Why Did Delphi Stop at Six Columns?

14 Aug, 2026

Nowadays, archaeology and AI can reconstruct ancient buildings, sculpture, and color with enough evidence and transparency that preserving a ruin is no longer automatically the more authentic choice. Conjectural reconstruction can preserve meaning when fragments alone cannot.

Preservation of Fragments Is Not Preservation of Meaning

A conjectural reconstruction is an evidence-based attempt to recover the lost appearance of a building, sculpture, painting, or other ancient object when the surviving evidence does not determine every detail with certainty. Some elements rest on direct physical evidence: foundations, surviving fragments, pigment traces. Others rely on comparison, structural logic, historical sources, and informed scholarly inference. Every conjectural reconstruction is a mix of the two, and the honesty of the reconstruction depends on how clearly it marks which is which.

Three cases test that honesty in 3 different ways. A ruined temple at Delphi, where the missing evidence is standing right where it always stood. A set of sculptures split by ownership between 2 cities, where the missing evidence is 1,500 miles away and perfectly well known. And the tools now available to handle both cases without pretending to more certainty than the evidence supports.

Berthold Werner, Temple of Apollo at Delphi, 8 October 2017. Photograph via Wikimedia Commons. Licensed under CC BY-SA 3.0.

Why did Delphi stop at 6 columns?

The ancient Greeks called Delphi the center of the world. The omphalos marked the navel of the earth. Kings, generals, city states, and ordinary pilgrims climbed the Sacred Way to consult Apollo. At the center of that sacred landscape stood the Temple of Apollo, the building that made Delphi Delphi.

A visitor who climbs the Sacred Way today reaches a platform, scattered architectural remains, and 6 standing columns. The temple that stood there was a Doric peripteral building with 6 columns across the front and 15 down each flank, on a footprint of roughly 60 by 24 meters. That colonnade held 38 columns. Six of them stand.

Modern archaeology treats that restraint as responsible conservation. I want to question whether restraint is what actually happened.

Those 6 columns are not ancient survivors. Henri Ducoux, an architect of the French School at Athens, raised them in the years just before and during the Second World War. The campaign finished under the German occupation, and it stopped there. The entablature never went back up. No conservation charter drew that line. A war did, and 85 years of institutional inertia have since ratified it as though it were a principle.

That is a different story from the one the site tells about itself.

The case against reconstruction deserves a fair hearing. Archaeologists inherited a long history of restorers who moved stones, filled gaps with speculation, damaged ancient material, and produced monuments that reflected modern imagination as much as ancient evidence. Conservation moved toward anastylosis, the reassembly of surviving original elements, and toward interventions that stay limited, documented, and reversible. Those principles protect archaeological evidence, and they were earned the hard way.

Delphi itself supplies the cautionary example. The City of Athens sponsored the reconstruction of the Treasury of the Athenians between 1903 and 1906, and Joseph Replat directed it. Replat worked from Albert Tournaire's excavation drawings because the official publication of the building did not appear until 1933. The reconstruction contains errors that scholars have acknowledged ever since. The metopes a visitor sees on the building are casts; the originals sit in the museum. So the treasury is both the argument for reconstruction and the argument against it. It is the one building on the Sacred Way that reads as architecture, and it is also a building that was put back together on incomplete information by a man who could not yet read the definitive study of what he was rebuilding.

Conjectural reconstruction of the Sanctuary of Apollo at Delphi by Albert Tournaire

Albert Tournaire, conjectural reconstruction of the Sanctuary of Apollo at Delphi, 1894. École nationale supérieure des Beaux Arts. Public domain, via Wikimedia Commons.

What a ruin cannot tell you. A ruin does not necessarily communicate the building that once stood there, and non-reconstruction is not a neutral act. The 6 columns establish height. They cannot establish mass, enclosure, rhythm, or hierarchy. The ancient visitor met a large Doric temple with continuous colonnades, substantial walls, an interior, an entablature, and a roof, dominating everything around it. The modern visitor meets an archaeological diagram and a view. We understand the footprint intellectually. We do not experience the architecture. The absence changes the building's character in specific ways. The temple now appears open rather than enclosed, skeletal rather than massive, and picturesque rather than imposing. Preservation protects the authentic stones while obscuring the authentic architectural experience. The contradiction is sharpest at Delphi, where the subsidiary treasury reads as a building and Apollo's temple reads as archaeology.

Sicily teaches both lessons at once. Selinunte shows what recovered elevation does. Under Jole Bovio Marconi, workers re-erected the 6 by 15 colonnade of Temple E between 1956 and 1959. The eye comprehends the temple immediately, and the collapsed monuments nearby make the comparison unavoidable. One field of stones communicates archaeological remains. Temple E communicates a building. It is also contested. Italian conservation scholarship, including Treccani's assessment, criticizes the intervention for requiring radical and irreversible integrations, and for producing an assembly that has weathered unevenly ever since. Anyone citing Temple E as a success has to cite the criticism with it.

Agrigento sharpens both edges. Between 1922 and 1924, the English captain Alexander Hardcastle financed the raising of 8 columns at the Temple of Heracles, and those columns restore enough vertical structure for a visitor to read the colonnade's rhythm. Hardcastle lost his fortune in the 1929 crash and died in a psychiatric hospital in 1933. A few hundred meters away, the celebrated 4-column corner of the Temple of the Dioscuri is a 19th century assembly: 3 columns raised in 1836 by Valerio Villareale and Saverio Cavallari under the Duke of Serradifalco, a 4th added in 1856, and a richly carved geison that did not belong to the temple at all. Pirro Marconi criticized it in the 1920s. It remains the most photographed object in the valley. The Dioscuri is the real warning, and it is more specific than "reconstruction is risky." Reconstruction misleads when confidence exceeds evidence and when the record of what was invented does not travel with the monument. That argues for disciplined reconstruction, not for refusing reconstruction.

Paestum and Knossos mark the two poles. Paestum makes the point from the other direction. The Temple of Neptune retains its full sequence of columns, much of its entablature, and both pediments, and it owes that to robust construction and the gradual abandonment of the city rather than to any modern campaign. Nobody standing in front of it needs a floor plan to know they are looking at an enormous building. We value Paestum partly because chance preserved the architectural information that collapsed everywhere else. Where chance failed, conservation often treats the loss as permanent by default.

Knossos marks the opposite pole. Arthur Evans excavated from 1900 to 1930 and rebuilt in ways nobody would repeat, most extensively through Piet de Jong's reinforced concrete work from 1922 onward. The errors are documented and severe. The faces of the "Ladies in Blue" are entirely invented, since no facial fragments survived. The "Saffron Gatherer" turned out to be a monkey. Mary Beard's formulation is the one worth keeping: the more famous the image is now, the less of it is actually ancient. Knossos is also, uncomfortably, the most comprehensible Bronze Age site in the Mediterranean. Visitors experience stairs, rooms, levels, courtyards, circulation, color, and enclosure. The palace becomes a place rather than a plan. Evans crossed lines that should not be crossed, and he also demonstrated that highly conservative presentation loses something real. Both things are true, and the second one is the part conservation orthodoxy rarely says out loud.

What Ducoux could not do, and what changes now

Ducoux worked with drawings, judgment, and whatever he could lift. He raised 6 columns because that was the limit of what physical anastylosis and interwar-era evidence allowed, and then a war ended the project before anyone had to decide where the next limit was.

That limit has moved. The old choice was binary. An archaeologist drew 1 reconstruction, defended it, and everyone treated it as more certain than it was, because a finished drawing looks finished. What has changed is not that AI produces a better drawing. It is that AI makes multiple, differently confident drawings cheap enough to produce and compare, and that surviving evidence itself can now be measured, matched, and modeled at a scale no interwar architect had access to.

A 2023 paper in Advances in Archaeological Practice, by Matthew Magnani and Jon Clindaniel, used generative AI to visualize competing interpretations of the same evidence rather than a single authoritative past, illustrating contrasting scenarios of Neanderthal behavior side by side instead of picking a winner in advance. That is the right use of the technology: not "what did this look like," but "what range of things could this evidence support, and how strongly does it support each."

Five applications follow from that reframing.

Fragment matching. The Stanford Digital Forma Urbis Romae Project has spent more than 2 decades scanning the 1,186 surviving pieces of an 18-by-13-meter marble map of ancient Rome, carved between 203 and 211 CE, with computational fragment-matching as the long-term goal. The same logic applies to architectural blocks, sculpture fragments, and fresco pieces. An ICCV 2025 paper, ReassembleNet, combines learned keypoints, graph-based pooling, and diffusion-based pose estimation to reassemble fragmented frescoes from geometry and texture together, a hybrid approach rather than a single technique.

Alternative reconstructions. Instead of 1 drawing of the Temple of Apollo's elevation, a system constrained by the surviving blocks, comparable 4th-century Doric temples, and structural limits could generate several plausible elevations for scholars to argue over and eliminate.

Polychromy. A 2023 study in Applied Sciences, examining a single Attic funerary stele at the Metropolitan Museum, combined multiband imaging, digital microscopy, portable and scanning XRF, and micro-invasive Raman spectroscopy to identify Egyptian blue, azurite, cinnabar, and ochre on a surface that looks like plain marble to the eye. That is a case study, not a general survey, but the method generalizes: instruments can establish what pigments were present in specific areas without recovering the full original scheme, and AI could combine those anchor points with securely documented parallels to propose a family of possible completions rather than 1 speculative brightly painted statue.

Structural plausibility. A reconstruction has to work as a building, not just resemble one. Historic Building Information Modeling already reconstructs lost architectural states from heterogeneous evidence; a 2024 Heritage Science study applied it to Suwon Hwaseong, a Joseon-dynasty Korean fortress, integrating 3D geometry with documented repair history. Nothing about the method is specific to wood or to Korea. The same modeling discipline could test whether a proposed Doric entablature would actually stand.

Explicit uncertainty. This is the part that matters most, and it is not hypothetical at Delphi. A current French School and Bordeaux project is performing a block-by-block digital anastylosis of the temple's peristyle columns, for a reason that should give any reconstruction advocate pause: scholars still disagree on whether each column carried 12, 13, or 14 drums. Courby, Ducoux, and the modern publication of the temple do not agree, and the height of the colonnade remains an open question 85 years after the columns went up. That is exactly the kind of disagreement AI-assisted modeling could hold open rather than paper over. Instead of 1 image labeled "Temple of Apollo, 330 BC," a system could tag each element with its evidentiary basis: footprint from excavated foundations, near certain; column positions from surviving members, very high confidence; drum count and roof form, actively disputed; painted decoration from pigment traces and parallels, low confidence. The numbers in a table like that are illustrative, not measurements, but the discipline of assigning them is real. It forces the reconstruction to say what it does not know, which is exactly what a single finished rendering never does.

None of this requires abandoning anastylosis. It extends the same logic. Original blocks whose positions can be established confidently return to those positions. Structural pieces needed to support them use identifiable modern material. Elements known with high confidence receive limited reconstruction. Conjectural sculpture, painting, and decoration stay digital, or stay absent, until the evidence improves. The choice conservation has treated as binary, rebuild the monument or leave it a ruin, was never actually binary. The tools now exist to work the middle ground honestly, tier by tier, with every tier labeled.

Nobody needs to propose a new temple at Delphi. The question is narrower: could enough elevation be recovered to restore the scale that 6 columns cannot communicate? At present, the burden falls almost entirely on intervention. Archaeologists ask whether sufficient evidence exists to justify putting something back. Delphi invites the inverse question. If sufficient evidence exists to reconstruct part of a monument responsibly, what do we lose by declining? We lose scale. We lose enclosure. We lose architectural hierarchy. We lose the relationship between the human body and the monument. At Delphi we may also lose some understanding of why generations of Greeks judged the place worth the walk.

The same question, asked of the Parthenon

Once reconstruction is framed as recovering a lost whole from surviving evidence, a second question follows naturally: does it matter where the surviving pieces of that whole physically sit? Delphi's problem is a building missing most of its elevation. The Parthenon's problem is different. Much of the original sculptural program is lost, but roughly half survives, divided principally between Athens and London. Unlike Delphi, the problem with much of what survives is not uncertainty about where it belongs. We know where it belongs.

The Parthenon sculptures were never independent artworks. The metopes, frieze, and pedimental figures were carved as parts of 1 building's single sculptural program, and roughly half of what survives is in the Acropolis Museum in Athens while most of the rest is in the British Museum in London. The Acropolis Museum's Parthenon Gallery makes the split impossible to ignore: original marble sculptures are displayed alongside plaster casts of the pieces held abroad, with glass walls giving visitors a direct sightline to the actual Parthenon standing on the hill outside. In at least 1 documented case, South Frieze Block 1, the arrangement is literal: the left half is genuine marble in Athens, and the right half is a plaster cast of the original that sits in London. The museum is performing a conjectural reconstruction of a block whose missing half is not conjectural at all. We know exactly where it is.

That is a sharper argument than the ownership question usually allows for. The standard framing asks who has the better legal claim. A reconstruction framing asks a different question: an archaeological object is evidence for a lost whole, and if scholarship is trying to reconstruct that whole, separating securely associated components across museums actively impairs the reconstruction. Not "who owns this," but "where can this best recover its original meaning."

The British Museum's answer is coherent and worth stating fairly. Its trustees maintain that Lord Elgin's activities were investigated by a Parliamentary Select Committee in 1816 and found lawful, that Parliament's purchase and vesting of the sculptures that year settled the legal question, and that the sculptures now form part of "this interconnected world collection," conveying, in the museum's words, the influences between Egyptian, Persian, Greek, and Roman cultures. The trustees describe a legal and moral duty to preserve the collection intact for the public.

The Greek government's position, laid out on the Ministry of Culture's site, argues instead for restoring "the integrity of the monument in its historic, cultural and natural environment," on the grounds that no researcher or visitor can fully appreciate sculptures that were designed as part of a specific building once they are fragmented and exhibited as standalone works. UNESCO's Intergovernmental Committee for the Return of Cultural Property has treated the case as active for years and, at its most recent session in May 2026, formally called on the UK to show "tangible goodwill" and intensify efforts toward reunification. The UK government continues to reject binding return, offering only loan arrangements, which Greece has rejected as inadequate. As of this writing, the impasse holds.

A hierarchy, not a blanket rule. The reconstruction argument does not imply that every Greek antiquity in a foreign museum should return to Greece, and it would be weaker if it tried to claim that. A complete vase excavated in Greece but standing on its own carries much of its interpretive context with it wherever it travels. A statue removed intact from a sanctuary raises a stronger question, because it once stood in a specific relationship to a specific space. A sculpture carved as part of a standing building's architectural program raises a stronger question still, because it was never meant to be freestanding at all. A fragment whose other half remains at the original site, or in a museum built to face that site, is the strongest case of the 4, because the missing information is not lost. It is 1,500 miles away. The Parthenon sculptures sit at the far end of that spectrum, and South Frieze Block 1 is close to a proof by demonstration.

Reunification does not require putting the originals back outside on the Acropolis, which would create exactly the conservation problems that motivated moving them indoors in the first place. It suggests a 3-layer arrangement instead, the same 3-layer logic that applies at Delphi: the Acropolis Museum holds the reunited originals, arranged in their architectural sequence; the Parthenon itself remains the monument, with any physical infill using clearly identifiable modern material; and digital, AI-assisted reconstruction fills the gaps that no physical intervention should touch, with the confidence level of every restored color, missing face, or conjectural pose stated rather than implied.

The thread connecting all 3

The Temple of Apollo asks whether leaving architecture in ruins distorts our understanding of what it was. Painted Greek sculpture asks whether presenting it in bare white marble, a habit that started with the Renaissance mistaking weathered stone for the ancient ideal, distorts our understanding of what it looked like. The Parthenon sculptures ask whether scattering the surviving pieces of 1 building's single program across 2 cities distorts our understanding of the whole it was designed to be part of.

All 3 point at the same proposition. Keeping a fragment safe is not the same as keeping its meaning intact. Conservation has spent a century getting very good at the first job. The tools now exist to get serious about the second, without pretending, the way 20th-century restorers sometimes did, that conjecture is fact. Ancient Greeks built temples, not ruins, and they built the Parthenon as 1 building, not 2 collections. The obligation runs the other way now: state what is known, state what is inferred, and stop treating the distance between them as a reason to do nothing.

My blog