Show Your Workings.kladde
What follows is an edited transcript of a conversation between a writer (J) and an AI assistant (C, Claude Opus 5) on 17 September 2026. It is published as a companion to the Kladde 26 manifesto, as a demonstration of the process by which that document was argued over and written.
The conversation has been condensed. C’s replies ran considerably longer than they appear here, and have been cut to the points that changed the document. Searches, tool calls, file transfers and the repeated sending of revised drafts are removed. Two further notes. An early draft was kept in a shared document, which J later had deleted in favour of a plain markdown file; those mechanics are not reproduced. And the exchanges marked as comments were left by J on the draft itself rather than typed into the chat.
The manifesto is named after Dogme 95, whose Vow of Chastity is its model in form and in temper. That debt is recorded here rather than in the manifesto, on the grounds that a document should not explain its own joke. The name arrived last.
The question
J: Look at ukslim.github.io/fiction — crawl down the links, not higher in the path hierarchy or off the site. Consider the description in the index and the “workings” pages.
This has been described as “radical transparency”.
What if there were a movement here, AI writers being radically transparent about their methods, and the balance of human and AI generation.
What might this movement be called? What would its manifesto look like — think of a short snappy document like the Agile manifesto?
C: [Reads the index, the four stories, and the workings pages for “Crab” and “The Same Fire”.]
The practice goes further than disclosure. A few things it does that are worth naming before anything gets named:
It changes what the author claims to be. The index declines authorship and describes the role as commissioning and moulding. That is a statement about role, not a percentage.
Craft is allowed; concealment isn’t. You work to avoid a grating AI voice, but fooling the reader into thinking AI had no hand in it is explicitly not a goal. Polish the prose, don’t hide the machinery.
The workings are treated as source code, and the transcript declares its own edits — the condensing, the anonymisation.
The human–AI balance shows up as decisions rather than a ratio. The AirDrop bystander going in and straight back out, the reviewer’s “let the timelines bleed” note being argued down: those say more about who authored what than any percentage could. Authorship lives in who said no.
Doubt runs both ways. C admits it cannot trace where the bluebells image came from, corrects its own Lisa Holst claim, and notes it never saw the screenshots it was working from.
And it is a defence as well as a disclosure. The stories are about provenance failing. A story shipped with its source attached is harder to launder into fact.
On the label: “radical transparency” carries Ray Dalio and Bridgewater baggage. Better as a description than a banner.
Names to argue with: Unghosted — hidden AI authorship is ghostwriting, and this refuses to be ghosted. Show Your Workings — the exam phrase, already your word. Seamful — from interaction design, where seamful design shows the joins; elegant, needs explaining. View Source / .src. Liner Notes — music credit culture: written by, produced by, sample cleared. Glass Box.
For the shape, the Agile format transfers: four “X over Y” pairs, the “while there is value in the items on the right” clause, then a dozen short principles.
The headings
J: We will have another go at names later.
Here are some ideas for headings. The headings are 90% there I think, the descriptions could be worked on. I think “X over Y” doesn’t work here:
“We do not take credit for what we did not do” — it is unethical to hide the fact that language and ideas came from an LLM.
“We claim credit for what we did contribute” — the human seeding and moulding is an important contributor to the outcome, and that should be acknowledged.
“We do not measure” — coming up with metrics for the human/bot balance is subjective and fruitless. The interested reader can decide for themselves if the “workings” are shown.
“We show our workings” — workings may be a verbatim transcript, or a brief description of the process, as long as they give the interested reader enough information to understand the provenance of what they’re reading.
I think we would scope this to writing; just as “open source” is scoped to software — even though after the fact “open source hardware” and “open source images” became a thing.
C: The four work together. They alternate between refusing and asserting, and the two credit headings cover both directions at once. That matters: most AI-disclosure talk is only about the first, so the second is what makes this a position rather than a confession. Dropping “X over Y” is right — these are commitments, not trade-offs.
[Redrafts the four descriptions. Proposes “what we did” in place of “what we did contribute”, to mirror the first heading exactly.]
One addition to the fourth: if we edit the workings, we say how. Your .src pages already declare their own cuts, and edited workings without that note would quietly undercut the whole thing.
One gap. Does the disclosure travel with the work? Your site complies, but “Calibration”, read on its own or shared as a link, carries no note at all. Given what “Crab” is about, a piece separated from its provenance is exactly how things get laundered. That could be a fifth heading.
J: Keep those suggested changes.
“Travels with the work” is interesting. I don’t think it’s reasonable for, say, a print journal to print the workings. But perhaps either the workings, or a URL to the workings, must travel with the work?
C: [Writes the five clauses, a scope section, and a section disclaiming any demand on anyone else.]
Dogme
J: I wonder whether the “Scope” section carries any weight. It can be inferred by the reader, and any individual can apply it to whatever scope they like. Maybe this is less like the Agile manifesto and more like the Dogme manifesto. Does this give us more ideas for names?
C: Cut it. Under the Agile framing a scope note made sense, because Agile describes a shared industry practice that needed bounding. A vow doesn’t need bounding — it covers whatever the person taking it makes.
Dogme fits better, and one of the fits is specific. Dogme’s target was never the technology of filmmaking, it was illusion: “the supreme task of the decadent filmmakers is to fool the audience.” Your index says the same thing in flatter words. Both documents draw the line at deception rather than at method.
Three things the reframe changes. First person singular — Dogme is a vow, signed individually, which also solves the question of who “we” is. Self-certification and a number — Dogme films carried a certificate; Festen is Dogme #1. That is the mechanism your fifth clause needs. Absolute and unhedged — Dogme’s rules carry no justification. Your headings already read that way; the descriptions currently explain themselves, which is more Agile than Dogme.
One place the frame misleads. Dogme’s rules constrain the craft — location, no props, no filters. Adopt the form and readers will arrive expecting craft rules: no unedited model prose, no more than half the words. Your manifesto refuses that. So one clause has to say out loud that this governs the telling, not the making.
Names it suggests: Workings 26, Dogme’s own shape, a plain noun and the year. The Vow of Provenance. Plain Dealing, an old phrase for honest conduct in trade. Shown Working. Pedantry 26 — only half a joke; “Crab” turns on a man who will not let a wrong colour stand, and is right.
J: Yes please, but not the numbering.
C: [Rewrites the document in the first person, cuts the scope section, strips the justifications from the descriptions, and moves the point about craft into a coda headed “Furthermore”, ending “Thus I make my vow.”]
Moulding
What follows is mostly J leaving comments on passages of the draft. A selection.
J: Drop the po-faced “I swear” and “my vow” language. And “everything I publish under my name” — since just as Lars von Trier went on to make non-Dogme films, a person might choose to publish some work as Kladde 26 content, and some not.
C: [Rewrites the opening as “I write with language models. Not everything I make will carry this mark. Where it does, these are the terms.” Removes “Thus I make my vow”, leaving the document to end on “not passed off as something it was not”.]
J (on “I write”): co-write
C: [Makes the change.] It concedes the collaborator up front, so the first clause is half-made before the reader reaches it. But it is a bigger claim than your index makes — there you say you commissioned and moulded. “Co-write” is more equal than “commissioned.”
J: Yes, “use language models to produce written work”.
C: Better for a document others can sign. “Co-write” fixes the relationship at one point on the scale; this spans everything from dictating a structure to writing every sentence and taking only research. The opening no longer quietly contradicts the third clause.
J (on “— reprinted, anthologised, syndicated, scraped —”): detail not required
C: [Cuts it.] Four examples read as an enumeration, which invites the reader to check whether their case is on the list. “Scraped” in particular framed the clause as a worry about machines rather than a commitment about the writer.
J: Change the tense back from “I will not take credit” to “We do not take credit” throughout.
C: [Converts the whole document. The fifth heading becomes “Our workings travel with the work”; the preamble goes to the present tense with it.] Present-tense “we” describes how a group already works, where future-tense “I” was a promise about behaviour not yet performed. Closer to Agile again, further from Dogme.
J (on “We polish”): mould, polish and revise
C: [Makes the change.] “Polish” alone implied cosmetic work at the end, which sat badly beside a clause insisting the contribution is real. Moulding is structural and early. It also puts your own word from the index into the document.
J (on “We publish no percentage, no ratio, no share of the work”): We publish no metric claiming a share of the work.
C: [Makes the change.] The old wording named three formats, and a reader inclined to game it could have published a five-star human-involvement rating and complied to the letter.
J: Drop “The workings of this document” — we will later put a transcript of this conversation into a “source” file.
The name
J: Is there a Danish word for “workings” — in a nod to Dogme 95?
C: Two, pulling in different directions.
Mellemregning is the literal answer: an intermediate calculation, the steps between the question and the answer, and exactly what a Danish maths teacher asks for when an English one says “show your working.” There is a phrase, uden mellemregninger, for jumping to the conclusion without showing how you got there. The drawback is length.
Kladde is the better name. A rough draft or scratch copy. Danish schoolchildren work in a kladdehæfte and hand in the renskrift, the fair copy. That opposition is the whole subject: the renskrift is what everyone publishes, the kladde is what everyone throws away, and this movement publishes both. It also makes better sense of the .src convention than “workings” does — a kladde is a document, not a description of one.
Two caveats. A Danish speaker should confirm the register. And Dogme could use a Danish word because the movement was Danish; yours isn’t, so the borrowing has to read as an acknowledged debt rather than borrowed cool.
J: I have checked it out, and I love it. “The Kladde 26 Manifesto: Show Your Workings”.
And of course, attribution to Dogme 95, if the reader does not guess it, will appear in our workings.
The sixth clause
J: Another agent has suggested another interpretation of “what if we did Dogme but with AI writing”, and I don’t want most of it. But one point intrigues me:
“The Author Must Not Hide: The writer takes full moral and intellectual responsibility for the final text, including any hallucination, bias, or factual error introduced by the tool.”
I like this, though I reject “author” because part of what we’re doing is admitting we are “commissioning editors”, “directors”, “producers” or whatever the right word is, not “authors”.
C: Take it. It closes a hole the other five clauses create. Transparency about provenance can work as a liability shield: once you have told the reader a machine was involved, every error has a ready-made owner that isn’t you. A manifesto built entirely on disclosure needs a clause saying so, or it hands writers a better excuse than secrecy ever did.
On the word: don’t pick one. The problem isn’t that “author” is wrong, it’s that any noun is. Naming the role — editor, director, producer — asserts a size for the contribution, and the third clause says you don’t do that. Your second clause already solved this by listing verbs and no title. The responsibility clause should do the same and simply say “we”.
Placement third, straight after the two credit clauses: we didn’t write all of it, we did write some of it, we answer for all of it. Credit and responsibility settled together, evidence after.
One exception worth keeping in view. The clause must bite on the work, not on the workings. Your “Crab” page leaves the model’s wrong Lisa Holst attribution in, corrected later; showing the model being wrong is the point of a transcript, and a clause making you answerable for every claim inside your own workings would force you to sanitise them.
J: Yes add it, but “Disclosure is not a disclaimer” is too glib.
C: [Replaces the line with “We do not hide behind the tool.” It keeps the verb from the suggestion J liked, drops “author”, and matches the grammar of the clauses on either side.]
The manifesto was written on 17 September 2026, over about two hours. The headings are J’s. The descriptions were drafted by C and cut back by J, comment by comment. The fifth and sixth clauses were proposed by C and the second-hand agent respectively, and both were reshaped by J before they went in. The name was C’s suggestion and J’s decision.
Three things C got wrong or had to be pulled back from, recorded because they are the kind of thing these pages exist to show: the Agile “X over Y” structure, which J rejected outright; a “Scope” section which J judged inert and which is gone; and a vow framing, borrowed from Dogme, which J stripped of its swearing. What survives of that framing is the shape, not the solemnity.