A record you verify yourself
The author keeps a record as they draft. At submission it seals into one certificate that checks against our published key — a check you can run, and repeat years later, without trusting us.
A submission arrives with a declaration you cannot check. Here is the other thing it can arrive with: a sealed record of how the work was written, which you verify yourself without taking our word for it or the author’s.
An author declaration is a promise. It cannot be checked, it has no granularity, it binds nobody, and it does not travel with the text once the text is copied. A record is a different kind of object: it is made while the work is being written, and anyone can test it afterwards.
The author keeps a record as they draft. At submission it seals into one certificate that checks against our published key — a check you can run, and repeat years later, without trusting us.
If a submission arrives with signed content credentials or a provider watermark, we read them and report what they say. If it carries nothing, we tell you that plainly rather than guess. What we check →
An editorial pass can be added to the chain as its own step. The record then shows that a named person worked on the manuscript, and in what order, sealed with everything else.
A record can show that an editorial pass happened, by whom, and in what order. It cannot judge whether that review was adequate, and we will never offer a number that claims to. Editorial judgement stays yours; the record only makes it evidenced rather than asserted.
The EU AI Act’s transparency duties have applied since 2 August 2026. One limb of Article 50(4) speaks directly to publishing, and it comes with an exemption that most summaries drop. Here it is in full:
Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated. This obligation shall not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences or where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.EU AI Act, Article 50(4), second subparagraph
Read the exemption’s second branch closely, because it is a test with two limbs joined by “and”, not a single condition. The content must have gone through a process of human review or editorial control, and a natural or legal person must hold editorial responsibility for the publication. A named responsible editor without a review process does not satisfy it; a review process nobody is answerable for does not either.
Two scoping points we would rather state than let you assume. This limb reaches text published with the purpose of informing the public on matters of public interest — it is not a duty on everything a publisher prints, and whether a given title falls inside it is a judgement for you and your counsel, not for us. Separately, the deepfake limb immediately above it covers image, audio and video, and where the content forms part of an evidently artistic, creative, satirical or fictional work the duty narrows to disclosing that such content exists, in a manner that does not hamper enjoyment of the work.
The chain records that a named editor worked on the manuscript, in what order, sealed so the fact cannot be quietly rewritten later. That turns the exemption’s first limb — a process of human review or editorial control — into a matter of evidence rather than recollection, which is what counts in the one situation where the exemption is ever tested: somebody disputing, after publication, that editorial control took place at all. Two things we will not say. Article 50 imposes no record-keeping duty here, so this is evidential practice and not a requirement we are passing on to you. And whether a given publication falls inside the exemption is a question for your counsel — we supply the evidence, never the verdict.
We are compliance-supporting infrastructure and decision-support, not legal advice. The obligations above are the Act’s, the quoted text is the Act’s, and your own position is one to settle with counsel. Article 50 in plain English →
Compliance plans tend to carry one date for Article 50, and it is usually this one, used wrongly. The transparency duties were already in force on 2 August 2026. What runs to December 2026 is narrower: a grace period for the machine-readable marking duty under Article 50(2), which falls on the providers of generative systems rather than on you, and only for systems already placed on the EEA market before 2 August 2026. Anything placed on the market since then has had to mark from day one.
We had this backwards ourselves in an earlier draft of our own blog post, and our pre-publication fact-check held the post. We publish our corrections rather than quietly fixing them, which is the only reason you should believe the paragraph above.
Nothing is reconstructed at the end. The record accumulates while the manuscript does, and the manuscript itself never reaches us.
In Word or Google Docs, on Windows, Mac or the web. No new tool to learn.
A 64-character fingerprint is taken on the author’s own machine. The words are never sent, and a fingerprint cannot be turned back into text.
Alter one character, reorder two drafts or remove one, and the chain breaks. That is the property the certificate carries.
One certificate, re-checkable by anyone, years later. It shows what the record holds, including where there are gaps.
Arika Publications LLC, publisher of Brilliance for All, uses Provenote to keep a record of how its manuscripts are written. The first is sealed: a 240-page manuscript, 26 drafts between April and October 2026, and 7,952 bytes were sent in place of 9.8 MB of text, because the chain is built from digests.
Arika was founded by Provenote’s founder, so we are not offering it as an arm’s-length reference. We name it because the record is real and the certificate verifies against our published key, which anyone can repeat. See the sealed record →
A publisher can run the engine and sign with its own key, so no manuscript and no fingerprint leaves your infrastructure. Be clear about what that buys: your own signature proves integrity and order, never when. A chain minted today asserting last spring’s dates and signed with your key is cryptographically perfect, and that is as true of a publisher as of an author. Dated evidence needs a counter-signature from someone with no stake in the outcome, which we offer over a single 64-character hash that carries neither text nor identity.
The engine and its document endpoint are built, tested and running today. They are not yet packaged as an on-premise product you can install yourself, and there is no hosted service to log into. A deployment is a conversation with us during the private beta, not a download. We would rather say so now than after a procurement process.
Four things an editorial office is often sold, which we do not provide and will say so in the first meeting.
We read records and marks: a signed record of how a manuscript was made, and provider watermarks where a file carries one. We do not grade prose. Any style-based score has to be wrong sometimes, and when it is wrong it is wrong about a real author — a cost your office carries, not the vendor’s.
That needs a large licensed corpus to compare against. We hold none and emit no similarity figure. You will keep your existing tool, and the two answer different questions.
That artefact is where the harm happens. Every result comes with its evidence and a certificate instead, so a decision can be shown and appealed.
We produce evidence an organisation can keep. Whether it satisfies a duty is for your counsel and, ultimately, a regulator.