They Opened Them to Read Them

In 79 CE Vesuvius buried the Villa of the Papyri and carbonised more than eighteen hundred scrolls.

The eruption did not leave papyrus intact. It made black brittle cylinders that could survive underground for two millennia and could not be read.

That condition has invited the obvious remedy ever since they were recovered. Open them.

So the record has to distinguish two kinds of damage. Vesuvius did the first. Eighteenth- and nineteenth-century excavators and scholars did the second, destroying many scrolls while trying to unroll material that heat had already made fragile.

Recovery was not the opposite of loss here. It became another cause of it.

The object carries two damage histories #

Of the Herculaneum papyri now catalogued, only a few hundred survive nearly complete. Most are partially decayed and several hundred are fragments. Those are not three degrees of one event. The eruption both destroyed and preserved the corpus, and the handling history that followed is a separate record entirely. A single fragment can carry the result of volcanic heat and the result of an attempt to reverse volcanic heat.

Treat both as original condition and the intervention has vanished from its own evidence.

This is a familiar error precisely because the later actor is trying to help. Nobody handling those cylinders needed to be careless to make the loss unreconstructable. A physical unrolling has no neutral position available to it. It applies force to a singular object and hands the next investigator a changed one, usually with no way back to the prior arrangement. So the question is not whether a recovery succeeded. It is whether the state before recovery, the act itself, and the alternatives that were rejected all remain distinguishable afterwards.

Where they do not, a repaired object can no longer be used to audit the repair.

Excavation consumes the thing it records #

Archaeologists have a name for the part that disappears first, and it is not the material. It is context.

A deposit, a cut or a structure is not simply stuff in a hole. Its relationship to what lies above, below and beside it establishes the sequence that makes the material mean anything. Single-context recording was built around that. Each deposit, cut and structure gets its own record sheet, plan and section drawing, and the Harris matrix is compiled from those records during the dig rather than reconstructed afterwards from an empty trench.

The timing is the entire point. A contradiction can still be checked against the ground while the ground is still there.

The older Wheeler-Kenyon approach makes the same decision in earth instead of paperwork. Excavation proceeds in a grid and leaves balks, usually about a metre wide, standing between the squares. A balk is unexcavated material kept deliberately as a vertical, checkable profile. It costs exposed area and buys an inspectable sequence. Open-area excavation makes the opposite trade and pays for it with continuous recording. Neither approach pretends observation is free.

Reading without opening #

Herculaneum now has a different method. The Vesuvius Challenge, launched in 2023, uses X-ray phase-contrast tomography and machine-assisted virtual unwrapping, and by early 2024 it had revealed a small percentage of the text of one scroll without physically opening it.

That is not a recovered corpus, and the figure should not be made to carry one.

It is evidence for something narrower and more useful: a procedure obtained new information while leaving the object, and the possibility of a later procedure, intact. The difference is not that the modern method is more powerful. It is that the old one consumed its source while working and the new one does not.

Data repair rarely has the drama of a scroll breaking under a tool. It has the same structure.

A vendor file arrives with an unexpected code. A cleaning job standardises it, overwrites the raw delivery, and produces a field that finally joins. The anomaly might have been a supplier error, an undocumented code change, a bad extract, or the single discriminator separating two records that are not the same record.

Once the raw file is gone, the improved field remains and the evidence that could decide between those four does not.

The backfill is the quieter version. A team keeps the output of a failed run and loses the source snapshot it was generated from. It then knows the failure happened and cannot reproduce it, cannot compare alternative repairs, and cannot establish whether the fault was in the code, the input, or a source that has since changed underneath. The record of the intervention has replaced the record of the case.

Where the material stops being the constraint #

The comparison breaks at the object. A Herculaneum scroll is singular. Excavation removes a layer from one place and the layer cannot be put back. Digital files copy for almost nothing, and a transformation can in principle preserve input and output together.

In practice, retention rules, access controls, expiring vendor deliveries and production overwrite paths make a digital repair behave exactly like the physical case.

Cheap copying helps nobody if the copy was not retained, cannot be inspected, or cannot be tied to the transformation that produced the thing replacing it.

None of which argues for not repairing anything. Archaeologists use flotation to recover charred seeds and small bone that a trowel would miss, and a procedure can certainly preserve more evidence than it consumes. The constraint is retention beforehand: immutable input, the context record, the transformation, and the alternatives that were considered and dropped.

The scrolls that were opened are legible in the sense that somebody once looked at them, and they are gone.

The one being read now is still a cylinder. A small part of it has been read and the rest is still there, waiting for whatever comes after this.

That is the whole of the difference, and it is not about which method was cleverer. It is about which one left something for the next one.