Which Version Is the Original?

If you walk into Saint Bavo's Cathedral in Ghent today and stand in front of the most famous painting in Belgium, seven of the panels you are looking at are photographs.

They are very good photographs, and the cathedral says so openly. The upper register of the Ghent Altarpiece is across town at the Museum of Fine Arts, in the third phase of a conservation campaign that began in 2012, and the panels will be back when the work is done. In the meantime the object in the case is part original, part reproduction, and nobody is pretending otherwise.

I keep coming back to that arrangement because it is an honest answer to a question that has no clean one. When a thing has been cleaned, repaired, cut down, overpainted, unoverpainted, copied and reassembled across six centuries, which state of it is the real one?

Hubert and Jan van Eyck, The Adoration of the Mystic Lamb, completed 1432. Saint Bavo's Cathedral, Ghent. Public domain, via Wikimedia Commons.

Five objects, five different answers

I want to put five cases next to each other, because the conservation field has been arguing about this for forty years with far more rigour than the software field applies to the same problem, and the disagreements are instructive.

1. The Sistine ceiling, and the possibility that the cleaning removed the artist

Michelangelo, The Creation of Adam, Sistine Chapel, in its post-restoration state. Public domain, via Wikimedia Commons.

The Vatican Museums cleaned the ceiling in phases between 1980 and 1994, with the lunettes first and the Last Judgment unveiled by John Paul II on 8 April 1994. Gianluigi Colalucci was chief restorer. Nippon Television paid for it, four point two million dollars, in exchange for exclusive reproduction rights.

The objection, made most publicly by the Columbia art historian James Beck in an open letter to La Repubblica on 1 February 1987, was not that the cleaning was clumsy. It was that the layer being removed was Michelangelo's. Beck and ArtWatch argued that the artist had finished the frescoes with glue-based painting applied a secco, onto dry plaster, and that stripping it left the bare fresco exposed and the modelling flattened.

The Vatican's reply was not rhetorical, and this is the part worth keeping. Fabrizio Mancinelli pointed to salt crystals found between the paint layer and the glue layer. Salts take time to form. If they sit between the two, the glue went on long after the fresco was finished, which means it was not Michelangelo's finishing but somebody else's later varnish.

The argument was settled, to the extent that it was settled at all, by reading the order of the layers.

2. Ghent, where roughly seventy percent of the surface was somebody else

The Royal Institute for Cultural Heritage in Brussels has been working on the altarpiece since October 2012, in a glass-walled workshop inside the Museum of Fine Arts so the public can watch. The Museum's own summary of what phase one found is blunt: close to seventy percent of the altarpiece's total surface had been overpainted and was therefore not of Van Eyck's hand.

That headline number flattens something more interesting. The Institute's own synthesis breaks it down: around seventy percent on the closed exterior, around half of the original painting covered on the central Lamb panel, and only ten to fifteen percent on the lower side panels. The overpainting was not a uniform varnish. It was a series of decisions, made by different people, at different times, for different reasons.

The most famous of those decisions concerned the Lamb's face. The Van Eycks painted it with forward-facing, close-set, human-like eyes. A sixteenth-century hand repainted it into a naturalistic, passive, side-eyed sheep. In 2019 the conservators took the overpaint off and the original came back, staring straight out.

Hélène Dubois, who ran the project, describes the effect as a more intense interaction with the viewer. The art historian Koenraad Jonckheere describes the original intervention as adapting the work to the taste of the time.

Note what had to happen before the paint moved. The Institute's synthesis records that the advice of the international committee of experts to uncover the original was accepted by all parties involved. An irreversible change to a six-hundred-year-old object required a standing committee, a documented recommendation, and sign-off from the owner, the conservator and the state. That is a change-control process, and it is stricter than the one guarding most production databases.

3. Dresden, where a layer of dirt settled the question

Johannes Vermeer, Girl Reading a Letter at an Open Window, Gemäldegalerie Alte Meister, Dresden, shown in its state before the 2021 restoration. The rear wall is empty. It is not any more. Public domain, via Wikimedia Commons.

An X-ray in 1979 showed a painting within the painting: a naked Cupid on the rear wall, hidden under plain paint. For four decades the assumption was that Vermeer had painted him out himself, changing his mind.

The research project that began in 2017 overturned that, and the evidence is the cleanest thing in this whole article. The museum's own wording:

Aged layers of binding agent and a layer of dirt exist between the original paint layers applied by Vermeer and the paint used for painting over the Cupid picture. A period of several decades must therefore have intervened.

Dirt is a timestamp. You cannot get dirt between two coats of paint applied in the same week. The physical stratigraphy of the object dated the edit, and the date placed it well after the artist was dead. The decision to remove the overpaint was taken in early 2018, and Christoph Schölzel took it off with a fine scalpel under a microscope over the following three years.

This is provenance in the strict sense, and it is worth separating from the loose sense. Provenance is usually discussed as a chain of owners. What actually decided this case was a chain of events, recoverable from the artefact itself, in order, with intervals. Nobody had to be trusted. The layers were read.

4. Amsterdam, where the fix was not applied to the original at all

The Night Watch was trimmed in 1715 to fit a wall: 64.4 cm off the left, 7 cm off the right, 23.3 off the top, 11.3 off the bottom. The cut-off figures survive only in a smaller seventeenth-century copy, probably by Gerrit Lundens.

In 2021 the Rijksmuseum reconstructed the missing strips. The Lundens copy supplied the composition; neural networks trained on Rembrandt's handling supplied the brushwork, correcting for Lundens' own style. Robert Erdmann, the senior scientist on the project: it is thanks to artificial intelligence that we can so closely simulate the original painting.

And then the part that matters. The reconstructed sections were printed on panels and placed around the original painting. The canvas was not touched.

The reconstruction was hung as a view of the object, not as a change to it.

That is the best answer anyone in this article gives to the question in the title. You do not have to decide which version is the original if you never overwrite the original. You render the other version alongside it, label it, and let the viewer hold both.

Video. "Operation Night Watch," Rijksmuseum.

5. Borja, where the failed restoration became the economy

In August 2012 an eighty-one-year-old parishioner named Cecilia Giménez repainted a deteriorating fresco of Christ in the Sanctuary of Mercy in Borja, Zaragoza. She worked in daylight and did not hide it. The BBC's correspondent described the result as a crayon sketch of a very hairy monkey in an ill-fitting tunic.

The sanctuary had been drawing five or six thousand visitors a year. In the year after, it drew more than forty thousand, with over fifty thousand euros in donations, which went to church staff and a retirement home. By 2016 the figure was fifty-seven thousand and an interpretation centre opened. It has since settled at something like fifteen to twenty thousand a year, which is still triple the old number. Giménez died in December 2025, aged ninety-four.

I am not showing you the repainted version, and the reason is itself part of the argument. Giménez asserted copyright over the altered work, and a merchandising arrangement followed. The original fresco by Elías García Martínez is public domain. The failed restoration of it is somebody's property.

Table 1. The same question, five times, with five different procedures and one non-procedure.

What a versioning discipline looks like when the stakes are real

Conservation has three habits that any field handling irreplaceable artefacts could borrow, and mostly does not.

• Reversibility is a design constraint, not a nice-to-have. The profession's working preference is for interventions that can be undone, and where that is impossible it says so explicitly and raises the bar for approval. Notice that the two cleanest outcomes above, Dresden and Amsterdam, are the two where the reasoning was documented in public before the work happened.

• The artefact is its own log. Layers, salts, dirt, the order of paint films. None of that was recorded by anyone at the time, and all of it was recoverable, because physical material accumulates evidence whether or not anybody intends it to. Digital artefacts do not have this property. A file that has been edited in place carries no stratigraphy at all. Whatever history you want later, you have to decide to keep now.

• A surrogate is not a lie if it is labelled. Seven photographs in a cathedral, printed panels around a Rembrandt, a 1945 copy standing in for a stolen panel. All fine, all honest, because in each case the institution says which is which.

The third one is where I think most working practice goes wrong, and not only in conservation. The failure is rarely that a substitute was used. It is that the substitute quietly stopped being marked as one, and a generation later nobody can tell.

The panel that is not there

Ghent has one more turn in it. On the night of 10 April 1934 the Just Judges panel was lifted out of its frame and taken. A note was left reading that it was taken from Germany by the Treaty of Versailles. A ransom of a million Belgian francs was demanded, the government refused, and at least eleven letters were exchanged. The thief returned the other half of the same double-sided panel as a gesture of good faith. On his deathbed in November 1934, Arsène Goedertier said he alone knew where it was and would take the secret with him.

It has never been found. In 1945 Jef Van der Veken painted a replacement, working from a sixteenth-century copy by Michiel Coxie, waxed it to match the patina of the rest, and signed it covertly by giving one of the horsemen the face of King Leopold III.

That copy hangs in the altarpiece now and functions as one of its twelve panels. I could not find any statement from the cathedral or the conservation institute defining its status: original, surrogate, or labelled reproduction. The absence is the most interesting fact in this article. Ninety years on, the substitute has simply become the thing.

A deep dive into this one object, panel by panel, is on my Korean blog: 베이니월드.

Sources

Ghent Altarpiece conservation phases, the seventy percent figure and the Lamb's face: KIK-IRPA and MSK Ghent. Current display state of the cathedral's panels: Saint Bavo's Cathedral. The Dresden Vermeer, including the quoted passage on the dirt layer: Gemäldegalerie Alte Meister. The Night Watch reconstruction: Rijksmuseum. Sistine restoration dates, personnel and the Beck controversy: Vatican Museums press record and contemporaneous reporting, February 1987.

Two things I could not pin down and have not asserted: the chemical composition of the AB57 solvent used on the ceiling, which circulates online without a primary conservation-science source, and the exact overall dimensions of the Ghent Altarpiece, for which published figures genuinely disagree.