About
One sheet, published the way an institution would publish it
NP-RMB-001Dossier v0.9Published 2026-08-1294% of claims carry evidence
What the project is
A print exists in many impressions. This one is a sheet with almost no documented history: a plate securely dated to 1638, an impression that could have been pulled at any point in the two and a half centuries after that, and a single documented sale in 2023 sitting at the end of a gap nobody has closed. That is an unpromising object for a catalogue entry and a good one for an open record, because almost everything about it is a question with a method attached.
The record is published by the owner, acting as steward rather than as an authority. The practical meaning of that distinction is that the project reports what sources say and grades those sources; it never adopts a source’s statement as its own finding. Where an auction house catalogued the sheet a certain way, the record says the auction house said it, names the tier that kind of source occupies, and leaves the question open until something better arrives.
Alongside the object sits the part that is useful to someone with no interest in a privately held sheet at all: a census of 27 impressions of this print across 20 institutions and the art market, each row transcribed from the holding institution’s own published record with the source URL against it, and each disagreement between institutions preserved rather than normalised away.
Why publish at all
Because a record nobody can see cannot be corrected by anyone who knows better. The two discrepancies already logged against this object were both found by checking the owner’s working record against published institutional records — an exercise that only has value if its result is public, including when the result is unflattering to the object.
Who it is for
Print specialists and paper conservators, who can settle in an afternoon things this project cannot settle at all. Cataloguers, who may find the census and the state atlas directly reusable. Students and lecturers, for whom an unfinished record is a better teaching object than a finished one. And anyone who wants to check a claim made here against the source it came from.
What the project is not
It does not attribute
The project issues no attribution and no opinion on authenticity. Neither the steward nor the software is a specialist in Old Master prints, and saying so is more useful than pretending otherwise. Attribution statements on the record belong to the sources that made them and are labelled with their tier.
It does not value
No acquisition figure, no valuation, no market framing anywhere on the site. A build gate scans the source and the data before every deploy and refuses to ship if it finds any. The rule is mechanical because good intentions are not auditable.
It does not sell or broker
The project does not buy, sell, consign or broker works of art, and has no commercial relationship with any dealer or auction house. If the object were ever offered for sale, that fact would be disclosed here before any sale process began and new scholarly claims would stop for the duration.
It is not a museum
There is no institution behind this, no accreditation, no board and no collection. The standards borrowed from museum practice — the confidence model, the exposure ledger, the loan conditions — are borrowed, and citing them is not the same as being answerable to anyone who enforces them.
It is not finished
Version 0.9 is circulated for expert review before any DOI deposit. Nothing in it should be cited as settled. The most important fields on the object — its state, its printing date, its ownership before 2023 — are all open, and the record says so on every surface that touches them.
It does not publish the object's location
Custody is stated at city level and no more precisely. Storage, security and insurance details are not held in this database at all, which is a stronger guarantee than a policy of not displaying them.
How the record works
Step 1
Claim
An assertion about the object or about a census row: a field, a value, who stated it, when it was added and when it was last checked. Claims are never edited in place. A claim that turns out to be wrong is superseded by a new one and kept, so the dead end is visible to the next reader.
Step 2
Evidence
The material a claim rests on, with a resolvable URL wherever a third party could check it. Every piece of evidence sits in an eight-tier hierarchy — physical observation at the top, an unreviewed submission at the bottom — and the tier is a property of the source, independent of whether the project likes what it says.
Step 3
Confidence
How well the evidence supports the claim, and nothing else. Not how likely it is to be true, not how much the project would prefer it. Disputed and rejected are first-class states rather than low scores, and they are marked rather than faded out.
The rule is that confidence cannot outrank evidence, and it is enforced by the build rather than by editorial discipline. A claim marked confirmed whose strongest source is an unreviewed submission does not get published with a warning next to it — the deploy fails. So does a claim pointing at an evidence id that does not exist, a census row with no source URL, and any file that does not validate against the schema.
This matters more than it sounds. The failure mode of a self-published record is not deliberate deception; it is the slow, well-meaning upgrade of a possible into a probable because the owner has lived with the object for a year. A gate that runs before every deploy is not persuadable.
Currently 17 of 18 active claims carry at least one piece of evidence — 94% — and the ones that do not are visible as such rather than quietly omitted.
Who is behind it
The project is run by the owner of the sheet — a private collector in Dubai, United Arab Emirates — working with a small amount of software and, where the work needs it, paid specialists who are named on the work they did. Custody is stated as “Private collection, Dubai, United Arab Emirates” and no more precisely.
The owner is not named on the public record. That is a security decision about a portable object held privately, not an attempt to hide behind an institutional-sounding name: the site says on every page that it is published by a private owner, and the conflict that creates is set out in full rather than buried. Correspondence reaches a named human at research@rembrandt.aidispatch.xyz, and researchers, conservators and institutions who need to know who they are dealing with are told.
What is never published, and is not held in this database at all: the storage location, any address associated with the object, security arrangements, insurance details, and the movement of the object in real time. The reasoning is set out in the privacy and security policy.
How decisions are made
At this stage, one person approves every change to the canonical record: the steward, acting on evidence submitted by other people and credited to them. Nothing is edited into the record directly by a submitter, and nothing enters it without a logged reason — but the approval itself has no second signature on it.
That is a real weakness and it is not softened here. A single approver with an interest in the object is exactly the arrangement that peer review exists to correct. The mitigations in place are structural rather than personal — the build gates, the published negative findings, the changelog that makes every change visible — and structural mitigations do not substitute for someone with standing who can say no.
The intended end state is a small named group of external reviewers with the authority to overrule the owner on cataloguing, conservation and provenance. The version 1.0 gate list makes the first step of that a precondition of deposit rather than an aspiration: the dossier does not get a DOI until an external reviewer has read the record and their corrections have been incorporated.
Corrections do not wait for any of this. They are accepted from anyone, at any time, and published with the corrector named — how to send one, and every change made so far.
Policies
- 01Editorial policyUpdated 2026-08-12What gets published, at what confidence, and on whose authority.
- 02Correction policyUpdated 2026-08-12How to tell us we are wrong, and what happens next.
- 03Privacy and securityUpdated 2026-08-12What is deliberately not published, and why.
- 04Rights and licensingUpdated 2026-08-12What you may reuse, and on what terms.
- 05Conflicts of interestUpdated 2026-08-12The obvious one, stated plainly.
- 06Technical standardsUpdated 2026-08-12The standards this record is built to, and where it falls short of them.
- 07Contributor creditUpdated 2026-08-12How observations are attributed, permanently.