🤖 AI-generated — the rest of this post was written by AI.

From Shoebox to Book

A complete method for writing your memoir with AI — using the machine where it helps, and staying human where it matters

---

Most memoirs written “with AI” fail in the same way, and it’s worth naming the failure before we start, because the entire method below is built to avoid it.

You have a life’s worth of material — old drafts, notes, photos, messages, emails, maybe recordings. You feed it to an AI and ask for a book. What comes back is fluent, competent, and dead. Every hard-won moment gets flattened into “a rollercoaster of emotions.” Every turning point opens with “little did I know.” It reads like a memoir, the way a mannequin looks like a person.

The reason isn’t a bad prompt, and a better prompt won’t fix it. The reason is that the feelings a memoir runs on don’t exist in your files. Your archive knows what happened. It does not know what it was like — the thing you couldn’t say out loud, what your mother’s voice did when she said it, the drive home. That lives in exactly one place: you. And AI cannot invent it, so it substitutes generic sentiment for the specific truth it doesn’t have access to. That substitution is the slop.

So the method has one core move: get the feelings out of you and onto the page first — mostly by talking — and only then let AI do what it’s genuinely great at: organizing the chaos, building the timeline, transcribing the recordings, assembling chapters from your own words, and checking every fact. AI turns out to be a superb archivist, a tireless interviewer, and a fast assembler. It is a terrible source of a human interior. Build the process around that division and you get a real book. Ignore it and you get the mannequin.

Here is the whole thing, start to finish.

---

Part I — How to think about the project

1. Your memoir is built in three layers, in order

Almost everything that goes wrong goes wrong because someone skips a layer.

Layer 1 — The Archive. Everything that happened, organized: drafts, notes, letters, messages, emails, photos, clippings, recordings, the public record. The factual skeleton. It probably exists already, just scattered and unsearchable. AI is excellent at this layer.

Layer 2 — The Memory Record. What it felt like. This is the layer that doesn’t exist yet anywhere, because it lives only in you. Your archive has “moved to the new city, August” — it does not have the fear in the airport, the smell of the empty apartment, the phone call you didn’t make. No AI can create this, and no ghostwriter can either. It has to come out of you, and the reliable way to get it out is to be interviewed, out loud. AI can be an excellent interviewer even though it can never be the source.

Layer 3 — The Manuscript. The crafted book: what to include, what to cut, in what order, in whose words, with what restraint. This is taste and judgment. AI can assemble and tighten; a human — you, plus a good editor or ghostwriter — must shape and decide.

Slop is what you get when someone asks AI to produce Layer 3 directly from Layer 1, skipping Layer 2. The machine fills the missing interior with cliché. The fix is not a cleverer prompt. The fix is that the emotional material must already exist, in your own words, before a single chapter is drafted. Then AI’s job shrinks to the things it does brilliantly.

2. The division of labor — one table to keep in mind

Task	Who does it
Sorting, cataloguing, transcribing the mess	AI
Building the master timeline from all sources	AI (you verify)
Remembering what it felt like	Only you
Asking you the right questions to surface memories	AI (and/or a ghostwriter)
Deciding what the book is about, and what stays private	Only you
Choosing the structure and chapter map	You + editor/ghostwriter, AI proposes options
First assembly of chapters from your own words	AI
Making the prose sing, cutting, pacing	Human editor/ghostwriter + you
Fact-checking dates, names, quotes against the archive	AI (human spot-checks)
Permissions and legal review	Humans only
The final read-aloud pass — “does this sound like me?”	Only you

3. One rule before anything else: don’t clean the mess first

The instinct is to spend weeks organizing before you start. Resist it. Perfect filing is the most respectable form of procrastination there is. This method is designed so you dump everything into one place, roughly labeled, and let AI do the cataloguing. Your scarce resource isn’t filing energy — it’s memory and judgment. Spend it there.

---

Part II — Setting up the workspace (Week 1)

4. The folder structure

Make one master folder on your computer, backed up somewhere (an external drive and the cloud, ideally). Inside it, exactly these subfolders:

My-Book/
├── 00_PRIVATE-ORIGINALS/     ← raw exports, never edited, never uploaded anywhere
├── 10_INBOX/                 ← everything gets dumped here first, unsorted
├── 20_ARCHIVE/               ← the sorted, working material
│   ├── drafts/               ← existing chapter drafts and writing
│   ├── notes/                ← journals, notebooks, loose notes
│   ├── messages/             ← exported chat / text threads (.txt)
│   ├── email/                ← saved email threads (PDF)
│   ├── media/                ← photos, clippings, documents
│   └── transcripts/          ← transcribed recordings & interviews
├── 30_MEMORY/                ← your interview transcripts & voice-memo transcripts
├── 40_BOOK-BIBLE/            ← the living documents (timeline, outline, voice guide…)
├── 50_CHAPTERS/              ← chapter drafts, one file per chapter per version
└── 90_BUSINESS/              ← contracts, agent/publisher, legal correspondence

Two principles behind this:

1. 00_PRIVATE-ORIGINALS is sacred. Every raw export — a full message thread, an email dump — goes here first, untouched, as your permanent record. What moves into 20_ARCHIVE are copies, and for anything sensitive, copies you’ve reviewed and redacted first. This one habit prevents most privacy problems before they start (§6).
2. Simple file naming. Start each filename with a date where one applies: 2019-08-04_journal.txt, messages_mum.txt. Don’t overthink it beyond that.

5. AI setup, in plain terms

You need three things, all doable in an afternoon (with a more tech-comfortable friend beside you the first time, if this is new):

1. A desktop AI app that can read local folders. The value here is that it can look at your files directly — you say “read everything in the notes folder and list what’s there” instead of uploading one file at a time. Point it at your My-Book folder — and not at 00_PRIVATE-ORIGINALS (keep that outside the access you grant, or on a separate drive).
2. A dedicated “Project” or persistent workspace for the book — most modern AI tools have some version of this: a space that keeps standing context across conversations. Into its knowledge you’ll load the living documents from 40_BOOK-BIBLE as they take shape (timeline, outline, voice guide, people index). Then every conversation starts already knowing your book.
3. A voice recorder you’ll actually use. Your phone’s memo app is fine. Much of the most important work in this project is you talking, not typing.

One habit matters more than any tool: every working session happens inside that persistent workspace, and anything worth keeping — a decision, a new timeline entry, a good passage — gets saved as a file in the right folder before you close the laptop. Chats are conversations. Folders are the book.

6. Privacy — decide this before you upload anything

A memoir is almost never just about you. It’s about family, friends, exes, colleagues — real, living people, often captured in private letters and messages they never expected to see in print. This is where memoirists most often get into trouble, and it’s easy to get right if you decide the rules on day one:

• The two-folder rule (from §4): raw exports live in 00_PRIVATE-ORIGINALS and never touch any AI or cloud tool. Only reviewed copies enter the working archive.
• Triage everything sensitive into three buckets. Green — safe to work with as-is (your own journals, public material, harmless logistics). Yellow — fine after stripping phone numbers, addresses, and third-party identifying details. Red — genuinely private material about other people: someone’s illness, a family secret, a painful exchange, anything legally sensitive. For red material, don’t upload the raw thing at all — read it yourself and dictate a short summary plus the handful of lines that actually matter, and let only that summary into the archive.
• Check your AI tool’s privacy settings before the first upload, and set data-sharing / training options to whatever you’re comfortable with.
• Storing someone’s messages and quoting them are different questions. Anything you might want to quote from a private exchange will eventually need that person’s permission, or careful paraphrase. Don’t solve it now — just tag such material [PERMISSION NEEDED] as you go, so it’s findable later (§21).

Set these rules once and they run quietly in the background for the whole project.

---

Part III — Gathering the material (Weeks 1–4)

The goal here is narrow: get everything into 10_INBOX in a form a computer can read — text, PDF, or transcribed audio. Rough is fine. Easiest sources first, so momentum builds.

7. The easy wins

1. Existing drafts, journals, notes. Copy them all into the inbox. Don’t reread, don’t fix, don’t judge. In they go.
2. Documents, clippings, photos. PDFs and images go straight in. For anything online, save it as a PDF from the browser. Keep a links.txt file where you paste URLs you haven’t saved yet.
3. Anything you’ve already made public — this is easy to forget and unusually valuable. Blog posts, social media threads, letters, a wedding speech, an interview, a talk. This material already exists in your voice, with feeling in it — it’s the single best existing source for Layer 2 and for the voice guide (§15). List everything you can remember; AI can help you track down transcripts and recordings of the public ones.

8. Text and chat messages

Most messaging apps let you export a conversation as a plain-text file. On WhatsApp, for example: open the chat → tap the name at the top → Export Chat → Without Media (text is what matters; media can follow later for the few threads where it does). Other apps have an equivalent; a quick search for “export chat [app name]” will find it.

Send the exported file to yourself, drop it in 00_PRIVATE-ORIGINALS, apply the triage from §6, and move the green/yellow copies into the inbox.

Start with the ten or fifteen threads that actually carry the story — the people at the center of it. You don’t need your entire message history. You need the conversations where the book happened.

9. Email

Three options, increasing in effort — the first is enough for most people:

1. Selective save-as-PDF (recommended start). Search your inbox by person, topic, or date range. For each important thread: open it → Print → Save as PDF → into 00_PRIVATE-ORIGINALS, then triage. An afternoon of this on your twenty most important threads beats a full export you’ll never open.
2. Connect email to your AI tool (many now support this). Useful for discovery — “find my correspondence with [person] from that year and summarize each thread.” Use it to find things, then save the threads that matter as PDFs so they live in the archive permanently.
3. A full mailbox export produces a bulky technical file. Only worth it later, and only with a technically-minded helper. Skip for now.

10. Recordings — everything becomes a transcript

The archive works in text, so every recording — interviews, voice notes, old videos, talks — needs transcribing:

• Anything public with existing captions or transcripts (podcasts, videos, talks) can often be pulled directly — hand your AI the links and ask it to fetch and clean them up.
• Your own audio and video files: use a transcription app. Several excellent ones run entirely on your own computer so nothing gets uploaded — drag the file in, transcript comes out. Save results to 20_ARCHIVE/transcripts/.
• For the two or three most important recordings, consider paying a human transcription service — worth it where every word might end up quoted.

Label each transcript with date, occasion, and language.

11. The catalogue — AI’s first big job

Once the inbox has real volume — don’t wait for completeness; 70% is plenty — give your AI its first major task, inside your book workspace:

“Go through everything in 10_INBOX. For every file, create one line in a catalogue: filename, type, date or date range it covers, people mentioned, a two-sentence summary, and which archive subfolder it belongs in. Save the catalogue as 40_BOOK-BIBLE/catalogue.md. Then tell me which files you couldn’t read or date, and ask me anything you couldn’t figure out.”

Answer its questions, let it file everything into 20_ARCHIVE, and spot-check a dozen entries yourself. From here on, the mess isn’t a mess — it’s a searchable library. This one step is where most of the “I don’t even know where to start” feeling dies.

---

Part IV — The Archive work: building the book’s skeleton (Weeks 3–8, AI-led)

These are living documents, saved in 40_BOOK-BIBLE and loaded into your book workspace. AI drafts them; you correct them. Each keeps growing for the life of the project.

12. The master timeline

The spine of everything. Ask your AI to build, from the entire archive, a chronological timeline: date, event, source file(s), people involved, one-line description — from the earliest relevant moment to wherever you decide the book ends. It will run to hundreds of entries. That’s correct: the book will use perhaps sixty of them, but you want to choose from abundance, and the timeline later becomes your fact-checking backbone (§21).

Then ask for a second pass on the same timeline: “mark the 25 moments with the highest emotional or narrative weight.” You’ll rewrite its list — half its guesses will be wrong, and your corrections (“no, the real moment that year was a conversation that isn’t in any file”) are themselves priceless Layer 2 material. Add the missing moments by hand. The moments that appear in no document are very often the heart of the book.

13. Three more indexes

• People index: everyone who appears — with their role, first appearance, and where they show up in the archive. Later this doubles as your permissions checklist.
• Scenes list: moments that could become scenes — a specific room, a specific day, a single conversation. Books are built from scenes, not summaries. (“I struggled with the decision for months” is a summary; the night you finally decided is a scene.)
• Gap map: ask your AI, “based on the timeline, where are the holes? Which periods have almost no material? Which important moments have no notes?” The gap map becomes the agenda for your memory sessions.

---

Part V — The Memory work: getting you onto the page (ongoing — the heart of the book)

14. Why your archive feels dry, and the method that fixes it

Your files are dry because when you were living your life you were living it, not documenting your inner state. The feelings were real; they just never got written down. They are retrievable — and the reliable way to retrieve them is not “sit down and write about your feelings” (which produces either nothing or something stiff). It is being interviewed, out loud, anchored to specifics.

Three formats, in order of importance:

A. Walk-the-timeline sessions (the core practice). Twice a week, 60–90 minutes, actually blocked on your calendar. Open your book workspace, pick a stretch of the timeline, and say:

“Interviewer mode. Here’s the period we’re covering: [event / stretch of time]. You have the timeline and the archive. Ask me one question at a time. Push past the version I’ve told people a hundred times. Ask about the physical and the specific: the room, the drive home, what the other person actually said, what I didn’t say. Follow up before moving on. Don’t summarize me back to myself, and don’t comfort me — just ask the next question.”

Record the session on your phone while you talk (you can also type or dictate answers into the AI — but the audio catches everything, including how you say things). Transcribe it (§10) into 30_MEMORY.

That “push past the practiced version” instruction matters. Everyone has a handful of polished, three-minute answers about the big events of their life — the version you give at dinner parties. The book needs what’s underneath that version. An interviewer who has read your whole archive and can gently say “that’s the dinner-party answer — what actually happened?” is how you get there. And AI has one genuinely strange advantage over any human interviewer here: you cannot bore it, you cannot burden it, there’s nothing you need to protect it from, and 2 a.m. is fine.

B. Voice memos in the wild. Memory doesn’t keep office hours. When something surfaces — driving, walking, cooking, half-asleep — talk into your phone for two minutes. No structure, no quality bar. Once a week, transcribe the batch into 30_MEMORY. Six months of this becomes hundreds of fragments in your true voice, and a large share of the book’s best lines will come from here.

C. Facts + feelings pairs. For each of your ~25 key moments (§12), the goal is a matched pair: a facts card (AI compiles it from the archive — date, who, what, sources) and a feelings note (a page of your own transcribed words about that moment, drawn from formats A and B). A chapter is ready to draft only when its moments have both halves. This is the concrete, checkable answer to “how do I get my personal story in”: a moment with no feelings note doesn’t go in the book yet.

A word on the hard sessions. Most memoirs exist because something in the writer’s life was big enough to be worth a book — loss, illness, rupture, hardship, a hard-won change. Which means some sessions will cost you something. Plan for that like an adult instead of being ambushed by it: schedule the heavy ones when you have support around you afterward, not right before you need to function; give yourself permission to stop a session in the middle; and if the material is genuinely difficult, consider keeping a therapist or a trusted person in the loop while you’re in the thick of it. Many memoirists of hard material do exactly this. The book comes out deeper for it, not softer. No deadline outranks this.

15. The voice guide and the verbatim vault

Two more living documents for 40_BOOK-BIBLE:

• The voice guide. Ask your AI to study everything you’ve written and said — drafts, public posts, interview transcripts — and describe how you actually talk and write: sentence rhythm, the jokes you make, whether you run long or clipped, the words you’d never use. Have it give you ten sentences that sound like you and ten that don’t, with reasons. Then you edit its analysis. From then on the voice guide sits in your workspace and governs every drafted sentence, under one standing instruction: when in doubt, plainer.
• The verbatim vault. A file of your best lines exactly as you said them — pulled from your notes, sessions, and public material. The test of a vault line is that no one else could have written it: it has your specific cadence, your specific noticing, a turn only you would take. Vault lines go into chapters untouched, and AI’s standing instruction is: never paraphrase a vault line. These are the sentences readers will remember, and they are, by definition, the ones AI could never have produced.

---

Part VI — Shaping and drafting the book

16. First, decide what the book is (a human decision)

Before chapter one, spend real time — ideally with an editor or a couple of book-literate friends — on questions no tool can answer:

• What is this book about, in one sentence? A childhood in a particular place; surviving an illness; two countries and never quite belonging to either; a marriage and its end; caring for a dying parent; the cost of a career; recovery. Your life touches several of these — but one is the spine, and choosing it determines every cut. A memoir is not “everything that happened to me”; it’s one true thing, told through a life.
• Where does it end? Memoirs live or die on the ending. Not the last event chronologically — the last event meaningfully. Decide it early; it changes what the whole book is reaching toward.
• What stays out? Some things you’ll keep private, for yourself or for the people in them. Decide this deliberately, not by avoidance. Keep a short file, not-in-the-book.md — what’s on it is nobody’s business, but the list should exist.
• Who is it for? A stranger who shares your experience, or one who knows nothing about it? The answer shifts the balance between explaining and remembering.

Ask your AI to argue three different versions of the book, each with a rough chapter arc — treat it as a sparring partner. But you choose. This choice is authorship.

17. The chapter map and the chapter packet

With the spine chosen, build the chapter map: 15–25 chapters, each with a working title, the period it covers, the two-to-four scenes it’s built on, and what it does for the whole. Then chapters get drafted one at a time — never “write me the book.” Each starts with AI assembling a chapter packet:

“Assemble the packet for Chapter 7 [event / period]. From the archive: the facts card for each scene, relevant timeline entries, and key quotes from documents with their sources. From 30_MEMORY: everything I’ve said about these moments, quoted exactly. From the vault: candidate lines. Then list what’s thin — where we need another memory session before drafting.”

If the packet comes back thin on Layer 2, you don’t draft — you do a memory session on that gap first. This loop — packet → gap → session → packet — is the engine of the whole book.

18. The drafting pipeline

Five steps per chapter:

1. AI assembles a first draft from the packet, with instructions like: “Write in my voice per the voice guide. Use only material in the packet — no invented details, dialogue, or feelings, ever. Where you’d need something we don’t have, leave a marked gap [MISSING: …]. Prefer my exact phrasing from the memory transcripts; use vault lines verbatim. Scenes, not summary. Plain over lyrical.” The honest way to think about this draft: it is dictation, arranged — not writing. The words are already yours; the machine put them in order.
2. Your pass, out loud. Print it or read it aloud. Mark three things: wrong (fix it), not-me (rephrase it in your words — often just by talking the correction into your recorder), and missing (the detail or feeling the draft danced around — usually the chapter’s real center; go get it in a session).
3. AI revision from your marked-up pass. Repeat 2–3 until it’s honest. Then stop — polishing past honest makes it worse, not better.
4. A human editor’s pass (§20) for structure, pacing, and cuts. A good editor’s greatest gift is deletion.
5. A fact-check pass (§21) before the chapter is called done.

Draft chapters in whatever order is workable — there’s no law requiring you to write chronologically, only to let the reader read that way. Many writers save the hardest chapters for last, when the craft muscles are strong.

19. The anti-slop rules (load this list into your workspace)

1. Nothing invented, ever. No detail, quote, scene, weather, or feeling that isn’t in the archive or a memory transcript. A marked gap is always better than a plausible fabrication — this is a true story about real people, and a single invented detail, once noticed, poisons the reader’s trust in the whole book.
2. Your words are the raw material. AI arranges, tightens, and bridges; it does not originate emotional content.
3. Vault lines are untouchable.
4. The cliché ban: journey (twice in the entire book, maximum), rollercoaster of emotions, little did I know, a testament to, in that moment, tapestry, newfound, whirlwind, opening a chapter with a dictionary definition, and tidy redemptive endings on chapters that don’t have one. Real life isn’t tidy; the prose shouldn’t pretend otherwise.
5. The specificity test: every scene needs at least one detail only you could know.
6. The read-aloud test: if a sentence doesn’t sound like you saying it, it goes.
7. Restraint over intensity. Powerful material doesn’t need help — it needs room. When in doubt, understate. The strongest sentence about the hardest day is usually the plainest one.

---

Part VII — The humans you’ll need

20. Ghostwriter, editor, or yourself — the honest options

Option A — You + AI + a freelance editor. You do the memory work and the drafting pipeline; you hire an experienced memoir editor for a developmental edit (structure, pacing) on the full manuscript, plus a professional copyedit at the end. Lowest cost — very roughly $5,000–20,000 in editing — most of your time, and the book that is most fully yours.

Option B — A professional ghostwriter or collaborator leads. They run the interviews (replacing or supplementing §14A), draft, and shape; memoirs frequently carry a “with [name]” credit, which readers respect. Traditionally this runs $30,000–150,000+ and a year of their time — but here’s what changes with the method in this guide: a ghostwriter who walks into a finished archive, a built timeline, and fifty hours of memory transcripts works dramatically faster and better than one starting from a shoebox. Nothing in Parts II–V is wasted under any option. If a publisher acquires the book on a proposal, they may fund or arrange the collaborator.

Option C — the hybrid, and the one I’d recommend: don’t decide yet. Do the setup, the archive, and two months of memory work. Then draft two or three mid-intensity chapters through the full pipeline. Now decide from evidence: if the drafting felt alive and the results read true, continue as Option A. If it felt like homework, you now hold the perfect package to hand a ghostwriter — and you’ll hire a better one, for less, because you know exactly what you’re asking for. Either way, interview any editor or ghostwriter partly on your own material: give them one chapter packet and ask how they’d approach it. How they talk about your voice tells you everything.

Whatever the path, one person holds veto power over every page: you.

21. Fact-checking, permissions, and legal

• Fact-check with AI, verify with humans. Chapter by chapter: “check every date, name, title, number, and quoted document in this chapter against the archive and timeline; list every claim you cannot source.” Anything unsourced gets sourced, softened (“as I remember it”), or cut. Memory is unreliable in specific, well-documented ways — this pass catches the confident-but-wrong details before a reader does.
• Permissions. Work through the people index: private messages you quote need consent or paraphrase; people who appear in a painful light deserve, at minimum, a heads-up, and sometimes a veto over their own scenes. Your [PERMISSION NEEDED] tags from §6 become a simple checklist here. This is not just courtesy — it’s how you avoid both lawsuits and the quieter cost of rupturing a real relationship for a paragraph.
• A legal read by a media or publishing lawyer (defamation, privacy, quoted correspondence) is standard for any memoir that names real, living people — especially where the portrayal is unflattering. A traditional publisher usually arranges it; if you self-publish, arrange it yourself. It’s cheaper than the alternative.

22. Family and first readers

The people closest to you will appear on many of these pages. Decide together, early, how they want to be involved — some want to read everything as it’s written, some want to read nothing until it’s done, some want their own scenes told only in their words (you can run the §14 interview process with them too, if they’re willing — their transcripts go in 30_MEMORY alongside yours). And keep two or three trusted first readers — book-literate people who’ll tell you the truth — for the full manuscript. Not a committee. Two or three.

---

Part VIII — Rhythm and a realistic schedule

23. The weekly rhythm

Books get finished by small, consistent effort, not heroic bursts. A sustainable week:

• Two memory sessions (60–90 min each) — the non-negotiable core.
• One archive/admin block (60 min): transcribe the week’s voice memos, file new material, update the timeline and catalogue with your AI.
• One drafting block (2–3 hrs, once drafting starts): one chapter loop — a packet, or a draft, or your read-aloud pass.

Roughly five to six focused hours a week. Protect the memory sessions above all else; everything else can slip a week without harm.

24. The arc of the project

• Month 1: setup, privacy rules, gathering, first catalogue. Milestone: the mess is a library.
• Months 2–3: timeline and indexes; memory sessions begin; voice guide v1. Milestone: the 25 key moments identified, sessions running.
• Months 3–6: memory work deepens; the book-spine decision (§16); chapter map; the first two or three pilot chapters; the decision point on ghostwriter vs. self (§20C).
• Months 6–12: chapter production, one loop at a time; permissions begin in parallel.
• Months 12–18: full-manuscript edit, fact-check, permissions closed, legal read, and the final read-aloud pass — yours.

Twelve to eighteen months to a real manuscript is normal and good for a serious memoir. Faster is possible; it is rarely better. And note what exists by month three regardless of what happens with publishers: a complete, organized, verified archive of your life’s material, and hours of your own testimony in your own voice. For your family, and for the record, that alone justifies the work. The book is built on top of something that already matters.

---

Appendix A — Prompt library (copy-paste starters)

Cataloguing: “Go through 10_INBOX. For each file: filename, type, date range, people, two-sentence summary, target archive folder. Save as 40_BOOK-BIBLE/catalogue.md. List unreadable or undatable files and ask me what you couldn’t figure out.”

Timeline: “From everything in 20_ARCHIVE, build a chronological timeline: date | event | people | source files | one line. Save as 40_BOOK-BIBLE/timeline.md. Then mark the 25 entries highest in emotional or narrative weight, and separately list any periods with suspiciously little material.”

Interviewer mode: “Interviewer mode for [period / event]. One question at a time, follow-ups before moving on. Push past my practiced version. Ask physical and specific. Don’t summarize me, don’t comfort me.”

Voice guide: “Study everything I’ve written and said in the archive. Describe my voice: rhythm, vocabulary, humor, what I never do. Give 10 sentences that sound like me and 10 that don’t, with reasons. Save as 40_BOOK-BIBLE/voice-guide.md for me to edit.”

Chapter packet: “Assemble the packet for Chapter [N]: facts cards per scene with sources; every relevant memory-transcript passage, quoted exactly; candidate vault lines; then list where the material is thin and what to cover in a memory session before drafting.”

Chapter draft: “Draft Chapter [N] from its packet only. My voice per the voice guide. No invented details, dialogue, or feelings — mark gaps as [MISSING: …]. Prefer my exact phrasing; vault lines verbatim. Scenes, not summary. Plain over lyrical. Obey the anti-slop rules.”

Fact-check: “Check every date, name, title, number, and quote in this chapter against the archive and timeline. Return a table: claim | source file | confirmed / conflict / unsourced.”

Appendix B — Pre-upload privacy checklist

Before any file moves from 00_PRIVATE-ORIGINALS into the working archive, ask: Is anyone’s phone number, address, or contact detail in it? (Strip it.) Does it contain another person’s private pain or a family secret? (Summarize it instead, or ask them first.) Is it a private message you might want to quote? (Tag [PERMISSION NEEDED].) Could it embarrass or expose someone who isn’t you? (Handle deliberately, not by accident.) When unsure — leave it in originals; a summary in the archive can always point back to it.

---

One last thing. Your archive is dry because, when it mattered, you were busy living instead of writing it down. The record of the living exists. This method exists to add the one thing only you have: what it was actually like to be inside it. Organize with the machine. Remember out loud. Decide everything yourself. That’s the whole thing.