Open the register, write, hand it over

One

Open the register, then write the way you always write

The register is opened when you start writing, not when you have finished.

You open it by saying so, in the chat. There is no command to run and no button: you are talking to the assistant you were going to write with anyway, and it answers.

you   open the register, I am about to write an article

it    Before we start — may what you tell me be quoted in a
      record that is handed to other people?

you   no, it's for a client

it    Understood: I'll record what you asked for, never your
      words. What are you writing, and where will it go?

you   a piece on X for Y, about 1,500 words — I'll draft it
      myself and use you for research and the edit

Your answers are the words you would have used anyway. There is nothing to phrase carefully: the one question that matters has a yes or a no in it, and anything that is not a clear yes is treated as the careful answer.

Those two exchanges go down as the first entries, before a line of the text exists. It also tells you two things at the start rather than at the end — that the record only sees this conversation, so anything you write elsewhere it will not know about, and that nothing in the record is ever deleted. Both of those go in as declared limits.

From then on you write, and the work is recorded as it happens — research, outline, drafting, revision, titling, every change and who made it.

There is nothing to remember and nothing to fill in. The two things that are asked of you are the two a machine cannot decide: whether the piece is one where your instructions can be quoted or only described, and, at the end, a last read of what the record says about other people before it is sealed.

Applying the method to a text that is already finished is allowed, and it produces a reconstructed estimate rather than a measurement. That is a different claim, and it has to be declared as one — the record says which it is, so a reader is never left to assume.

Two

Close the case, and see the numbers for the first time

When the text is done you say so, and the closing runs: the work is attributed span by span, measured, sealed with a signature and two independent timestamps, and packed into a single file with the verifier inside it. You are told three things and asked one.

you   the piece is done

it    Of the words in the final text, 47% are the model's
      and 53% yours; of the ideas, 31% the model's and 69%
      yours. Almost all of the model's words are in the
      revision — the first draft is 86% yours.

      Two things this does not prove: it can't show the
      record is complete, only that nothing in it was
      changed after the fact, and the ideas figure is a
      judgement, not a count.

it    One thing before I seal it. The record still holds
      what you told me about other people, and passages the
      article no longer has — this is the last point at
      which any of that can come out for free. Do you want
      to read it? It's about forty lines.

you   yes

The numbers are an example — yours will be yours. What is not an example is the shape: the limits arrive in the same breath as the figures, not in a footnote under them, and the last read is offered once whether or not there is anything in the record to find.

Two checks have to pass before any number exists, and they are what make the number worth printing. One rebuilds your published text out of the annotated pieces and compares it, character for character. The other re-derives the measurement from the record. If either fails you are told, and nothing is published.

Then the disclosure is generated — never written by hand, because a hand-written note freezes numbers that were true the day you typed them. It is added at render time, from the measurement, and it does not go in your text.

Three

Hand it over — and sign it, if you want your name on it

The document comes out with the record and the verifier inside it. Whoever you send it to can drop that one file on the verifier and check everything, offline, without asking you for anything.

The last decision is yours and it is optional. A qualified electronic signature over that document covers the article and the evidence in one act, and because a supervised trust service identified you first, the reader gets a person rather than a key. It is the only place identity enters this method at all.

Embed, then sign, in that order. A document signed the other way round shows a perfectly valid signature over pages that hold none of the evidence — the failure that looks most like success. The tools refuse the dangerous half of it: a document that claims to carry the record but does not is not produced at all.

A case without a signature is a complete case. It simply makes no claim about who wrote the text, which is the honest position when nothing has been published to back one.