# 12 ยท What it costs, and where it loses *After this chapter you will know four situations in which the method in this book is the wrong one, what it charges that nobody mentions, and what the six days did not produce.* --- ## What was not produced Start with the largest fact, because it is the one most likely to be glossed over. In six days and eighty-eight releases, this project did not write any of the second book. Not one chapter. The `/v2/` tree at v0.5.11 contains an extraction pipeline, a document reader, a core graph, an identity ledger, a deterministic transformer with twelve operators, a file explorer, nineteen briefs, thirty-five registered methods, a retrospective, and three empty book folders. It contains no prose of the book it exists to produce. This is not a hidden failure. It is what the founder asked for on day three, in brief 20, in these words: build all the reference material first, decompose it, and only then draw the plot lines. The retrospective closes the v0.4 era saying the same thing in the agent's voice: "The surface is built; v0.5 is what gets made with it." But it is a cost, and it is the cost you should weigh first. If your deadline is Friday and what you need is chapters, this method will hand you a very good machine and no chapters. ## What was extracted The second fact of the same shape. The universe names twenty-one source documents. One has been extracted. The retrospective says so plainly, under a heading that reads "Observations, held loosely": > One document is extracted; twenty wait. Every mechanism built this era (folders, gates, > shards, ledgers, graph pages) was designed to fan out without redesign. v0.5 finds out. At v0.5.11, v0.5 still has not found out. The fan-out has not happened. Everything in Chapter 5 and Chapter 9 is built on a single pilot document, and the claim that it generalises across twenty more is a design intention with good evidence behind it, not a demonstrated result. Where the corpus is silent, this book says so, and here it is silent. ## The costs nobody mentions ### Attention, not time The obvious cost is hours, and the hours were real: two of the six days ran past fourteen hours between first and last release, and one spanned twenty-two. The less obvious cost is the shape of the attention. Every thirty minutes, something arrives that requires a judgement: is this right, is this what I meant, is this good enough to build on. There is no long stretch of unbroken making, because the making is being done elsewhere. Some people find this energising and some find it exhausting, and the difference is worth knowing about yourself before you commit six days to it. The method does not require this pace. Nothing in the loop breaks if a round takes a day. But the loop is only as good as the review at the end of it, and the review is the part you cannot delegate. ### You will ship things you have not read At twenty releases a day, nobody reads every line. This project's answer is the gates: eighty-four unit tests, sixteen validator checks, byte-identical rebuild proofs, replayed example vectors. Those catch a great deal. They do not catch everything, and the honest statement is that this method involves publishing work you have not personally verified line by line, on the strength of machine checks and the fact that you looked at the result. If that is unacceptable in your domain, this pace is unacceptable in your domain. ### The record has to be public to be worth anything Most of the value in this book comes from the record: the verbatim briefs, the honest release notes, the debt written down, the hour lost to a zombie browser reported the same day. That record is only useful if it is honest, and it is only reliably honest if it is public. A private log of your own failures is a log you will edit. The line "the round also burned an hour on a lesson worth recording" was written into a public table by an agent who knew it would be read. If you are not willing to publish the failures, you will get a much weaker version of this method, because the incentive that keeps the record straight will be missing. ### Debt accumulates in the open, which is not the same as not accumulating The component that draws the graph went from 202 lines to 434 against a stated budget of 250, over roughly two days, and was still at 434 at v0.5.11. Every release that grew it recorded the growth. The remedy was named at v0.4.13 and has not been carried out. The retrospective's framing is "honest debt beats hidden debt", which is true. It is also true that the debt is still there. A public record of a problem is not a solution to it, and a method that makes it easy to record debt makes it easy to keep recording debt. ### The version number is not a measure of anything Eighty-eight releases sounds like a lot, and some of them are one button label. Release count measures the granularity of your release process, not the amount of work. It is a useful number for understanding the cadence and a misleading one for understanding output. The word counts, test counts and file counts in this book are more honest measures, and they are all in Appendix B and Appendix C, computed rather than recalled. --- ## Four situations where you should do something else The first edition of the book this project is writing has a page about where its own argument loses. The same courtesy applies here. ### 1. Your book has a fixed, known shape If you already know the chapters, the order, and roughly what each says, the whole apparatus in this book is overhead. Build the universe first is advice for the case where you do not yet know what the book is. If you do know, write it: an agent will still help enormously with the production, and you need one branch, one gate on links, and a release table, not twelve operator folders. The tell: if brief 20's complaint, "we are already defining the answer before we know what questions we're answering", does not describe a real risk for you, you are not in this book's situation. ### 2. You cannot publish the working record Confidential material, an unfinished argument you are not ready to be seen holding, a publisher who wants an exclusive. If the transcripts and the failures have to stay private, most of the discipline this book describes stops being self-enforcing, and you should expect the record to drift towards flattery. ### 3. You are not going to look at the output every round The loop's quality comes entirely from step 5. An agent shipping every thirty minutes to nobody produces a large quantity of confident, plausible work in a direction nobody checked. That is worse than slow work, because it is expensive to unpick and it looks finished. If you can only review once a week, slow the loop to once a week. Do not let it run without you. ### 4. The work is mostly a single act of judgement Some books are one argument, made once, that either lands or does not. Poetry. A short polemic. A memoir whose value is in the voice. The method in this book is built for work with a lot of structure and a lot of evidence, where the labour is in organising and checking rather than in composing. Applied to a book whose whole value is one person's sentences, it will produce a large amount of machinery around a thing that did not need any. --- ## What it does buy Against all of that, the honest positive case, stated in the terms this book has used throughout. **Six days of work with a complete, checkable record.** Every claim in this book was verified against the repository. Every screenshot was re-taken from the tag its caption names. That is not normal. Most projects, six days later, cannot tell you why a decision was made. **A rate of correction that changes what you can attempt.** When being wrong about a design question costs thirty minutes, you ask different questions. "Meaning without connectivity" was tested by running it, not by arguing about it. That is worth more than the speed itself. **Judgement concentrated where it belongs.** The founder spent six days deciding things and none of it typing. The parts of the work that are uniquely human, knowing what the book is for, noticing something subtly wrong, choosing between two right answers, are exactly the parts the arrangement leaves with the person. **Infrastructure that compounds.** Thirty-five techniques, each used in earnest before being written down, each naming the release where it first shipped, and the superseded ones still listed with their supersession recorded. That register is worth more on day thirty than on day six, and it exists because writing it down was part of the loop rather than a task for later. --- ## The honest summary This method produced, in six days: a working surface for writing a book, an unusually good record of how it was made, and no book. Whether that is a success depends entirely on whether you agree with the decision taken on day three, which was to build the universe before finding the plot. The founder's argument for it is in brief 20 and it is a good argument. It is also a bet, and at v0.5.11 the bet has not yet paid out. This book's own position, marked as the writing agent's judgement: the bet is sound for a book of this kind, where the material already exists and the difficulty is finding the story in it. It would be a bad bet for most books. And the parts of the method that are unambiguously worth stealing, the verbatim briefs, the gates, the honest release notes, the state pane, every-bug-becomes-a-test, are worth stealing regardless of which way you bet. --- **Where the live estate shows this.** The first edition's own account of where its argument loses is at `/v1/about/participant.html`. The v0.4 retrospective, including the observations held loosely, is at `/v2/dev-pack/retro-00-the-v04-retrospective.html`. The three empty book folders are at `v2/books/`, each carrying its own commissioning prompt.