument, compressed: computation in clinical reporting is now free and excellent, and the ecosystem that made it so is candid that it ships capability without assurance. The industry’s most trusted evidence ritual — independent double programming — remains artisanal. Between validated monoliths and loose components sits an empty cell: a composed environment that generates its own accountability. Filling that cell is an engineering discipline — contracts, hashes, gates, locked references, shipped validation — learnable by a small team, and it is the book’s subject from Part III onward.

The companion volume to this book is a research preprint, Human-Governed Validation of R Packages for Regulated Statistical Computing; where chapters lean on its results, they say so. Its sibling volumes — Modern R in Practice and Clinical R in Practice — carry the craft and the ecosystem map respectively. This one carries the why and the how-it-fits-together.

Read it with suspicion, please. The tests at the end of each principle chapter exist so you can argue with the book using systems rather than adjectives. Trust, as the title of the first draft had it, is a layer — and a layer is only real when you can show where it starts and where it stops.

— Jaime Yan, October 2026