A proposal. Find, image, translate and grade everything that survives of the human record — and release all of it into the public domain. Ptolemy’s method was the right ambition and the wrong architecture. He built one library and it went. This builds one index and no library.
Almost everything written in the ancient world is gone — not burned in one famous fire, but simply not recopied. Varro wrote some 620 books and one survives complete. And those are only the works we know were written, because somebody who survived happened to name them. The true loss cannot be measured. The Missing Ancient Works sets out the arithmetic. The proposal memo carries the full argument.
The part that can still be acted on is this: a very large body of material exists, physically, right now, and has never been read. Half a million fragments from one Egyptian rubbish dump, of which about five thousand are published. Some 1,800 carbonised scrolls whose first words were read in 2023. Roughly nine in ten excavated cuneiform tablets. Eight million objects in one museum alone. The constraint is not secrecy. It is that nobody is paid to do it.
Find them, buy what is at risk, image them at archival resolution, publish. The unit cost is already known — the British Library’s Endangered Archives Programme has funded over 500 projects in more than ninety countries and digitised over sixteen million images, at a cap of around £60,000 per project. The originals never leave the country. What EAP does not do is release the high-resolution files openly, and that is the single cheapest change available.
Log and scan every object, with its findspot. Photogrammetry needs a phone; reflectance imaging needs a camera, a light and a shiny ball — and recovers worn inscriptions that flat photographs lose entirely. There is no global catalogue of museum holdings and no way to say how many objects are held worldwide. A system that produced that number would be the first of its kind.
What was built, where, when, by what evidence, who ordered it and what survives — standing, ruined, relocated, lost, or known only from description. This is the leg where dating is hardest, because a building is usually dated by what was found in it, and the finds were often dated by the building. The works table on Egypt before the Ptolemies is the prototype.
Translate all of it. Two problems, not one. The ancient languages — Akkadian, Sumerian, Coptic, Syriac, Hittite — where a handful of people worldwide can read some of them, and where machine translation is already working. And the modern scholarly languages, which is the larger and stranger gap: two centuries of German, French, Russian, Italian, Chinese and Japanese scholarship that has never been translated. The field has been reading itself in fragments.
One index, no library. Nothing is centralised and nothing changes hands. The objects and manuscripts stay exactly where they are; what is unified is the catalogue. This is not theoretical — the International Dunhuang Project did it with 50,000 manuscripts split across five countries by the collectors who removed them. The physical objects will never be reunited. The images were, because nobody had to concede anything.
Public domain from capture. The underlying works are public domain by age — a Sumerian tablet has no copyright and neither does Livy. So reproductions of them — photographs, scans, 3D models, transcriptions, translations and derived data — are released into the public domain. No non-commercial clause, no attribution requirement, no database rights asserted. Every restriction is a place where a future institution can build a toll booth, and “non-commercial” alone blocks textbooks, documentaries and training data, which is most of the actual use.
No ruling on meaning. The project produces images, records and translations. It takes no position on what a document means, who it belongs to, or where it should be held. Those disputes continue in parallel and do not block the scanning. This is the design rule that lets all nations sign: an image makes no claim.
Everything graded, including the machine. A recovered text is not automatically true and a machine-restored line is not evidence. Ithaca, the best-published tool for restoring damaged Greek inscriptions, is right about 62 per cent of the time alone and around 72 with a historian. A reading that is 62 per cent likely is a hypothesis with a number on it. It is recorded as one.
The reading is a separate claim from the object. A contemporary inscription in a well-attested language with bilingual keys is not the same evidence as a contemporary inscription in Sumerian, which has no known relatives and was deciphered indirectly, or in Etruscan, which can be read aloud and largely not understood. Three dots for the object and one for the reading are different claims. Most registers, including this site’s, currently collapse them.
An object without a findspot can be read but cannot be dated, placed, or connected to anything. Looting does not merely steal objects; it destroys the information that made them evidence. Ten cases, each a different failure.
| Case | Where | What happened | The failure |
|---|---|---|---|
| Nefertiti | Berlin, Neues Museum | Berlin withheld its 3D scan for years. In 2016 two artists released a high-resolution model, claiming a covert scan; the museum then published its own data under a non-commercial licence. The bust is about 3,400 years old and in the public domain by any measure. The restriction is on the scan, not the object. | Ownership claimed in the reproduction |
| Palmyra | Syria | Structures destroyed in 2015. The models that survive were built from tourist photographs gathered after the fact, not from any systematic record. A scan is not the object, but it survives what the object does not. | No record made before loss |
| The Dunhuang manuscripts | split across five countries | Some 50,000 manuscripts from one sealed cave, dispersed by the collectors who removed them. The physical objects will never be reunited. The images were — by agreement on a common standard, with nobody conceding custody. | Dispersal, solved digitally |
| Ebla and Ugarit | Syria | Both produced large tablet archives by accident. Nobody was looking for either. There are hundreds of unexcavated tells across Iraq and Syria and no way to know which hold archives. | Unknown until dug |
| The Villa dei Papiri | Herculaneum | About 1,800 carbonised scrolls recovered, and they are a Greek Epicurean library. A Roman villa of that scale is expected to have held a Latin library too. The villa is not fully excavated. | Known target, undug |
| Ashurbanipal's library | Nineveh | Found in 1853. Tens of thousands of tablets went to the British Museum; the site has never been exhausted and the excavation methods of the 1850s destroyed context that cannot be recovered. | Dug badly, then abandoned |
| Qin Shi Huang's tomb | Xi'an | The terracotta army was found in 1974 in the outer pits. The tomb mound itself has never been opened, by deliberate policy, pending methods that will not destroy what is inside. | Deliberately unopened |
| The excavation backlog | worldwide | Most excavated material is in storage, uncatalogued and unpublished. Many digs were never written up at all, which means the findspot survives only in a dead archaeologist's notebooks — if those survive. | Context recorded, never published |
| Market antiquities | private and museum collections | An object without a findspot can be read but cannot be dated, placed, or connected to anything. Looting does not merely steal objects; it destroys the information that made them evidence. | Context destroyed |
| Reproduction fees | many museums | Some institutions fund conservation from image licensing. Abolishing the fees without replacing the income gives the institutions that most need the money a real reason to refuse. This is the trade the funding has to price. | The honest obstacle |
Every artefact needs an identifier and a tamper-evident record. How that record is held is a policy question, not an engineering one — it depends on whether the system is for institutions, for private holders, or for both. Four options, in ascending order of cost and reach. They are not mutually exclusive.
| Option | What it is | What it gets you | What it does not | Cost |
|---|---|---|---|---|
| A. Public timestamp only | A cryptographic hash of each scan and its record, published to a public timestamping service. No chain, no token, no ledger of its own. | Cheapest by a wide margin. Proves the object existed in that form on that date. Tamper-evident: alter the scan and the hash no longer matches. Uses existing infrastructure. | The record still sits in one institution's database, which can be amended or lost. Proves integrity, not independence. | Near zero |
| B. Public timestamp, mirrored catalogue | As above, plus the catalogue itself published as an open dataset, mirrored by several institutions on different continents, each publishing its own hash of the whole. | No single holder can quietly amend the record, because the mirrors disagree and the disagreement is visible. Still no chain. This is how the Dunhuang and IIIF models already work. | Requires several institutions to commit to mirroring in perpetuity, which is an organisational problem rather than a technical one. | Low |
| C. Permissioned chain | A distributed ledger written to by accredited institutions only — museums, national archives, accredited excavation directors. | No single party controls the record. Write access is controlled, so junk entries are limited. Auditable. | Excludes private owners, who hold much of the undocumented material. And a consortium of accredited institutions is not obviously more trustworthy than a mirrored catalogue — it is the same institutions. | Moderate |
| D. Open public chain | Anyone may register an object: private owners, dealers, detectorists, families. | Reaches the material nobody else can reach. Undocumented private holdings are the largest unrecorded body of artefacts in existence. Registration is voluntary and the incentive is real — a documented object is worth more than an undocumented one. | The laundering risk. A looted object given a clean-looking ID and ten years of ledger history becomes saleable. An immutable false provenance is worse than none, because it looks authoritative. | Moderate to high |
Why not an NFT. The token never conferred ownership of the artwork — it was a pointer, often to a URL that later went dead. There was no underlying asset, so when the speculation stopped there was nothing left. This inverts every part of that. The object stays with its owner, nothing is tokenised or traded, and the scan is public domain, so the thing being created is not scarce and cannot be cornered.
And what no registry can do. A ledger records that somebody made a claim. It cannot verify the claim was true, and an immutable false provenance is worse than none, because it looks authoritative. Which is why the claim and its evidence are recorded separately.
| Grade | Basis | What it means |
|---|---|---|
| Three | Excavated in a recorded context, findspot published, stratum known. | The object and its information both survive. |
| Two | Known collection history to a stated date, no excavation context. | Can be read and studied. Cannot be placed. |
| One | Owner asserts a history. No documentation. | Recorded as an assertion, not as provenance. The registry states who claimed what, and when. |
| None | No history offered, or history contradicted by evidence. | Registered as unprovenanced. The record says so permanently. |
Fund the imaging. Not new digs — the material is already indoors. Archival-resolution capture of what museums and libraries already hold, released open. This is the cheapest historical work available and almost nobody funds it. Nothing is being scanned today because nobody is paid to scan it.
Buy what is at risk. Manuscripts in private hands, in family libraries, in unstable regions. Timbuktu’s were evacuated under threat in 2012 and 2013 by people acting on their own initiative. Purchase, with provenance properly established, is legitimate and urgent — and it is what the Ptolemies did with the money, minus the theft.
Pay for the reading. The Vesuvius Challenge was open data, a stated problem and a prize. It read words off a scroll nobody had opened in nineteen centuries, within months. The same structure applied to Oxyrhynchus, the cuneiform backlog and palimpsest imaging would move faster than any institutional programme.
Dig where the evidence says libraries are. The Villa dei Papiri at Herculaneum is not fully excavated, and its 1,800 recovered scrolls are a Greek library — a Roman villa of that scale is expected to have held a Latin one. Ebla and Ugarit both produced tablet archives by accident. There are hundreds of unexcavated tells across Iraq and Syria and no way to know which hold archives.
Who pays. Anyone, on the condition above. The precedent is not a corporate grant — Vesuvius was funded by two individuals and delivered in months. The research labs have a real interest here: ancient-language restoration is a hard, publishable, low-resource problem, and the material is copyright-free. DeepMind built Ithaca and published it. Nobody was asked to give anything away, because there is nothing to own. Funding is welcome from any nation or company; the output belongs to no one.
The honest obstacle. Some museums fund conservation from image licensing fees. The proposal has to replace that income, not simply abolish it — otherwise the institutions that most need the money have a real reason to refuse. That is the trade the funding has to price, and it is not priced here.
This is a proposal at first draft. It is a page on a personal research site, not an organisation, and it has no funding, no staff and no institutional backing. What it has is a stated architecture, a set of principles, and an argument that the work is cheap relative to its value and that nobody is doing it. Figures on this page are the orders of magnitude published by the projects and institutions named, and are not independently audited here.
If it becomes more than a page it will need its own home. For now it is a page, an argument, and an invitation. The honest promise is smaller and better than the truth: not the truth, but everything that survives, with what attests it stated.