In the book’s repository there is a file called STILE.md, and it has grown to sixty-six kilobytes of nothing but writing rules. It contains no mathematics. It contains the prohibitions I imposed on myself, and the longest section is the list of words I’m not allowed to use.
The first entry is “trivial”. Then come “obvious”, “clearly”, and the most insidious formula of them all, “it is easy to see that”.
These are words I met for years in mathematics books and always read the same way: if this step is obvious and I don’t get it, then I’m the problem. That is the exact moment a student closes the book. People who write those words almost never mean to humiliate anyone, they’re just flagging that a step is mechanical, but the effect on the reader struggling at precisely that point is to be dismissed while asking for help. My ideal reader is the person I was at nineteen, facing a smooth wall. You don’t say “trivial” to that reader.
The rules allow one exception worth describing: empathetic negations stay legal. “The domain is not always obvious” is permitted, because it doesn’t tell the reader they should already know, it tells them the book is aware this is a hard spot. The same word changes jobs depending on who it protects.
Then there’s the rule that catches everyone off guard when I describe it: the structure “sentence, colon, list of three things” is banned. It is the fingerprint of prose generated by a language model, and in a book written today, with the declared help of AI tools, that cadence is the first signal a reader recognises even without being able to name it. When I find one I dissolve it into prose. The only triplets that survive are the substantive ones, where the three cases really are three and really are exhaustive, because there the mathematics dictates the shape and there is nothing to disguise.
Not every rule is a prohibition, and some exist to protect me from proofreaders. In the book I use the decimal point rather than the comma, in an Italian text, and it makes anyone reading the first chapter sit up. It’s a decision, not a slip: the reader I’m addressing writes numbers with a point every day of their working life. I’m generous with the euphonic d, an Italian nicety many editors now prune. I use a conjunction pairing that style guides dislike. All of these would be silently “fixed” by any reviewer, human or otherwise, so they’re written down in the rules file under the most important label of all: do not flag.
And this is where STILE.md stops being a style document and becomes something else. It is a linter configuration, holding both the active rules and the declared exceptions. It serves me when I write at six in the morning and can no longer remember what I decided back in March, and it serves anyone reviewing the text by telling them in advance what is an error and what is my voice. Without that file every review pass would drag the book back toward average Italian, which is exactly where I’m trying to keep it from going.
A technical book isn’t written only by choosing what to say. It’s written mostly by deciding what you will never say, and then honouring that list for three hundred pages.
Last modified on 2026-06-28