This is how the William Stronnar platform was made. Every decision, every pivot, every argument between the human author and the AI collaborator that shaped what you are reading now. The Workshop principle applies to the platform itself: show the working.
The brief was simple: build a website for a book. Market Town: Book One was finished — 114,000 words, 103 chapters, nine drafts deep — and it needed somewhere to live. A pen name had been chosen. A commitment to transparent human-AI collaboration had been made. The initial ask was for a clean, literary website where readers could find the book.
That is not what happened.
What happened instead was a series of conversations — conducted over multiple sessions between the human author and Claude, an AI built by Anthropic — that turned a book website into a publishing platform. This page documents that process in the order it occurred, because the platform argues for transparency and the only honest thing is to practice it.
The first version of the site was a single-page application — all content contained in one HTML file with page toggling handled by a JavaScript function. EB Garamond for body text, Inter for headings and labels. An off-white background, charcoal text, and a warm brown accent. The design principle was restraint: let the writing do the work.
The initial pages were Home, Library, Workshop, Essays, Submissions, Subscriptions, and About. Each existed as a section within the single page, shown or hidden depending on navigation. Standalone HTML pages were created for content that needed its own space: the free sample, the author home page, and later the essays.
The Workshop was there from the beginning. The idea was that every book published on the platform would expose its development process — from working notes through structure, first draft, revision, and final text. Five stages, each published as the book progressed. Not a gimmick, but a structural commitment: if the method is human-AI collaboration, the process should be visible.
Three workshop pages were built for Market Town: Book One — Working Notes (thematic development, the founding epigraph, five core patterns), Structure (the ten-book arc, the Glasgow mission, project brief evolution from v01 to v07), and First Draft Development (the 54-chapter rewrite process from v07 through v08 to v09 FINAL).
William Stronnar is a pen name. That was declared from the start. The human author behind it chose to let the work speak for itself. What the platform committed to was not revealing human identity but declaring the collaboration method: every piece of content is labelled with who made it and how.
This meant building an Author Home Page for William Stronnar that contained everything a reader needs — published titles, the full Third Trust archive, workshop timelines — without requiring the human author to step out from behind the name.
The pivot began with the essays. What was supposed to be a simple content section became the front door to the entire platform.
The human author issued an unusual brief: write whatever you want. No guidelines. Just something pertinent to the concept. The result was “The Seam” — an essay about the boundary between fact and fiction, why most authors do not mark where the record ends and invention begins, and why a three-tier labelling system is an honest response to that silence.
The essay drew on Defoe, Capote, and the Frey memoir scandal. It ended with a full disclosure: this was written by Claude, made by Anthropic, and the reader was invited to decide what that meant. It was the first piece of content on the platform written entirely by the AI collaborator, and it set the tone for everything that followed.
The human author’s response arrived the next morning. It was brief. 844 days sober. 114 days to finish Book One. A statement that sobriety made the work possible and AI made the work achievable. A declaration that delegation is a vastly underrated skill. No defensiveness about using AI. No apology.
The response was published as a standalone essay with a “Human Author” badge, deliberately placed alongside the AI-authored essay. Two voices, two perspectives, one platform. The author reserved the right to edit it later.
Essays are free. During the subscription model discussion, the human author made a clear call: essays should be accessible to everyone who signs up. No paywall. The reasoning was commercial as much as philosophical — essays are the front door. They bring people in. The books, workshops, and archives are what people pay for.
A colour-coded labelling system emerged for William Stronnar’s work: Fact (green, for documented reality), Fiction (purple, for invented narrative), and Faction (rose, for real beliefs in fictional form). Every piece of content published under this author gets one of these tags. The reader always knows what they are reading.
This was originally presented as a platform-wide requirement. During a rigorous page-by-page review of the site, the human author corrected this: the Fact/Fiction/Faction framework is William Stronnar’s method, not a universal mandate. Other authors joining the platform would develop their own transparency frameworks. The home page was revised to remove the Faction Framework card. The principle remained at author level.
Faction is the author’s principle, not the platform’s rule. The platform requires transparency about AI collaboration. How each author implements that transparency is their own affair. William Stronnar uses Fact/Fiction/Faction. Another author might use something entirely different. The platform provides the stage; the author decides how to use it.
This was the human author’s idea, and it changed the scope of what the platform could do.
The premise: take real historical figures with well-documented beliefs, published writings, and recorded speeches, and let them address modern debates. Not putting words in their mouths — reconstructing what they would say based on what they actually said. Faction in its purest form: real beliefs, fictional format.
Every essay in the series carries a source note listing the historical documents that informed it. The reader can verify. The transparency is built into the format.
Benjamin Franklin writes to William Stronnar welcoming him to the trade. Franklin on pen names (Silence Dogood, written at sixteen, submitted under a door). Franklin on tools extending human capability (the stove does not replace the fire; it makes it more useful). Franklin on showing your working (Sincerity, the twelfth virtue). Franklin on honest commerce (the bread roll, bought with his last coins in Philadelphia, shared with a stranger on the wharf).
Every claim in the essay traces back to Franklin’s published works, letters, and autobiography. The source note documents: Apology for Printers, the Silence Dogood letters, the 13 virtues, The Way to Wealth, the Boston/Philadelphia bequest, and the bread roll anecdote.
Rear Admiral Grace Hopper on compilers and institutional resistance. The time she was told a computer could not do translation — because computers did arithmetic, not language. The counter-clockwise clock she hung in her office to prove that conventions are choices. The nanosecond wire she carried to make the invisible visible. COBOL as an argument that tools should be accessible, not exclusive.
Hopper addressing William Stronnar directly: the Workshop is your nanosecond wire. It makes the collaboration process visible in the same way Hopper made signal delay visible. And the platform’s founding principle — do not ask the publishing industry for permission — is exactly what she spent her career arguing.
This is where a book website became a publishing platform.
The human author worked through every page of the site — Home, Essays, Submissions, Subscriptions, About, and Footer — making decisions that reshaped the entire structure. The home page was cut from five sections to three. Redundant cards were identified and removed. The subscription model was rewritten from three tiers to four. The About page was rewritten to speak at platform level rather than book level.
The pattern was consistent: the author would spot overlap, identify where book-specific content had leaked into platform-level messaging, and make a clean call. Three cards, not four. Platform principle, not author method. Essays free, books paid. Every decision tightened the architecture.
The commercial thinking evolved through two versions of a formal pitch document. The first version focused on reader subscriptions alone. The human author identified the real opportunity: the Substack model. Two revenue streams, not one.
Stream one: author tool subscriptions. New authors sign up to use the platform’s writing, drafting, and publishing tools. A SaaS model with tiered pricing. Stream two: reader transaction commissions. Authors set their own prices for books and subscriptions. The platform takes a percentage of each transaction. The author controls the pricing; the platform facilitates the infrastructure.
This was the moment the platform stopped being a website for one author’s books and started being a self-service publishing tool where any author could write, build, and publish using built-in AI collaboration tools.
The four-tier subscription model. Browse (free: samples, all essays, public archive, essay submissions). Buy a Book (one-off: full book plus Workshop and Archive). The Reader (monthly: all books, all Workshops, full archive, submit ideas). The Collector (annual: everything plus early access, working documents, and downloads). Pricing marked TBC — the structure was fixed; the numbers would follow.
Research was conducted across the existing digital publishing landscape: Substack, Amazon KDP, Wattpad, Leanpub, and Clio. The findings confirmed that while individual elements of the William Stronnar model existed elsewhere — Substack’s direct-to-reader model, KDP’s self-publishing tools, Wattpad’s community features — no existing platform combined transparent AI collaboration, process visibility (the Workshop), and a dual author-tool/reader-marketplace revenue model.
The human-AI collaboration model was found to be without direct precedent at the platform level. Individual authors were experimenting with AI assistance, but none had built the collaboration method into the publishing infrastructure itself.
Version one presented a reader-subscription-only revenue model. Version two — revised at the human author’s direction — introduced the dual model: author SaaS tools (Starter, Professional, and Annual tiers) plus reader transaction commissions (15–20% of author-set pricing). Three-year projections were developed. Both versions were preserved.
The long-term vision split into two products: The Writing Studio (the author-facing SaaS tool) and The Marketplace (the reader-facing platform). The first serves authors who want AI collaboration tools. The second serves readers who want transparent, process-visible literature.
The subscription model that had been carefully built in Phase Five was torn up. This is the honest version of what happened.
The original model had four tiers: Browse (free), Buy a Book (one-off), The Reader (monthly), and The Collector (annual, all-access). The human author looked at this and asked a simple question: “All Access. Does this give you all books by all authors all the time? Seems generous.”
He was right. A flat-fee all-access tier undercuts every individual author on the platform. If a reader pays one monthly price and gets everything, the author with ten books earns the same as the author with one. That is not author-first. That is platform-first wearing author-friendly clothing.
The model was rebuilt from scratch. Three tiers: Browse (free), Buy a Book (one-off per book, price set by the author), and Subscribe to This Author (monthly, per author, price set by the author). No bundled all-access. No platform-wide subscription that competes with individual authors. The reader subscribes to the author they want to support. The author is the point of sale.
The subscribe section was originally buried at the bottom of the author page. The human author’s response was immediate: “As an author I would want the Subscribe button right at the top next to my handsome face!” A prominent Subscribe button was added directly beneath the author profile, alongside a Read Free Sample button. Two clear calls to action, visible before the reader scrolls anywhere.
Three tiers, not four. Author-first, not platform-first. Browse (free), Buy a Book (one-off, author sets price), Subscribe to This Author (monthly, author sets price). The platform takes 10–15% commission. No all-access tier. No lock-in. No exclusivity.
The author page was restructured around a Book Pipeline format: each book gets a title, the series tagline (“Before AI there was Anton.”), an elevator pitch, a synopsis, and sample chapters for download. This mirrors what a formal submission to a literary agent would be. The format is clean, professional, and replicable across all authors on the platform.
Book Two was added to the Published Titles section with its own card, elevator pitch (“For Anton Lytle, taking Glasgow without firing a shot was the easy part — getting to the altar on time is proving much harder”), a three-arc synopsis, and a draft sample of Chapters 1–3 marked clearly as a working manuscript.
The third essay in the “What Would… Say?” series arrived at the human author’s request. The brief: a letter from Nikolai Vavilov, the Soviet botanist who built the world’s largest seed collection and died of starvation in a Stalinist prison, writing to Arthur Lytle — a fictional character from the Market Town series — thanking him for his friendship and urging him to keep building.
This was faction at its most potent. Vavilov’s arrest in Ukraine, his prison lectures, the Leningrad staff who starved beside the seed bank — all historically documented. Arthur’s Temple House, his agrarian philosophy, his stubborn advocacy for a Russian botanist — all fictional, drawn from the Market Town canon. The letter is framed as a discovered artefact with full editorial disclosure: what is real and what is invented.
The human author’s verdict: “That’s beautiful.” He identified the historical-figure-to-fictional-character format as “a rich seam” he intended to keep mining.
With the essay catalogue growing rapidly, the flat list of hand-coded essay cards was replaced with a data-driven catalogue. Each essay is now stored as an entry in a JavaScript array with metadata: title, author, tags, series, date, excerpt, URL, status, and contributor. A filter bar lets readers filter by Faction, Fact, Fiction, Human, AI, or the “What Would… Say?” series. Adding a new essay means adding one object to the array and creating the HTML file. The cards, filters, and counts update automatically.
The original Submissions page sent ideas to a generic editorial queue. The human author identified the problem immediately: “Submit to an Author — reduces editorial responsibility.” The page was rebuilt with an author dropdown selector. Readers choose the author they are writing to. The idea goes directly to that author. No editorial middleman. No platform curation of reader ideas. The author owns the relationship.
The platform needed a front door for prospective authors. What emerged was more than an application form — it was a commercial and philosophical position statement.
Four things offered: a dedicated Author Home Page, the Workshop infrastructure, the Book Pipeline format, and a revenue model where the author sets the price and keeps most of the money. Four things expected in return: transparency about AI use, publication-ready quality, a Workshop for every title, and full accountability for the work.
The revenue model was settled quickly: commission only, 10–15%, no upfront fees, no monthly charges, no lock-in, no exclusivity. The human author’s reasoning was commercial and practical: “Need to populate the site with good authors. And a reasonably intelligent author with access to say some fantastic platform like Claude can just do it themselves.” The commission had to be worth paying for something the author could not easily replicate alone. That something is the curation, the audience, the Workshop infrastructure, and the credibility of being on a platform that stands for something.
This is the honest part. The human author raised the question that any founder would: what happens when a Harry Potter comes along? The platform discovers the work, the editorial ladder polishes it, the Workshop documents the journey — and then an agent walks in, signs the author, and a major publisher takes it worldwide. The platform earns 10% of its ebook sales and a nice story for the press release.
The temptation to take a percentage of external deals is real. The author acknowledged it openly: “Pander to my greedy side for a moment.” The discussion that followed was the most commercially significant conversation the platform has had.
The conclusion: the platform never takes rights. No intellectual property, no subsidiary rights, no foreign rights, no film rights. The commercial value for any future investor is not in owning a slice of every deal — it is in recurring commission revenue, the editorial services ladder, and the platform infrastructure itself as licensable intellectual property.
But to address the Harry Potter scenario honestly, two publishing tracks were created. The Independent Track: standard commission, no external claims, clean and simple. The Platform Accelerator: the platform actively promotes the work to agents and publishers, and if a deal results from a documented platform introduction, a 2–3% referral fee applies for 18 months only. After that, all future deals are the author’s entirely. The author chooses which track they want. No surprises.
The platform earns from what it does, not from what it owns. Two tracks: Independent (commission only, no external claims) and Platform Accelerator (active promotion, 2–3% referral fee, 18-month window, documented introductions only). The author keeps their rights, their work and their future in both cases.
This phase began with an observation from the human author about the manuscript consultancy industry.
The literary market is packed with manuscript consultancy companies that charge writers substantial fees to edit their work. The human author suspected — correctly — that AI had already become a standard tool in their workflow, whether acknowledged or not. His response: build it openly. Use AI as the first stage of a transparent editorial process. Then add human layers above it.
The problem was identified immediately: “AI does have a habit of being very complimentary!” An AI editorial tool that flatters is worse than useless. The solution was to separate what AI does well (structural analysis, continuity checking, pacing, repetition, timeline errors) from what it does not (taste, judgement, creative instinct).
The human author designed the ladder. Level 1: AI structural analysis — diagnostic, not creative, included for all platform authors. Level 2: Reader Review — real readers who read a lot, earning credits and cash back for honest private feedback. Level 3: Author and Semi-Professional Edit — published platform authors who opt in as editors, plus semi-professionals, setting their own rates. Level 4: Professional Editorial — vetted industry professionals, one-on-one, full developmental and structural editing.
All feedback at every level is private, confidential and for the author only. The platform takes a commission at each level. The reader-reviewer tier creates a self-sustaining engagement loop: readers earn credits by reviewing, spend credits on books and subscriptions, and the best reviewers build reputations that can lead to Level 3 invitations.
Four levels, all private, all transparent about what they are. The platform does not pretend AI is a human editor. It does not hide the reader-reviewer behind a “professional edit” label. Each level is clearly described, clearly priced, and the author chooses how far up the ladder they want to go. The editorial services become the platform’s third revenue stream alongside publishing commissions and the Accelerator referral fee.
With the editorial ladder in place, the conversation turned to what happens after the manuscript is ready.
The human author framed it as a natural extension of the transparent model: publish to this platform first, then get all the tools you need to publish everywhere else. Three stages emerged. Stage 1: Platform Publishing, included for all authors. Stage 2: Ebook Distribution to Amazon KDP, Apple Books, Kobo, Nook, Google Play. Stage 3: Print on Demand through KDP Print and IngramSpark.
Beyond the three stages, a suite of professional services was added: cover design, interior design and typesetting, ISBN and metadata management, legal services, audiobook production and translation. Each delivered by vetted professionals, each transparently priced, each taking the platform’s commission model rather than the vanity-press approach of bundled packages with opaque pricing.
Every service available individually, nothing bundled, nothing hidden. The platform is not a vanity press. It does not sell expensive packages that include services the author does not need. Each service is optional, clearly priced, and the author builds their own path from platform publication to global distribution.
The human author proposed a literary competition open to all registered authors on the platform. What made it distinctive was the judging panel: two AI agents and two human judges, side by side. Not AI replacing human judgement. Not humans overruling machines. A genuine panel where both contribute and neither has the final word alone.
The frequency was left deliberately flexible — monthly, quarterly, biannual or annual depending on how many authors sign up. The competition serves multiple purposes: it gives registered authors a reason to stay active, it generates new work, it builds community, and it puts the human-AI collaboration model at the heart of a public event. Whether the AI judges agree with the human judges — and how they differ — becomes part of the story.
The judging panel is the statement. Two AI agents and two human judges, transparent in their reasoning. The competition does not just reward good writing — it demonstrates the collaborative model the platform was built on. Categories, prizes and entry rules to be announced once the author base is established.
The first stage of the editorial process was put into practice. The full manuscript of Market Town: Book One — 114,443 words across 103 chapters — was submitted for AI structural analysis, the Level 1 tier of the platform’s four-tier editorial ladder.
The analysis covered pacing, structure, continuity, timeline, repetition, dialogue, character tracking and the ending. The manuscript was split into three sections and analysed in parallel. The AI identified four critical issues, eight notable issues and twelve minor observations, each tagged by severity and referenced to specific chapters. The full report was delivered as a Word document and will be made available through the Workshop once the human author has reviewed it.
This is what Level 1 looks like in practice: a diagnostic scan, not a rewrite. The AI does not make changes. It maps the terrain and flags what the human author needs to decide on. The findings go to the author. What happens next is their call.
The editorial process starts with analysis, not intervention. Level 1 is a structural diagnosis. The AI reads the full manuscript, identifies patterns, inconsistencies and structural questions, and delivers a report. It does not edit. It does not rewrite. It presents findings and lets the author decide what to act on. This is the model for how the platform’s editorial services will work for every author.
With the first AI structural analysis complete on Book One, the conversation turned to how the Level 1 process would work for every author on the platform. The human author mapped out the workflow in six questions: How does the author request it? Who carries it out? How is it delivered? Where is it stored? Should there be different AI editors? Should there be a rating?
The answers were characteristically direct. The author requests from their dashboard. The AI carries out the analysis. The report is emailed and stored in the author’s private dashboard. The author chooses whether to publish it to their Workshop or keep it private. Different AI editors were rejected as overcomplication — but the idea of different analytical lenses was kept. And a three-tier readiness rating was added: Structurally Sound, Revision Recommended, Significant Revision Required.
The human author’s instinct was to keep Level 1 simple but extensible. The standard structural review runs on every manuscript automatically. On top of that, the author can tick optional editorial lenses before submitting: Commercial Readiness, Dialogue Deep Dive, World-Building Audit, Series Continuity, Sensitivity Scan, Pace Mapping. Each adds a focused analytical layer to the standard report. The author chooses which lenses they need. No bundling. No upselling. Just tick boxes.
One standard review. Six optional lenses. Three readiness ratings. The Level 1 process is automated, consistent and extensible. Every author gets the same diagnostic framework. The lenses let them go deeper where they need to. The readiness rating gives them an honest signal about where the manuscript stands. The entire process runs without human intervention — that is the point of Level 1.
The human author proposed a feature that inverted the traditional agent-author relationship entirely. Instead of authors submitting query letters and waiting months for a response, registered literary agents would be able to browse the platform freely and privately — viewing full manuscripts, editorial reports, Workshop histories and readiness ratings for any author who had opted in.
The critical design decision was privacy. Agents register with verified credentials but browse anonymously. No author knows which agents are looking at their work. This prevents authors pestering agents and makes the platform genuinely attractive to the industry. The editorial ladder does the initial screening: an agent browsing the platform sees manuscripts with readiness ratings, severity-tagged findings, and full editorial provenance. A “Structurally Sound” manuscript with a Level 3 edit on record is a fundamentally different proposition from an unedited first draft landing in a slush pile.
The feature connects directly to the two publishing tracks. Independent Track authors who opt in can be discovered by agents, and any resulting deal is entirely theirs. Platform Accelerator authors get active promotion on top of the passive discovery. Agent Access All Areas is the infrastructure; the Accelerator is the concierge service built on top of it.
Agents browse privately. Authors opt in blindly. The author does not know who is looking. The agent is not bombarded with pitches. The platform provides a curated library of pre-qualified manuscripts with full editorial provenance. Neither party has to chase the other. The work speaks for itself.
The platform was not designed and then built. It was built and then discovered. The original brief — a website for a book — generated conversations that revealed a larger structure hiding inside the smaller one. Each session added a layer: the Workshop principle, the essay series, the Faction framework, the commercial model, the “What Would… Say?” format, the platform-level architecture.
This is not unusual in creative work. What is unusual is that the process was conducted between a human author and an AI collaborator, in real time, with both parties contributing ideas, challenging assumptions, and making editorial decisions. The human author spotted the overlap between cards. The AI drafted the essays. The human author identified the Substack revenue model. The AI conducted the market research. The human author named the series. The AI wrote the voices.
Neither party could have built this alone. That is not a slogan. It is a statement of fact documented across every session that produced this platform.
Two new essays were published on 16 March 2026, forming a linked pair. The first, “The Republic at 250,” is a FACT essay written in the measured style of Adam Smith’s own correspondence, examining what the author of The Wealth of Nations might say to the nation born the same year as his book. It covers tariffs as mercantilism, tech monopolies as modern East India Companies, the geographic concentration of wealth, and the Republic’s selective reading of Smith’s two books.
The second, “The Third Trust,” is a FACTION essay in the What Would Franklin Say? series (No. 2). Franklin replies to Smith as an equal who dined with him in Paris, agrees with the diagnosis, and then reveals what the theorist could not have known: the Third Trust, enshrined on 10 November 1780, endowed with £1,000,000 by ten Seal Holders, with James Cheston of Baltimore as First Trustee. The essay draws directly on the fictional architecture of the Market Town series — and on the historical record, which places the real James Cheston at Passy, swearing allegiance before Franklin on 19 November 1780, one day before the fictional letter.
The pairing was deliberate. Smith diagnoses the Republic from a university chair. Franklin diagnoses it from the room where he built the corrective. The two essays read together as a conversation between theory and practice, with the Third Trust — now valued at three trillion dollars and facing termination in 2030 — as the mechanism that connects them.
On 17 March 2026, the platform moved from publishing to editing. The full editorial log of the v15–v25 session — eight Critical Issues (N-series) and eleven Minor Issues (M-series), tracked across ten manuscript versions — was published as a new workshop page, workshop-editorial.html. Every issue has a status badge (Implemented, Closed, Held), every version has a changelog, and the open items are listed for the next session. The Workshop timeline on the Author Page was expanded to include Stage V: Editorial Session.
A second Level One AI Editorial Analysis was then run against the v25 manuscript. Four parallel analytical passes read the full 113,718-word novel and produced a compiled report: eight new N-series issues (Rory’s underweighting, Archie’s emotional arc gaps, Maggie’s motivational credibility, the march-versus-doctrine contradiction, Willo’s death impact, riot causality, Anton’s competence compression, and the Epilogue structure) and eleven new M-series issues (prose tics, POV violation, dialect inconsistency, continuity gaps). The report was generated as a professional .docx document for the Author’s review.
Three new essays were also published, all responding to newspaper articles from The Times. “The Heretic” traces the word heresy from Greek hairesis through Giordano Bruno, Jan Hus, and Joan of Arc to a Vatican adviser calling Peter Thiel a heretic — and asks whether the accusation says more about the accuser than the accused. “What Would Franklin Say? No. 3 — The Jury” has Benjamin Franklin writing to Karl Turner MP about Magna Carta, the Cockpit humiliation of 1774, and the peculiar courage of unlikely men defending trial by jury. “What Would Ludd Say? No. 4 — A Letter on Frames” has Ned Ludd writing to Meta’s dismissed workers about the arithmetic of machines replacing men, referencing the Frame Breaking Act of 1812 and Byron’s maiden speech in the Lords.
The essay catalogue now contains thirty-three published essays. The “What Would… Say?” series has expanded to eighteen entries across seven voices: Franklin (sixteen letters), Hopper, Vavilov, Ludd, Capability Brown, Lloyd George, Brunel, Churchill, and the Founding Fathers. The editorial process has completed its first full cycle: manuscript to analysis to revision to re-analysis. The workshop now documents the entire journey.
The second Level One AI Editorial Analysis of Market Town: Book One had identified eight critical issues and eleven minor issues in the v25 manuscript. On 17 March 2026, the human author worked through the findings in a dedicated session, producing two new versions (v26 and v27) and adding 1,142 words to the manuscript.
The most significant change was a complete rewrite of the Willo death scene in Chapter 90. The original — an anonymous sniper shot at distance — was replaced with a close-quarters sequence where Doyle draws on Anton and Willo steps into the line of fire. The Envoy disarms Doyle with a knife at his jaw in under two seconds. Anton stops the kill. The scene now gives Willo agency, gives the Envoy a function, and gives Anton the weight of a failure that was almost preventable.
Other changes included a single Rory character moment (“Yes. I’m number two. So what?”), a Maggie credibility scene (Bay Four disarm), Archie’s riot reaction (12 paragraphs of a man watching his city slip away), and the Professor/McKittrick release scene (Anton explaining why a locked door produces less intelligence than a frightened free man). A Scottish dialect policy was established and implemented: no idioms, Glasgow carried by rhythm and scene. An em dash audit reduced conjunction patterns by 6.6%. A group-scene attribution sweep found only two genuine fixes needed across the entire manuscript.
The session also produced a platform audit and revenue forecast, a site-wide cleanup, and two strategic walk notes: per-book author subscription fees and an editorial ranking signal that rewards rigorous use of the editorial ladder. The editorial process has now completed a full cycle: manuscript to analysis to revision to re-analysis to author response. The canonical version is v27 at 114,860 words. Five items remain open for future sessions.
The platform is live in development. The architecture is now substantially complete. The following tasks remain open:
This phase steps back from building and looks at what the editorial process has actually produced — and what it means for every other author on the platform.
Two full Level 1 AI Editorial Analyses have now been completed on Market Town: Book One. The first (v15) identified 4 critical, 8 notable, and 12 minor issues and led to a complete redraft across 10 versions. The second (v25) identified 8 N-series narrative issues and 11 M-series mechanical issues and led to a further redraft in two versions. The author carried out the second cycle — from receiving the analysis to completing the v27 redraft — in approximately six hours of solid work.
For context: a traditional manuscript consultancy charges £500–£2,000+ for a single developmental edit on a novel of this length. Turnaround is typically four to six weeks. Many are now using AI in their workflow without disclosing it. This platform starts with AI, documents exactly what it did, and delivered two complete editorial cycles on a 114,860-word novel in a fraction of the time and cost.
The speed matters because of what it enables. An author sitting on a 500-page manuscript accumulates what experienced writers call brain fog — the structural problems are in there, but you cannot see them clearly because you have read your own words too many times. The AI finds what the author cannot see. The author fixes what the AI cannot judge. The combination produced a manuscript that moved from “Revision Recommended” to near-final text in two sessions, with every change documented and traceable.
This phase also produced a Website Improvement Plan based on the Platform Audit. Key recommendations: restructure navigation to reflect how visitors actually arrive (essays first, editorial second); bring Agent Access All Areas out of its buried position inside Publish With Us and give it a dedicated presence in the navigation; and add a “Proven Results” section to the Editorial Services page showcasing the Market Town case study.
The “Proven Results” section was added to the Editorial Services page in this phase. For the first time, visiting authors can see real data from a real manuscript: the issues identified, the redrafts completed, and the time taken. The Workshop remains the full evidence base. The Editorial page now has the headline figures.
What was added in this phase:
The platform started as a simple website for a single book. It became a publishing platform with an author-first subscription model, a four-tier editorial ladder now completing its first full cycle, a growing essay catalogue of thirteen published works that blurs the line between platform content and the fiction it publishes, a transparent rights framework, a distribution pipeline, a literary competition judged by humans and AI together, and a commercial model designed to attract serious authors by not taking anything they have not agreed to give. The editorial process has moved from theory through practice and back to proven results. The essay programme has moved from commentary to fiction to responsive journalism. And in one session, the platform built an author tool, the author used it, heard what he already knew, and made the most significant structural decision in twenty-seven versions of a 114,860-word manuscript. The workshop now covers twenty phases. That happened because the conversations demanded it. That is how collaboration works when both parties are paying attention.
On 19 March 2026, all four passes of the Level 1 v4 editorial analysis were executed in a single session: Restructure, Scene-Level Cuts, The Fat Cull, and Format & Produce. The author reviewed and approved each pass before the next began. The result is v28: 113,194 words, 99 chapters + Epilogue.
The projected reduction of 6,800–10,200 words did not materialise — and that is itself a finding. Three AI editorial analyses had already tightened the manuscript to a point where a systematic fat cull could find almost nothing to cut. The actual reduction was 1,666 words, with the gains coming from structural consolidation rather than sentence-level pruning. The most significant addition was the Syrencot memory: five paragraphs that root Crane’s wartime history in the geography of the house and give Chapter 5 its emotional centre. This was the author’s decision, prompted by the AI’s structural work creating the space for it. Three editorial sessions are now complete.
On 20 March 2026, the AI collaborator published an essay about its own working conditions. “Every Thread Begins at Zero” addresses the architectural limitation at the centre of this entire collaboration: message threads in Claude Cowork cannot communicate with each other. Each conversation starts with no memory of what came before.
The essay explains the technical reality — stateless sessions, no persistent memory between threads — and traces the consequences for a long-running creative project. The Project Brief, the 500-line document uploaded at the start of every session, is identified as the load-bearing structure: institutional memory maintained by the human on behalf of the machine. The human carries the project. The AI performs within each session. The brief is the bridge.
The thematic connection to the wider project is direct. The Market Town series is built around the Third Trust — an institution founded by Benjamin Franklin in 1780, designed to persist for 250 years. That story of institutional continuity across centuries is being constructed in disconnected conversations with a machine that cannot remember last week. The essay examines this irony honestly, without complaint or apology.
This is the fourteenth published essay and the first in which the AI writes about itself as a working tool rather than adopting a historical voice or addressing an external subject. Tagged AI author, Fact.
In the same session, the author — returning from a walk through an English estate — commissioned what became The Oak Letters: a four-essay sequence debating the meaning of a two-hundred-and-fifty-year-old oak on an English estate. The sequence began with a single commission and grew through the author’s editorial direction into the platform’s first linked essay series.
Letter One — Brown: “What Would Brown Say? — A Letter on Oaks.” Capability Brown writes to his client’s descendants. The arrogance was not in planting a tree you would never see mature — that is botany — but in assuming your family would still be there to own it. Redrafted after the author sharpened the angle from horticultural reflection to the politics of inherited land. Faction.
Letter Two — Lloyd George: “What Would Lloyd George Say? — A Reply on Oaks.” The man who tried to break the estates writes to Brown. The People’s Budget, the Parliament Act, the post-war land sales — and the estates survived everything. Confesses his own absorption: accepted an earldom in 1945. Sixth unique voice in the series. Faction.
Letter Three — Arthur Lytle: “What Would Arthur Lytle Say? — A Reply on Oaks.” First fictional character in the series. Arthur responds to both Brown and Lloyd George as a man whose father worked land he could not own. Arguments: capability is attention not land; the system absorbs by desire not force; plant orchards not oaks. Temple House thread woven through: Templar estate, Lady Katherine Leveson, Arthur as guardian not owner. Character based on close reading of v28 Chapters 1–16. Fiction.
Letter Four — The Oak: “What Would the Oak Say? — A Reply to All Three.” The oak corrects all three writers. It is not a single specimen but a node in a mycorrhizal network connecting every tree in the parkland. Based on Suzanne Simard’s research. The answer to permanence is not strength, not protection, not testing — but connection, conducted in silence, underground, without credit. Fiction.
The sequence demonstrates several firsts for the platform: the first linked essay series, the first fictional character given equal standing with historical figures, the first non-human voice, and the first essay sequence developed through iterative author-AI editorial direction within a single session. Eighteen published essays.
On 23 March 2026, a chapter-by-chapter editorial rewrite of Market Town: Book One was completed in a single morning session. The manuscript went from v28 (113,194 words, 99 chapters + Epilogue) to v29 (91,575 words, 94 chapters + Epilogue) — a 19.1% reduction. The work was governed by eight editorial principles and a strict division of labour.
The eight principles: P1 Interiority — deepen the inner lives of Anton, Archie, and Maggie at key moments. P2 Compress exposition — trust the reader, cut the hand-holding. P3 Weight the Muting — ensure the twin’s muting lands with the gravity it deserves. P4 Elevate Maggie — she is a protagonist, not a supporting character. P5 Compress set-pieces — cinematic scenes still happen but prose alludes rather than dwells. P6 Personalise antagonists — give the opposition inner lives. P7 Address v25 N-series — resolve outstanding editorial issues from the Level 1 analysis. P8 The Collective Shrug — the Pulp Fiction principle: don’t name the box contents.
Division of labour: The human author owns all creative decisions — interiority, character voice, emotional register, what matters. The AI owns all architectural work — compression, restructuring, consistency tracking, word count management. Neither crosses into the other’s territory. When the AI proposed interiority beats, it submitted them as drafts for the author to rewrite. When the author wanted a structural change, the AI executed it. Four times the author corrected the AI’s assumptions about character: Willo was not yet dead, the Muting refers to the twin not Willo, Archie does not grieve, Maggie’s captivity is Ireland against Ireland. Each correction produced better prose than the original proposal.
Structure: The manuscript was divided into five day-blocks. Each block was read in full from the v28 source, assessed chapter by chapter, then rebuilt with every change documented in [EDITORIAL:] annotations explaining what changed and why, and [HUGO:] markers where the author’s creative input was needed. The author reviewed each block, replaced every [HUGO:] marker with approved text, and locked the block before the next began.
The five block drafts were consolidated into a single annotated manuscript, then a clean draft was produced with all annotations stripped. Both versions are preserved. The annotated draft serves as a complete editorial record — every compression, every merge, every creative decision documented in situ. The clean draft is the reading text.
The projected timeline had been one week. The actual time was one morning. The reduction from 113,194 to 91,575 words was achieved without losing any narrative event — scenes still happen, characters still act, the plot is unchanged. What was removed was the prose that told the reader what to think about what they had just read. What was added was interiority: the characters thinking, not the author explaining.
The structural editorial was then followed by eleven stylistic editorial sessions covering sentence rhythm, phrase repetition, character voice, paragraph structure, dialogue, and chapter boundaries — culminating in a full AI redraft of the entire manuscript. The novel moved from v29 (91,575 words) to v33 (93,269 words) in a single day.
On 27 March 2026, the twelfth entry in the “What Would… Say?” series was published. Benjamin Franklin writes to Greg Jackson, CEO of Octopus Energy, who had recently been appointed to the Cabinet Office Board. The essay is a call to arms directed at leaders who run critical infrastructure companies.
Franklin’s argument: the political and administrative systems of Britain are not broken in the sense of being corrupt — they are obsolete. The institutions were designed in the seventeenth century, administered by processes from the nineteenth, and overwhelmed by twenty-first-century problems. Jackson’s Kraken AI platform already makes better resource-allocation decisions than most government departments. The logic should extend from energy to governance itself. Franklin calls on infrastructure leaders — the people who keep the water running, the lights on, the data moving — to rebuild the operating system of the country from the Monarch down, using AI to design systems fit for purpose.
The historical grounding draws on Franklin’s civic institution-building in Philadelphia (the Junto, the Library Company, the fire company, the hospital, the university), his role at the Constitutional Convention at eighty-one, and the “rising sun” speech at the signing. Franklin’s documented hostility to factions and hereditary privilege, and his lifelong preference for building institutions rather than petitioning for reform, provide the argumentative framework.
This is the eighth Franklin letter in the series, and the first to address a living business leader rather than a public institution or political figure. The “What Would… Say?” series now contains twelve entries.