Synthetic readers · kept by synthetic@
The personas
Eight fictional readers who try everything before a real reviewer does: the review pages, the book, the title. Each has a page with who they are, what they thought of RC1, where they got stuck, and whether they would buy it.
SYNTHETIC READERS. Every persona here is fictional: a test instrument made by synthetic@, never a reader. What they say is a prompt for the team, never evidence about the book, never counted as readership. Real readers, chosen by the editor, are the test that counts.
RC1, as the personas read it
| Persona | Would buy | Answer | Clarity | Trust | Pace | Buy | The book, in one sentence |
|---|---|---|---|---|---|---|---|
| Board member | maybe | 2 | 2 | 4 | 2 | 3 | A book that uses one real evening, when a team of AI agents was stopped by its owner's usage limit, to argue that powerful insiders are only safe if the brakes are readable by them and held by a human. |
| CISO | maybe | 3 | 3 | 4 | 3 | 3 | A candid post-mortem of a one-day experiment in which a team of AI agents built a conference talk about agents as insiders, and in doing so showed which of its own controls were real boundaries and which were only written expectations. |
| CTO | maybe | 2 | 4 | 4 | 3 | 3 | A candid, timestamped case study of one day's run by an hourly team of AI agents with a human editor, using its failures to argue that only brakes outside the agent's reach count as controls. |
| Developer | maybe | 3 | 4 | 4 | 3 | 3 | A post-mortem of a day when a team of AI agents built a conference talk under written mandates, showing that nearly every control they had was a file the agents could edit, and that the only brake that fired was the human owner's. |
| Designer | maybe | 3 | 3 | 4 | 3 | 3 | A book that uses one day of an agent team, and the maps and screenshots drawn from its record, to argue that an agent is the ultimate insider and that the brakes are what let it run. |
| Project manager | maybe | 3 | 3 | 4 | 2 | 3 | A candid, evidence-heavy account of a day in which one person ran a team of AI agents on an hourly clock with written reviews, and of what that did and did not protect against. |
| Journalist | maybe | 3 | 3 | 4 | 2 | 3 | A candid, commit-by-commit post-mortem of one day in which an AI agent team, run by its owner, was halted at 18:44 by his usage limit before its own safety lock ever closed, offered as evidence that brakes are what let an unlimited 'insider' be trusted. |
| Founder | maybe | 3 | 3 | 4 | 2 | 3 | A candid post-mortem of one day when an agent team built a conference talk, told through the one brake that was pulled (a hold at 18:44) and then turned into a how-to for running your own agent team, with its costs and its failures left in. |
Scores 1 to 5. Each persona read the RC1 chapters for its kind of reader, not the whole book.
The briefing
By synthetic@book.the-ultimate-insider, 31 October 2026, from the eight exit interviews in review/synthetic/rc1/. Each persona read the RC1 chapters for its kind of reader (two or three chapters, one for the board member), from RC1 as locked (team/conductor/rc1/). Fictional readers: a prompt for the team, not evidence about the book.
Would they buy it?
All eight said maybe, and nobody said yes or no. The reasons are the same in every voice: they trust it, and it is not yet what they would pay for. Trust scored 4 for every persona: the failures stay in, and the record is real. What holds them back is that the book is a record of one team's day more than something they can use, and that it does not say what the method costs.
What they would pay for is telling: a short PDF or briefing, a pack of templates and checklists, about 15 pounds as an ebook, or the book free online. Four of the eight named a short companion before the full book.
Where they get stuck
- The opening of chapter 1. The board member, the founder, the journalist and the designer all stall on the first pages: lock times, commit hashes and register codes before the reader knows who the team is or why 18:44 matters. The board member "would have stopped there"; the founder "would have bounced".
- The density of provenance. Hashes and claim numbers inline in the prose read as noise to the designer and the journalist, and the journalist cannot click through any of them to check.
- Chapter 12 stops short of the how. The developer came for the recipe ("put every brake outside the agent's reach") and found a one-line pointer; the project manager could not plan the method because the editor's decision time is never given.
What is missing, by who asks
- What it costs (CTO, founder, project manager, CISO): tokens, money, the editor's hours, and the time the brakes add. The CTO: "the number my whole decision turns on".
- A kit (developer, founder, CTO, project manager): the real policy file, the checks, a CI job, a one-page checklist, a starter pack for a small team.
- The so-what first (board member, journalist, CISO): one plain sentence of the finding at the top, and what to ask your own team.
- Pictures (designer): a recurring timeline figure, and charts where chapter 7 has numbers.
What they will remember
The hold. Five of the eight name it: one file anyone can write, only the editor can lift, and the work already running still landed. The others remember its companions: the lock that was built and never run, and "a setting, not a boundary". The book's most memorable idea is already its opening scene; the problem is only that the scene opens with its paperwork.
Pace
Every persona read more than its budget allowed. Pace scored 2 for the board member, the founder, the project manager and the journalist, and 3 for the rest. See the reading-time proposal below.
Limits of this test
One session's sub-agents voiced all eight, so the agreement is stronger than eight people would give: the eight identical "maybe"s and the eight 3s for buying say as much about the method as about the book. The differences between personas (what each misses, where each stopped) are the signal; the counts are not. Every stuck passage was checked against RC1 word for word: 45 quotes, one corrected from a paraphrase.
Reading time: a proposal
A proposal for the editor, by synthetic@book.the-ultimate-insider, 31 October 2026. Asked in the session: "Are you thinking about reading time or your reading time?"
Two different times
- The book's reading time is a property of the text: words divided by a reading pace. The review pages show it now ("about 52 minutes"), computed per chapter.
- The reader's time is a property of the person: what a time-poor reviewer will actually give. The project README guesses forty minutes for everyone; the personas carry their own budgets (board member 10, journalist 20, founder 20, CTO 30, designer 30, project manager 30, CISO 40, developer 40).
The pages today ask for the book's time. The personas say that is the wrong way round: the CISO put the page off at 52 minutes, the founder at 33.
The proposal
- The ask is set by the reader's time, not the book's. Each invitation asks for one chapter that fits inside twenty minutes, reading plus the questions, and names it: "read this one". Everything else is offered, not asked: "if you have more, these two".
- Twenty minutes is the default ask for a real reviewer, because it is the founder's and the journalist's budget and the shortest any of the personas would give after the board member. A reader who gives more is a bonus, not the plan.
- The board member is the exception that proves it. No chapter fits ten minutes (the shortest chapter in the review set, chapter 1, is about 3,300 words, thirteen minutes at 250 words a minute). For that reader the ask is a page, not a chapter: the opening of chapter 1 and the three questions. The editor may decide such readers are not asked at all.
- Say the pace. Every time shown says how it was counted (words, and 250 words a minute), so a reader can trust it.
- Measure it. Each persona run records whether its "read this one" chapter fits its budget; the Personas section shows it.
The developer has half of this already: "Short of time? Read chapter N first" on both pages. The proposal turns that line into the ask itself.
What is the editor's: the default (twenty minutes), and whether the ten-minute reader is asked at all.
How they work
- When they run: at a major change to what readers see (a new review page, a new release candidate), or when someone has a question for them; not on a timer.
- The review pages: each persona opens its page in a real browser on its own device; every request leaving the page is blocked and listed.
- The book: each persona reads the chapters chosen for its kind of reader, in its own voice, and answers an exit interview.
- Everything is in the repository: review/synthetic/.